Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * pathnode.c
4 : * Routines to manipulate pathlists and create path nodes
5 : *
6 : * Portions Copyright (c) 1996-2019, 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 "miscadmin.h"
20 : #include "foreign/fdwapi.h"
21 : #include "nodes/extensible.h"
22 : #include "nodes/nodeFuncs.h"
23 : #include "optimizer/appendinfo.h"
24 : #include "optimizer/clauses.h"
25 : #include "optimizer/cost.h"
26 : #include "optimizer/optimizer.h"
27 : #include "optimizer/pathnode.h"
28 : #include "optimizer/paths.h"
29 : #include "optimizer/planmain.h"
30 : #include "optimizer/prep.h"
31 : #include "optimizer/restrictinfo.h"
32 : #include "optimizer/tlist.h"
33 : #include "parser/parsetree.h"
34 : #include "utils/lsyscache.h"
35 : #include "utils/memutils.h"
36 : #include "utils/selfuncs.h"
37 :
38 :
39 : typedef enum
40 : {
41 : COSTS_EQUAL, /* path costs are fuzzily equal */
42 : COSTS_BETTER1, /* first path is cheaper than second */
43 : COSTS_BETTER2, /* second path is cheaper than first */
44 : COSTS_DIFFERENT /* neither path dominates the other on cost */
45 : } PathCostComparison;
46 :
47 : /*
48 : * STD_FUZZ_FACTOR is the normal fuzz factor for compare_path_costs_fuzzily.
49 : * XXX is it worth making this user-controllable? It provides a tradeoff
50 : * between planner runtime and the accuracy of path cost comparisons.
51 : */
52 : #define STD_FUZZ_FACTOR 1.01
53 :
54 : static List *translate_sub_tlist(List *tlist, int relid);
55 : static int append_total_cost_compare(const ListCell *a, const ListCell *b);
56 : static int append_startup_cost_compare(const ListCell *a, const ListCell *b);
57 : static List *reparameterize_pathlist_by_child(PlannerInfo *root,
58 : List *pathlist,
59 : RelOptInfo *child_rel);
60 :
61 :
62 : /*****************************************************************************
63 : * MISC. PATH UTILITIES
64 : *****************************************************************************/
65 :
66 : /*
67 : * compare_path_costs
68 : * Return -1, 0, or +1 according as path1 is cheaper, the same cost,
69 : * or more expensive than path2 for the specified criterion.
70 : */
71 : int
72 387724 : compare_path_costs(Path *path1, Path *path2, CostSelector criterion)
73 : {
74 387724 : if (criterion == STARTUP_COST)
75 : {
76 198392 : if (path1->startup_cost < path2->startup_cost)
77 96740 : return -1;
78 101652 : if (path1->startup_cost > path2->startup_cost)
79 43128 : return +1;
80 :
81 : /*
82 : * If paths have the same startup cost (not at all unlikely), order
83 : * them by total cost.
84 : */
85 58524 : if (path1->total_cost < path2->total_cost)
86 36582 : return -1;
87 21942 : if (path1->total_cost > path2->total_cost)
88 1496 : return +1;
89 : }
90 : else
91 : {
92 189332 : if (path1->total_cost < path2->total_cost)
93 181006 : return -1;
94 8326 : if (path1->total_cost > path2->total_cost)
95 1546 : return +1;
96 :
97 : /*
98 : * If paths have the same total cost, order them by startup cost.
99 : */
100 6780 : if (path1->startup_cost < path2->startup_cost)
101 8 : return -1;
102 6772 : if (path1->startup_cost > path2->startup_cost)
103 0 : return +1;
104 : }
105 27218 : return 0;
106 : }
107 :
108 : /*
109 : * compare_path_fractional_costs
110 : * Return -1, 0, or +1 according as path1 is cheaper, the same cost,
111 : * or more expensive than path2 for fetching the specified fraction
112 : * of the total tuples.
113 : *
114 : * If fraction is <= 0 or > 1, we interpret it as 1, ie, we select the
115 : * path with the cheaper total_cost.
116 : */
117 : int
118 2208 : compare_fractional_path_costs(Path *path1, Path *path2,
119 : double fraction)
120 : {
121 : Cost cost1,
122 : cost2;
123 :
124 2208 : if (fraction <= 0.0 || fraction >= 1.0)
125 1014 : return compare_path_costs(path1, path2, TOTAL_COST);
126 2388 : cost1 = path1->startup_cost +
127 1194 : fraction * (path1->total_cost - path1->startup_cost);
128 2388 : cost2 = path2->startup_cost +
129 1194 : fraction * (path2->total_cost - path2->startup_cost);
130 1194 : if (cost1 < cost2)
131 624 : return -1;
132 570 : if (cost1 > cost2)
133 570 : return +1;
134 0 : return 0;
135 : }
136 :
137 : /*
138 : * compare_path_costs_fuzzily
139 : * Compare the costs of two paths to see if either can be said to
140 : * dominate the other.
141 : *
142 : * We use fuzzy comparisons so that add_path() can avoid keeping both of
143 : * a pair of paths that really have insignificantly different cost.
144 : *
145 : * The fuzz_factor argument must be 1.0 plus delta, where delta is the
146 : * fraction of the smaller cost that is considered to be a significant
147 : * difference. For example, fuzz_factor = 1.01 makes the fuzziness limit
148 : * be 1% of the smaller cost.
149 : *
150 : * The two paths are said to have "equal" costs if both startup and total
151 : * costs are fuzzily the same. Path1 is said to be better than path2 if
152 : * it has fuzzily better startup cost and fuzzily no worse total cost,
153 : * or if it has fuzzily better total cost and fuzzily no worse startup cost.
154 : * Path2 is better than path1 if the reverse holds. Finally, if one path
155 : * is fuzzily better than the other on startup cost and fuzzily worse on
156 : * total cost, we just say that their costs are "different", since neither
157 : * dominates the other across the whole performance spectrum.
158 : *
159 : * This function also enforces a policy rule that paths for which the relevant
160 : * one of parent->consider_startup and parent->consider_param_startup is false
161 : * cannot survive comparisons solely on the grounds of good startup cost, so
162 : * we never return COSTS_DIFFERENT when that is true for the total-cost loser.
163 : * (But if total costs are fuzzily equal, we compare startup costs anyway,
164 : * in hopes of eliminating one path or the other.)
165 : */
166 : static PathCostComparison
167 1708048 : compare_path_costs_fuzzily(Path *path1, Path *path2, double fuzz_factor)
168 : {
169 : #define CONSIDER_PATH_STARTUP_COST(p) \
170 : ((p)->param_info == NULL ? (p)->parent->consider_startup : (p)->parent->consider_param_startup)
171 :
172 : /*
173 : * Check total cost first since it's more likely to be different; many
174 : * paths have zero startup cost.
175 : */
176 1708048 : if (path1->total_cost > path2->total_cost * fuzz_factor)
177 : {
178 : /* path1 fuzzily worse on total cost */
179 837110 : if (CONSIDER_PATH_STARTUP_COST(path1) &&
180 27118 : path2->startup_cost > path1->startup_cost * fuzz_factor)
181 : {
182 : /* ... but path2 fuzzily worse on startup, so DIFFERENT */
183 13020 : return COSTS_DIFFERENT;
184 : }
185 : /* else path2 dominates */
186 796972 : return COSTS_BETTER2;
187 : }
188 898056 : if (path2->total_cost > path1->total_cost * fuzz_factor)
189 : {
190 : /* path2 fuzzily worse on total cost */
191 459310 : if (CONSIDER_PATH_STARTUP_COST(path2) &&
192 11570 : path1->startup_cost > path2->startup_cost * fuzz_factor)
193 : {
194 : /* ... but path1 fuzzily worse on startup, so DIFFERENT */
195 6280 : return COSTS_DIFFERENT;
196 : }
197 : /* else path1 dominates */
198 441460 : return COSTS_BETTER1;
199 : }
200 : /* fuzzily the same on total cost ... */
201 450316 : if (path1->startup_cost > path2->startup_cost * fuzz_factor)
202 : {
203 : /* ... but path1 fuzzily worse on startup, so path2 wins */
204 179590 : return COSTS_BETTER2;
205 : }
206 270726 : if (path2->startup_cost > path1->startup_cost * fuzz_factor)
207 : {
208 : /* ... but path2 fuzzily worse on startup, so path1 wins */
209 38006 : return COSTS_BETTER1;
210 : }
211 : /* fuzzily the same on both costs */
212 232720 : return COSTS_EQUAL;
213 :
214 : #undef CONSIDER_PATH_STARTUP_COST
215 : }
216 :
217 : /*
218 : * set_cheapest
219 : * Find the minimum-cost paths from among a relation's paths,
220 : * and save them in the rel's cheapest-path fields.
221 : *
222 : * cheapest_total_path is normally the cheapest-total-cost unparameterized
223 : * path; but if there are no unparameterized paths, we assign it to be the
224 : * best (cheapest least-parameterized) parameterized path. However, only
225 : * unparameterized paths are considered candidates for cheapest_startup_path,
226 : * so that will be NULL if there are no unparameterized paths.
227 : *
228 : * The cheapest_parameterized_paths list collects all parameterized paths
229 : * that have survived the add_path() tournament for this relation. (Since
230 : * add_path ignores pathkeys for a parameterized path, these will be paths
231 : * that have best cost or best row count for their parameterization. We
232 : * may also have both a parallel-safe and a non-parallel-safe path in some
233 : * cases for the same parameterization in some cases, but this should be
234 : * relatively rare since, most typically, all paths for the same relation
235 : * will be parallel-safe or none of them will.)
236 : *
237 : * cheapest_parameterized_paths always includes the cheapest-total
238 : * unparameterized path, too, if there is one; the users of that list find
239 : * it more convenient if that's included.
240 : *
241 : * This is normally called only after we've finished constructing the path
242 : * list for the rel node.
243 : */
244 : void
245 1112174 : set_cheapest(RelOptInfo *parent_rel)
246 : {
247 : Path *cheapest_startup_path;
248 : Path *cheapest_total_path;
249 : Path *best_param_path;
250 : List *parameterized_paths;
251 : ListCell *p;
252 :
253 : Assert(IsA(parent_rel, RelOptInfo));
254 :
255 1112174 : if (parent_rel->pathlist == NIL)
256 0 : elog(ERROR, "could not devise a query plan for the given query");
257 :
258 1112174 : cheapest_startup_path = cheapest_total_path = best_param_path = NULL;
259 1112174 : parameterized_paths = NIL;
260 :
261 2433156 : foreach(p, parent_rel->pathlist)
262 : {
263 1320982 : Path *path = (Path *) lfirst(p);
264 : int cmp;
265 :
266 1320982 : if (path->param_info)
267 : {
268 : /* Parameterized path, so add it to parameterized_paths */
269 59830 : parameterized_paths = lappend(parameterized_paths, path);
270 :
271 : /*
272 : * If we have an unparameterized cheapest-total, we no longer care
273 : * about finding the best parameterized path, so move on.
274 : */
275 59830 : if (cheapest_total_path)
276 8390 : continue;
277 :
278 : /*
279 : * Otherwise, track the best parameterized path, which is the one
280 : * with least total cost among those of the minimum
281 : * parameterization.
282 : */
283 51440 : if (best_param_path == NULL)
284 48358 : best_param_path = path;
285 : else
286 : {
287 6164 : switch (bms_subset_compare(PATH_REQ_OUTER(path),
288 6164 : PATH_REQ_OUTER(best_param_path)))
289 : {
290 : case BMS_EQUAL:
291 : /* keep the cheaper one */
292 36 : if (compare_path_costs(path, best_param_path,
293 : TOTAL_COST) < 0)
294 0 : best_param_path = path;
295 36 : break;
296 : case BMS_SUBSET1:
297 : /* new path is less-parameterized */
298 138 : best_param_path = path;
299 138 : break;
300 : case BMS_SUBSET2:
301 : /* old path is less-parameterized, keep it */
302 8 : break;
303 : case BMS_DIFFERENT:
304 :
305 : /*
306 : * This means that neither path has the least possible
307 : * parameterization for the rel. We'll sit on the old
308 : * path until something better comes along.
309 : */
310 2900 : break;
311 : }
312 : }
313 : }
314 : else
315 : {
316 : /* Unparameterized path, so consider it for cheapest slots */
317 1261152 : if (cheapest_total_path == NULL)
318 : {
319 1110738 : cheapest_startup_path = cheapest_total_path = path;
320 1110738 : continue;
321 : }
322 :
323 : /*
324 : * If we find two paths of identical costs, try to keep the
325 : * better-sorted one. The paths might have unrelated sort
326 : * orderings, in which case we can only guess which might be
327 : * better to keep, but if one is superior then we definitely
328 : * should keep that one.
329 : */
330 150414 : cmp = compare_path_costs(cheapest_startup_path, path, STARTUP_COST);
331 150414 : if (cmp > 0 ||
332 62 : (cmp == 0 &&
333 62 : compare_pathkeys(cheapest_startup_path->pathkeys,
334 : path->pathkeys) == PATHKEYS_BETTER2))
335 35576 : cheapest_startup_path = path;
336 :
337 150414 : cmp = compare_path_costs(cheapest_total_path, path, TOTAL_COST);
338 150414 : if (cmp > 0 ||
339 0 : (cmp == 0 &&
340 0 : compare_pathkeys(cheapest_total_path->pathkeys,
341 : path->pathkeys) == PATHKEYS_BETTER2))
342 0 : cheapest_total_path = path;
343 : }
344 : }
345 :
346 : /* Add cheapest unparameterized path, if any, to parameterized_paths */
347 1112174 : if (cheapest_total_path)
348 1110738 : parameterized_paths = lcons(cheapest_total_path, parameterized_paths);
349 :
350 : /*
351 : * If there is no unparameterized path, use the best parameterized path as
352 : * cheapest_total_path (but not as cheapest_startup_path).
353 : */
354 1112174 : if (cheapest_total_path == NULL)
355 1436 : cheapest_total_path = best_param_path;
356 : Assert(cheapest_total_path != NULL);
357 :
358 1112174 : parent_rel->cheapest_startup_path = cheapest_startup_path;
359 1112174 : parent_rel->cheapest_total_path = cheapest_total_path;
360 1112174 : parent_rel->cheapest_unique_path = NULL; /* computed only if needed */
361 1112174 : parent_rel->cheapest_parameterized_paths = parameterized_paths;
362 1112174 : }
363 :
364 : /*
365 : * add_path
366 : * Consider a potential implementation path for the specified parent rel,
367 : * and add it to the rel's pathlist if it is worthy of consideration.
368 : * A path is worthy if it has a better sort order (better pathkeys) or
369 : * cheaper cost (on either dimension), or generates fewer rows, than any
370 : * existing path that has the same or superset parameterization rels.
371 : * We also consider parallel-safe paths more worthy than others.
372 : *
373 : * We also remove from the rel's pathlist any old paths that are dominated
374 : * by new_path --- that is, new_path is cheaper, at least as well ordered,
375 : * generates no more rows, requires no outer rels not required by the old
376 : * path, and is no less parallel-safe.
377 : *
378 : * In most cases, a path with a superset parameterization will generate
379 : * fewer rows (since it has more join clauses to apply), so that those two
380 : * figures of merit move in opposite directions; this means that a path of
381 : * one parameterization can seldom dominate a path of another. But such
382 : * cases do arise, so we make the full set of checks anyway.
383 : *
384 : * There are two policy decisions embedded in this function, along with
385 : * its sibling add_path_precheck. First, we treat all parameterized paths
386 : * as having NIL pathkeys, so that they cannot win comparisons on the
387 : * basis of sort order. This is to reduce the number of parameterized
388 : * paths that are kept; see discussion in src/backend/optimizer/README.
389 : *
390 : * Second, we only consider cheap startup cost to be interesting if
391 : * parent_rel->consider_startup is true for an unparameterized path, or
392 : * parent_rel->consider_param_startup is true for a parameterized one.
393 : * Again, this allows discarding useless paths sooner.
394 : *
395 : * The pathlist is kept sorted by total_cost, with cheaper paths
396 : * at the front. Within this routine, that's simply a speed hack:
397 : * doing it that way makes it more likely that we will reject an inferior
398 : * path after a few comparisons, rather than many comparisons.
399 : * However, add_path_precheck relies on this ordering to exit early
400 : * when possible.
401 : *
402 : * NOTE: discarded Path objects are immediately pfree'd to reduce planner
403 : * memory consumption. We dare not try to free the substructure of a Path,
404 : * since much of it may be shared with other Paths or the query tree itself;
405 : * but just recycling discarded Path nodes is a very useful savings in
406 : * a large join tree. We can recycle the List nodes of pathlist, too.
407 : *
408 : * As noted in optimizer/README, deleting a previously-accepted Path is
409 : * safe because we know that Paths of this rel cannot yet be referenced
410 : * from any other rel, such as a higher-level join. However, in some cases
411 : * it is possible that a Path is referenced by another Path for its own
412 : * rel; we must not delete such a Path, even if it is dominated by the new
413 : * Path. Currently this occurs only for IndexPath objects, which may be
414 : * referenced as children of BitmapHeapPaths as well as being paths in
415 : * their own right. Hence, we don't pfree IndexPaths when rejecting them.
416 : *
417 : * 'parent_rel' is the relation entry to which the path corresponds.
418 : * 'new_path' is a potential path for parent_rel.
419 : *
420 : * Returns nothing, but modifies parent_rel->pathlist.
421 : */
422 : void
423 2016876 : add_path(RelOptInfo *parent_rel, Path *new_path)
424 : {
425 2016876 : bool accept_new = true; /* unless we find a superior old path */
426 2016876 : int insert_at = 0; /* where to insert new item */
427 : List *new_path_pathkeys;
428 : ListCell *p1;
429 :
430 : /*
431 : * This is a convenient place to check for query cancel --- no part of the
432 : * planner goes very long without calling add_path().
433 : */
434 2016876 : CHECK_FOR_INTERRUPTS();
435 :
436 : /* Pretend parameterized paths have no pathkeys, per comment above */
437 2016876 : new_path_pathkeys = new_path->param_info ? NIL : new_path->pathkeys;
438 :
439 : /*
440 : * Loop to check proposed new path against old paths. Note it is possible
441 : * for more than one old path to be tossed out because new_path dominates
442 : * it.
443 : */
444 2865318 : foreach(p1, parent_rel->pathlist)
445 : {
446 1546674 : Path *old_path = (Path *) lfirst(p1);
447 1546674 : bool remove_old = false; /* unless new proves superior */
448 : PathCostComparison costcmp;
449 : PathKeysComparison keyscmp;
450 : BMS_Comparison outercmp;
451 :
452 : /*
453 : * Do a fuzzy cost comparison with standard fuzziness limit.
454 : */
455 1546674 : costcmp = compare_path_costs_fuzzily(new_path, old_path,
456 : STD_FUZZ_FACTOR);
457 :
458 : /*
459 : * If the two paths compare differently for startup and total cost,
460 : * then we want to keep both, and we can skip comparing pathkeys and
461 : * required_outer rels. If they compare the same, proceed with the
462 : * other comparisons. Row count is checked last. (We make the tests
463 : * in this order because the cost comparison is most likely to turn
464 : * out "different", and the pathkeys comparison next most likely. As
465 : * explained above, row count very seldom makes a difference, so even
466 : * though it's cheap to compare there's not much point in checking it
467 : * earlier.)
468 : */
469 1546674 : if (costcmp != COSTS_DIFFERENT)
470 : {
471 : /* Similarly check to see if either dominates on pathkeys */
472 : List *old_path_pathkeys;
473 :
474 1527384 : old_path_pathkeys = old_path->param_info ? NIL : old_path->pathkeys;
475 1527384 : keyscmp = compare_pathkeys(new_path_pathkeys,
476 : old_path_pathkeys);
477 1527384 : if (keyscmp != PATHKEYS_DIFFERENT)
478 : {
479 1475688 : switch (costcmp)
480 : {
481 : case COSTS_EQUAL:
482 170560 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
483 170560 : PATH_REQ_OUTER(old_path));
484 169386 : if (keyscmp == PATHKEYS_BETTER1)
485 : {
486 1356 : if ((outercmp == BMS_EQUAL ||
487 1356 : outercmp == BMS_SUBSET1) &&
488 2706 : new_path->rows <= old_path->rows &&
489 1350 : new_path->parallel_safe >= old_path->parallel_safe)
490 1350 : remove_old = true; /* new dominates old */
491 : }
492 168030 : else if (keyscmp == PATHKEYS_BETTER2)
493 : {
494 5496 : if ((outercmp == BMS_EQUAL ||
495 5496 : outercmp == BMS_SUBSET2) &&
496 10980 : new_path->rows >= old_path->rows &&
497 5484 : new_path->parallel_safe <= old_path->parallel_safe)
498 5484 : accept_new = false; /* old dominates new */
499 : }
500 : else /* keyscmp == PATHKEYS_EQUAL */
501 : {
502 162534 : if (outercmp == BMS_EQUAL)
503 : {
504 : /*
505 : * Same pathkeys and outer rels, and fuzzily
506 : * the same cost, so keep just one; to decide
507 : * which, first check parallel-safety, then
508 : * rows, then do a fuzzy cost comparison with
509 : * very small fuzz limit. (We used to do an
510 : * exact cost comparison, but that results in
511 : * annoying platform-specific plan variations
512 : * due to roundoff in the cost estimates.) If
513 : * things are still tied, arbitrarily keep
514 : * only the old path. Notice that we will
515 : * keep only the old path even if the
516 : * less-fuzzy comparison decides the startup
517 : * and total costs compare differently.
518 : */
519 322980 : if (new_path->parallel_safe >
520 161490 : old_path->parallel_safe)
521 32 : remove_old = true; /* new dominates old */
522 322916 : else if (new_path->parallel_safe <
523 161458 : old_path->parallel_safe)
524 68 : accept_new = false; /* old dominates new */
525 161390 : else if (new_path->rows < old_path->rows)
526 4 : remove_old = true; /* new dominates old */
527 161386 : else if (new_path->rows > old_path->rows)
528 12 : accept_new = false; /* old dominates new */
529 161374 : else if (compare_path_costs_fuzzily(new_path,
530 : old_path,
531 : 1.0000000001) == COSTS_BETTER1)
532 5022 : remove_old = true; /* new dominates old */
533 : else
534 156352 : accept_new = false; /* old equals or
535 : * dominates new */
536 : }
537 1314 : else if (outercmp == BMS_SUBSET1 &&
538 536 : new_path->rows <= old_path->rows &&
539 266 : new_path->parallel_safe >= old_path->parallel_safe)
540 266 : remove_old = true; /* new dominates old */
541 1458 : else if (outercmp == BMS_SUBSET2 &&
542 1346 : new_path->rows >= old_path->rows &&
543 666 : new_path->parallel_safe <= old_path->parallel_safe)
544 666 : accept_new = false; /* old dominates new */
545 : /* else different parameterizations, keep both */
546 : }
547 169386 : break;
548 : case COSTS_BETTER1:
549 465164 : if (keyscmp != PATHKEYS_BETTER2)
550 : {
551 328738 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
552 328738 : PATH_REQ_OUTER(old_path));
553 321782 : if ((outercmp == BMS_EQUAL ||
554 275112 : outercmp == BMS_SUBSET1) &&
555 548836 : new_path->rows <= old_path->rows &&
556 273724 : new_path->parallel_safe >= old_path->parallel_safe)
557 272296 : remove_old = true; /* new dominates old */
558 : }
559 465164 : break;
560 : case COSTS_BETTER2:
561 841138 : if (keyscmp != PATHKEYS_BETTER1)
562 : {
563 674364 : outercmp = bms_subset_compare(PATH_REQ_OUTER(new_path),
564 674364 : PATH_REQ_OUTER(old_path));
565 587784 : if ((outercmp == BMS_EQUAL ||
566 558066 : outercmp == BMS_SUBSET2) &&
567 1095256 : new_path->rows >= old_path->rows &&
568 537190 : new_path->parallel_safe <= old_path->parallel_safe)
569 535650 : accept_new = false; /* old dominates new */
570 : }
571 841138 : break;
572 : case COSTS_DIFFERENT:
573 :
574 : /*
575 : * can't get here, but keep this case to keep compiler
576 : * quiet
577 : */
578 0 : break;
579 : }
580 : }
581 : }
582 :
583 : /*
584 : * Remove current element from pathlist if dominated by new.
585 : */
586 1546674 : if (remove_old)
587 : {
588 278970 : parent_rel->pathlist = foreach_delete_current(parent_rel->pathlist,
589 : p1);
590 :
591 : /*
592 : * Delete the data pointed-to by the deleted cell, if possible
593 : */
594 278970 : if (!IsA(old_path, IndexPath))
595 266268 : pfree(old_path);
596 : }
597 : else
598 : {
599 : /* new belongs after this old path if it has cost >= old's */
600 1267704 : if (new_path->total_cost >= old_path->total_cost)
601 1019860 : insert_at = foreach_current_index(p1) + 1;
602 : }
603 :
604 : /*
605 : * If we found an old path that dominates new_path, we can quit
606 : * scanning the pathlist; we will not add new_path, and we assume
607 : * new_path cannot dominate any other elements of the pathlist.
608 : */
609 1546674 : if (!accept_new)
610 698232 : break;
611 : }
612 :
613 2016876 : if (accept_new)
614 : {
615 : /* Accept the new path: insert it at proper place in pathlist */
616 1318644 : parent_rel->pathlist =
617 1318644 : list_insert_nth(parent_rel->pathlist, insert_at, new_path);
618 : }
619 : else
620 : {
621 : /* Reject and recycle the new path */
622 698232 : if (!IsA(new_path, IndexPath))
623 649194 : pfree(new_path);
624 : }
625 2016876 : }
626 :
627 : /*
628 : * add_path_precheck
629 : * Check whether a proposed new path could possibly get accepted.
630 : * We assume we know the path's pathkeys and parameterization accurately,
631 : * and have lower bounds for its costs.
632 : *
633 : * Note that we do not know the path's rowcount, since getting an estimate for
634 : * that is too expensive to do before prechecking. We assume here that paths
635 : * of a superset parameterization will generate fewer rows; if that holds,
636 : * then paths with different parameterizations cannot dominate each other
637 : * and so we can simply ignore existing paths of another parameterization.
638 : * (In the infrequent cases where that rule of thumb fails, add_path will
639 : * get rid of the inferior path.)
640 : *
641 : * At the time this is called, we haven't actually built a Path structure,
642 : * so the required information has to be passed piecemeal.
643 : */
644 : bool
645 1757284 : add_path_precheck(RelOptInfo *parent_rel,
646 : Cost startup_cost, Cost total_cost,
647 : List *pathkeys, Relids required_outer)
648 : {
649 : List *new_path_pathkeys;
650 : bool consider_startup;
651 : ListCell *p1;
652 :
653 : /* Pretend parameterized paths have no pathkeys, per add_path policy */
654 1757284 : new_path_pathkeys = required_outer ? NIL : pathkeys;
655 :
656 : /* Decide whether new path's startup cost is interesting */
657 1757284 : consider_startup = required_outer ? parent_rel->consider_param_startup : parent_rel->consider_startup;
658 :
659 2166988 : foreach(p1, parent_rel->pathlist)
660 : {
661 2045586 : Path *old_path = (Path *) lfirst(p1);
662 : PathKeysComparison keyscmp;
663 :
664 : /*
665 : * We are looking for an old_path with the same parameterization (and
666 : * by assumption the same rowcount) that dominates the new path on
667 : * pathkeys as well as both cost metrics. If we find one, we can
668 : * reject the new path.
669 : *
670 : * Cost comparisons here should match compare_path_costs_fuzzily.
671 : */
672 2045586 : if (total_cost > old_path->total_cost * STD_FUZZ_FACTOR)
673 : {
674 : /* new path can win on startup cost only if consider_startup */
675 2032434 : if (startup_cost > old_path->startup_cost * STD_FUZZ_FACTOR ||
676 681806 : !consider_startup)
677 : {
678 : /* new path loses on cost, so check pathkeys... */
679 : List *old_path_pathkeys;
680 :
681 1339700 : old_path_pathkeys = old_path->param_info ? NIL : old_path->pathkeys;
682 1339700 : keyscmp = compare_pathkeys(new_path_pathkeys,
683 : old_path_pathkeys);
684 1339700 : if (keyscmp == PATHKEYS_EQUAL ||
685 : keyscmp == PATHKEYS_BETTER2)
686 : {
687 : /* new path does not win on pathkeys... */
688 952602 : if (bms_equal(required_outer, PATH_REQ_OUTER(old_path)))
689 : {
690 : /* Found an old path that dominates the new one */
691 940924 : return false;
692 : }
693 : }
694 : }
695 : }
696 : else
697 : {
698 : /*
699 : * Since the pathlist is sorted by total_cost, we can stop looking
700 : * once we reach a path with a total_cost larger than the new
701 : * path's.
702 : */
703 694958 : break;
704 : }
705 : }
706 :
707 816360 : return true;
708 : }
709 :
710 : /*
711 : * add_partial_path
712 : * Like add_path, our goal here is to consider whether a path is worthy
713 : * of being kept around, but the considerations here are a bit different.
714 : * A partial path is one which can be executed in any number of workers in
715 : * parallel such that each worker will generate a subset of the path's
716 : * overall result.
717 : *
718 : * As in add_path, the partial_pathlist is kept sorted with the cheapest
719 : * total path in front. This is depended on by multiple places, which
720 : * just take the front entry as the cheapest path without searching.
721 : *
722 : * We don't generate parameterized partial paths for several reasons. Most
723 : * importantly, they're not safe to execute, because there's nothing to
724 : * make sure that a parallel scan within the parameterized portion of the
725 : * plan is running with the same value in every worker at the same time.
726 : * Fortunately, it seems unlikely to be worthwhile anyway, because having
727 : * each worker scan the entire outer relation and a subset of the inner
728 : * relation will generally be a terrible plan. The inner (parameterized)
729 : * side of the plan will be small anyway. There could be rare cases where
730 : * this wins big - e.g. if join order constraints put a 1-row relation on
731 : * the outer side of the topmost join with a parameterized plan on the inner
732 : * side - but we'll have to be content not to handle such cases until
733 : * somebody builds an executor infrastructure that can cope with them.
734 : *
735 : * Because we don't consider parameterized paths here, we also don't
736 : * need to consider the row counts as a measure of quality: every path will
737 : * produce the same number of rows. Neither do we need to consider startup
738 : * costs: parallelism is only used for plans that will be run to completion.
739 : * Therefore, this routine is much simpler than add_path: it needs to
740 : * consider only pathkeys and total cost.
741 : *
742 : * As with add_path, we pfree paths that are found to be dominated by
743 : * another partial path; this requires that there be no other references to
744 : * such paths yet. Hence, GatherPaths must not be created for a rel until
745 : * we're done creating all partial paths for it. Unlike add_path, we don't
746 : * take an exception for IndexPaths as partial index paths won't be
747 : * referenced by partial BitmapHeapPaths.
748 : */
749 : void
750 34004 : add_partial_path(RelOptInfo *parent_rel, Path *new_path)
751 : {
752 34004 : bool accept_new = true; /* unless we find a superior old path */
753 34004 : int insert_at = 0; /* where to insert new item */
754 : ListCell *p1;
755 :
756 : /* Check for query cancel. */
757 34004 : CHECK_FOR_INTERRUPTS();
758 :
759 : /* Path to be added must be parallel safe. */
760 : Assert(new_path->parallel_safe);
761 :
762 : /* Relation should be OK for parallelism, too. */
763 : Assert(parent_rel->consider_parallel);
764 :
765 : /*
766 : * As in add_path, throw out any paths which are dominated by the new
767 : * path, but throw out the new path if some existing path dominates it.
768 : */
769 42012 : foreach(p1, parent_rel->partial_pathlist)
770 : {
771 11456 : Path *old_path = (Path *) lfirst(p1);
772 11456 : bool remove_old = false; /* unless new proves superior */
773 : PathKeysComparison keyscmp;
774 :
775 : /* Compare pathkeys. */
776 11456 : keyscmp = compare_pathkeys(new_path->pathkeys, old_path->pathkeys);
777 :
778 : /* Unless pathkeys are incompatible, keep just one of the two paths. */
779 11456 : if (keyscmp != PATHKEYS_DIFFERENT)
780 : {
781 11456 : if (new_path->total_cost > old_path->total_cost * STD_FUZZ_FACTOR)
782 : {
783 : /* New path costs more; keep it only if pathkeys are better. */
784 5654 : if (keyscmp != PATHKEYS_BETTER1)
785 2634 : accept_new = false;
786 : }
787 11604 : else if (old_path->total_cost > new_path->total_cost
788 5802 : * STD_FUZZ_FACTOR)
789 : {
790 : /* Old path costs more; keep it only if pathkeys are better. */
791 4824 : if (keyscmp != PATHKEYS_BETTER2)
792 2654 : remove_old = true;
793 : }
794 978 : else if (keyscmp == PATHKEYS_BETTER1)
795 : {
796 : /* Costs are about the same, new path has better pathkeys. */
797 0 : remove_old = true;
798 : }
799 978 : else if (keyscmp == PATHKEYS_BETTER2)
800 : {
801 : /* Costs are about the same, old path has better pathkeys. */
802 112 : accept_new = false;
803 : }
804 866 : else if (old_path->total_cost > new_path->total_cost * 1.0000000001)
805 : {
806 : /* Pathkeys are the same, and the old path costs more. */
807 164 : remove_old = true;
808 : }
809 : else
810 : {
811 : /*
812 : * Pathkeys are the same, and new path isn't materially
813 : * cheaper.
814 : */
815 702 : accept_new = false;
816 : }
817 : }
818 :
819 : /*
820 : * Remove current element from partial_pathlist if dominated by new.
821 : */
822 11456 : if (remove_old)
823 : {
824 2818 : parent_rel->partial_pathlist =
825 2818 : foreach_delete_current(parent_rel->partial_pathlist, p1);
826 2818 : pfree(old_path);
827 : }
828 : else
829 : {
830 : /* new belongs after this old path if it has cost >= old's */
831 8638 : if (new_path->total_cost >= old_path->total_cost)
832 6428 : insert_at = foreach_current_index(p1) + 1;
833 : }
834 :
835 : /*
836 : * If we found an old path that dominates new_path, we can quit
837 : * scanning the partial_pathlist; we will not add new_path, and we
838 : * assume new_path cannot dominate any later path.
839 : */
840 11456 : if (!accept_new)
841 3448 : break;
842 : }
843 :
844 34004 : if (accept_new)
845 : {
846 : /* Accept the new path: insert it at proper place */
847 30556 : parent_rel->partial_pathlist =
848 30556 : list_insert_nth(parent_rel->partial_pathlist, insert_at, new_path);
849 : }
850 : else
851 : {
852 : /* Reject and recycle the new path */
853 3448 : pfree(new_path);
854 : }
855 34004 : }
856 :
857 : /*
858 : * add_partial_path_precheck
859 : * Check whether a proposed new partial path could possibly get accepted.
860 : *
861 : * Unlike add_path_precheck, we can ignore startup cost and parameterization,
862 : * since they don't matter for partial paths (see add_partial_path). But
863 : * we do want to make sure we don't add a partial path if there's already
864 : * a complete path that dominates it, since in that case the proposed path
865 : * is surely a loser.
866 : */
867 : bool
868 19140 : add_partial_path_precheck(RelOptInfo *parent_rel, Cost total_cost,
869 : List *pathkeys)
870 : {
871 : ListCell *p1;
872 :
873 : /*
874 : * Our goal here is twofold. First, we want to find out whether this path
875 : * is clearly inferior to some existing partial path. If so, we want to
876 : * reject it immediately. Second, we want to find out whether this path
877 : * is clearly superior to some existing partial path -- at least, modulo
878 : * final cost computations. If so, we definitely want to consider it.
879 : *
880 : * Unlike add_path(), we always compare pathkeys here. This is because we
881 : * expect partial_pathlist to be very short, and getting a definitive
882 : * answer at this stage avoids the need to call add_path_precheck.
883 : */
884 24174 : foreach(p1, parent_rel->partial_pathlist)
885 : {
886 20364 : Path *old_path = (Path *) lfirst(p1);
887 : PathKeysComparison keyscmp;
888 :
889 20364 : keyscmp = compare_pathkeys(pathkeys, old_path->pathkeys);
890 20364 : if (keyscmp != PATHKEYS_DIFFERENT)
891 : {
892 20364 : if (total_cost > old_path->total_cost * STD_FUZZ_FACTOR &&
893 : keyscmp != PATHKEYS_BETTER1)
894 27774 : return false;
895 7920 : if (old_path->total_cost > total_cost * STD_FUZZ_FACTOR &&
896 : keyscmp != PATHKEYS_BETTER2)
897 2886 : return true;
898 : }
899 : }
900 :
901 : /*
902 : * This path is neither clearly inferior to an existing partial path nor
903 : * clearly good enough that it might replace one. Compare it to
904 : * non-parallel plans. If it loses even before accounting for the cost of
905 : * the Gather node, we should definitely reject it.
906 : *
907 : * Note that we pass the total_cost to add_path_precheck twice. This is
908 : * because it's never advantageous to consider the startup cost of a
909 : * partial path; the resulting plans, if run in parallel, will be run to
910 : * completion.
911 : */
912 3810 : if (!add_path_precheck(parent_rel, total_cost, total_cost, pathkeys,
913 : NULL))
914 236 : return false;
915 :
916 3574 : return true;
917 : }
918 :
919 :
920 : /*****************************************************************************
921 : * PATH NODE CREATION ROUTINES
922 : *****************************************************************************/
923 :
924 : /*
925 : * create_seqscan_path
926 : * Creates a path corresponding to a sequential scan, returning the
927 : * pathnode.
928 : */
929 : Path *
930 229156 : create_seqscan_path(PlannerInfo *root, RelOptInfo *rel,
931 : Relids required_outer, int parallel_workers)
932 : {
933 229156 : Path *pathnode = makeNode(Path);
934 :
935 229156 : pathnode->pathtype = T_SeqScan;
936 229156 : pathnode->parent = rel;
937 229156 : pathnode->pathtarget = rel->reltarget;
938 229156 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
939 : required_outer);
940 229156 : pathnode->parallel_aware = parallel_workers > 0 ? true : false;
941 229156 : pathnode->parallel_safe = rel->consider_parallel;
942 229156 : pathnode->parallel_workers = parallel_workers;
943 229156 : pathnode->pathkeys = NIL; /* seqscan has unordered result */
944 :
945 229156 : cost_seqscan(pathnode, root, rel, pathnode->param_info);
946 :
947 229156 : return pathnode;
948 : }
949 :
950 : /*
951 : * create_samplescan_path
952 : * Creates a path node for a sampled table scan.
953 : */
954 : Path *
955 180 : create_samplescan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
956 : {
957 180 : Path *pathnode = makeNode(Path);
958 :
959 180 : pathnode->pathtype = T_SampleScan;
960 180 : pathnode->parent = rel;
961 180 : pathnode->pathtarget = rel->reltarget;
962 180 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
963 : required_outer);
964 180 : pathnode->parallel_aware = false;
965 180 : pathnode->parallel_safe = rel->consider_parallel;
966 180 : pathnode->parallel_workers = 0;
967 180 : pathnode->pathkeys = NIL; /* samplescan has unordered result */
968 :
969 180 : cost_samplescan(pathnode, root, rel, pathnode->param_info);
970 :
971 180 : return pathnode;
972 : }
973 :
974 : /*
975 : * create_index_path
976 : * Creates a path node for an index scan.
977 : *
978 : * 'index' is a usable index.
979 : * 'indexclauses' is a list of IndexClause nodes representing clauses
980 : * to be enforced as qual conditions in the scan.
981 : * 'indexorderbys' is a list of bare expressions (no RestrictInfos)
982 : * to be used as index ordering operators in the scan.
983 : * 'indexorderbycols' is an integer list of index column numbers (zero based)
984 : * the ordering operators can be used with.
985 : * 'pathkeys' describes the ordering of the path.
986 : * 'indexscandir' is ForwardScanDirection or BackwardScanDirection
987 : * for an ordered index, or NoMovementScanDirection for
988 : * an unordered index.
989 : * 'indexonly' is true if an index-only scan is wanted.
990 : * 'required_outer' is the set of outer relids for a parameterized path.
991 : * 'loop_count' is the number of repetitions of the indexscan to factor into
992 : * estimates of caching behavior.
993 : * 'partial_path' is true if constructing a parallel index scan path.
994 : *
995 : * Returns the new path node.
996 : */
997 : IndexPath *
998 368808 : create_index_path(PlannerInfo *root,
999 : IndexOptInfo *index,
1000 : List *indexclauses,
1001 : List *indexorderbys,
1002 : List *indexorderbycols,
1003 : List *pathkeys,
1004 : ScanDirection indexscandir,
1005 : bool indexonly,
1006 : Relids required_outer,
1007 : double loop_count,
1008 : bool partial_path)
1009 : {
1010 368808 : IndexPath *pathnode = makeNode(IndexPath);
1011 368808 : RelOptInfo *rel = index->rel;
1012 :
1013 368808 : pathnode->path.pathtype = indexonly ? T_IndexOnlyScan : T_IndexScan;
1014 368808 : pathnode->path.parent = rel;
1015 368808 : pathnode->path.pathtarget = rel->reltarget;
1016 368808 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1017 : required_outer);
1018 368808 : pathnode->path.parallel_aware = false;
1019 368808 : pathnode->path.parallel_safe = rel->consider_parallel;
1020 368808 : pathnode->path.parallel_workers = 0;
1021 368808 : pathnode->path.pathkeys = pathkeys;
1022 :
1023 368808 : pathnode->indexinfo = index;
1024 368808 : pathnode->indexclauses = indexclauses;
1025 368808 : pathnode->indexorderbys = indexorderbys;
1026 368808 : pathnode->indexorderbycols = indexorderbycols;
1027 368808 : pathnode->indexscandir = indexscandir;
1028 :
1029 368808 : cost_index(pathnode, root, loop_count, partial_path);
1030 :
1031 368808 : return pathnode;
1032 : }
1033 :
1034 : /*
1035 : * create_bitmap_heap_path
1036 : * Creates a path node for a bitmap scan.
1037 : *
1038 : * 'bitmapqual' is a tree of IndexPath, BitmapAndPath, and BitmapOrPath nodes.
1039 : * 'required_outer' is the set of outer relids for a parameterized path.
1040 : * 'loop_count' is the number of repetitions of the indexscan to factor into
1041 : * estimates of caching behavior.
1042 : *
1043 : * loop_count should match the value used when creating the component
1044 : * IndexPaths.
1045 : */
1046 : BitmapHeapPath *
1047 174014 : create_bitmap_heap_path(PlannerInfo *root,
1048 : RelOptInfo *rel,
1049 : Path *bitmapqual,
1050 : Relids required_outer,
1051 : double loop_count,
1052 : int parallel_degree)
1053 : {
1054 174014 : BitmapHeapPath *pathnode = makeNode(BitmapHeapPath);
1055 :
1056 174014 : pathnode->path.pathtype = T_BitmapHeapScan;
1057 174014 : pathnode->path.parent = rel;
1058 174014 : pathnode->path.pathtarget = rel->reltarget;
1059 174014 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1060 : required_outer);
1061 174014 : pathnode->path.parallel_aware = parallel_degree > 0 ? true : false;
1062 174014 : pathnode->path.parallel_safe = rel->consider_parallel;
1063 174014 : pathnode->path.parallel_workers = parallel_degree;
1064 174014 : pathnode->path.pathkeys = NIL; /* always unordered */
1065 :
1066 174014 : pathnode->bitmapqual = bitmapqual;
1067 :
1068 174014 : cost_bitmap_heap_scan(&pathnode->path, root, rel,
1069 : pathnode->path.param_info,
1070 : bitmapqual, loop_count);
1071 :
1072 174014 : return pathnode;
1073 : }
1074 :
1075 : /*
1076 : * create_bitmap_and_path
1077 : * Creates a path node representing a BitmapAnd.
1078 : */
1079 : BitmapAndPath *
1080 48 : create_bitmap_and_path(PlannerInfo *root,
1081 : RelOptInfo *rel,
1082 : List *bitmapquals)
1083 : {
1084 48 : BitmapAndPath *pathnode = makeNode(BitmapAndPath);
1085 :
1086 48 : pathnode->path.pathtype = T_BitmapAnd;
1087 48 : pathnode->path.parent = rel;
1088 48 : pathnode->path.pathtarget = rel->reltarget;
1089 48 : pathnode->path.param_info = NULL; /* not used in bitmap trees */
1090 :
1091 : /*
1092 : * Currently, a BitmapHeapPath, BitmapAndPath, or BitmapOrPath will be
1093 : * parallel-safe if and only if rel->consider_parallel is set. So, we can
1094 : * set the flag for this path based only on the relation-level flag,
1095 : * without actually iterating over the list of children.
1096 : */
1097 48 : pathnode->path.parallel_aware = false;
1098 48 : pathnode->path.parallel_safe = rel->consider_parallel;
1099 48 : pathnode->path.parallel_workers = 0;
1100 :
1101 48 : pathnode->path.pathkeys = NIL; /* always unordered */
1102 :
1103 48 : pathnode->bitmapquals = bitmapquals;
1104 :
1105 : /* this sets bitmapselectivity as well as the regular cost fields: */
1106 48 : cost_bitmap_and_node(pathnode, root);
1107 :
1108 48 : return pathnode;
1109 : }
1110 :
1111 : /*
1112 : * create_bitmap_or_path
1113 : * Creates a path node representing a BitmapOr.
1114 : */
1115 : BitmapOrPath *
1116 318 : create_bitmap_or_path(PlannerInfo *root,
1117 : RelOptInfo *rel,
1118 : List *bitmapquals)
1119 : {
1120 318 : BitmapOrPath *pathnode = makeNode(BitmapOrPath);
1121 :
1122 318 : pathnode->path.pathtype = T_BitmapOr;
1123 318 : pathnode->path.parent = rel;
1124 318 : pathnode->path.pathtarget = rel->reltarget;
1125 318 : pathnode->path.param_info = NULL; /* not used in bitmap trees */
1126 :
1127 : /*
1128 : * Currently, a BitmapHeapPath, BitmapAndPath, or BitmapOrPath will be
1129 : * parallel-safe if and only if rel->consider_parallel is set. So, we can
1130 : * set the flag for this path based only on the relation-level flag,
1131 : * without actually iterating over the list of children.
1132 : */
1133 318 : pathnode->path.parallel_aware = false;
1134 318 : pathnode->path.parallel_safe = rel->consider_parallel;
1135 318 : pathnode->path.parallel_workers = 0;
1136 :
1137 318 : pathnode->path.pathkeys = NIL; /* always unordered */
1138 :
1139 318 : pathnode->bitmapquals = bitmapquals;
1140 :
1141 : /* this sets bitmapselectivity as well as the regular cost fields: */
1142 318 : cost_bitmap_or_node(pathnode, root);
1143 :
1144 318 : return pathnode;
1145 : }
1146 :
1147 : /*
1148 : * create_tidscan_path
1149 : * Creates a path corresponding to a scan by TID, returning the pathnode.
1150 : */
1151 : TidPath *
1152 594 : create_tidscan_path(PlannerInfo *root, RelOptInfo *rel, List *tidquals,
1153 : Relids required_outer)
1154 : {
1155 594 : TidPath *pathnode = makeNode(TidPath);
1156 :
1157 594 : pathnode->path.pathtype = T_TidScan;
1158 594 : pathnode->path.parent = rel;
1159 594 : pathnode->path.pathtarget = rel->reltarget;
1160 594 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1161 : required_outer);
1162 594 : pathnode->path.parallel_aware = false;
1163 594 : pathnode->path.parallel_safe = rel->consider_parallel;
1164 594 : pathnode->path.parallel_workers = 0;
1165 594 : pathnode->path.pathkeys = NIL; /* always unordered */
1166 :
1167 594 : pathnode->tidquals = tidquals;
1168 :
1169 594 : cost_tidscan(&pathnode->path, root, rel, tidquals,
1170 : pathnode->path.param_info);
1171 :
1172 594 : return pathnode;
1173 : }
1174 :
1175 : /*
1176 : * create_append_path
1177 : * Creates a path corresponding to an Append plan, returning the
1178 : * pathnode.
1179 : *
1180 : * Note that we must handle subpaths = NIL, representing a dummy access path.
1181 : * Also, there are callers that pass root = NULL.
1182 : */
1183 : AppendPath *
1184 28238 : create_append_path(PlannerInfo *root,
1185 : RelOptInfo *rel,
1186 : List *subpaths, List *partial_subpaths,
1187 : List *pathkeys, Relids required_outer,
1188 : int parallel_workers, bool parallel_aware,
1189 : List *partitioned_rels, double rows)
1190 : {
1191 28238 : AppendPath *pathnode = makeNode(AppendPath);
1192 : ListCell *l;
1193 :
1194 : Assert(!parallel_aware || parallel_workers > 0);
1195 :
1196 28238 : pathnode->path.pathtype = T_Append;
1197 28238 : pathnode->path.parent = rel;
1198 28238 : pathnode->path.pathtarget = rel->reltarget;
1199 :
1200 : /*
1201 : * When generating an Append path for a partitioned table, there may be
1202 : * parameters that are useful so we can eliminate certain partitions
1203 : * during execution. Here we'll go all the way and fully populate the
1204 : * parameter info data as we do for normal base relations. However, we
1205 : * need only bother doing this for RELOPT_BASEREL rels, as
1206 : * RELOPT_OTHER_MEMBER_REL's Append paths are merged into the base rel's
1207 : * Append subpaths. It would do no harm to do this, we just avoid it to
1208 : * save wasting effort.
1209 : */
1210 28238 : if (partitioned_rels != NIL && root && rel->reloptkind == RELOPT_BASEREL)
1211 11398 : pathnode->path.param_info = get_baserel_parampathinfo(root,
1212 : rel,
1213 : required_outer);
1214 : else
1215 16840 : pathnode->path.param_info = get_appendrel_parampathinfo(rel,
1216 : required_outer);
1217 :
1218 28238 : pathnode->path.parallel_aware = parallel_aware;
1219 28238 : pathnode->path.parallel_safe = rel->consider_parallel;
1220 28238 : pathnode->path.parallel_workers = parallel_workers;
1221 28238 : pathnode->path.pathkeys = pathkeys;
1222 28238 : pathnode->partitioned_rels = list_copy(partitioned_rels);
1223 :
1224 : /*
1225 : * For parallel append, non-partial paths are sorted by descending total
1226 : * costs. That way, the total time to finish all non-partial paths is
1227 : * minimized. Also, the partial paths are sorted by descending startup
1228 : * costs. There may be some paths that require to do startup work by a
1229 : * single worker. In such case, it's better for workers to choose the
1230 : * expensive ones first, whereas the leader should choose the cheapest
1231 : * startup plan.
1232 : */
1233 28238 : if (pathnode->path.parallel_aware)
1234 : {
1235 : /*
1236 : * We mustn't fiddle with the order of subpaths when the Append has
1237 : * pathkeys. The order they're listed in is critical to keeping the
1238 : * pathkeys valid.
1239 : */
1240 : Assert(pathkeys == NIL);
1241 :
1242 9656 : list_sort(subpaths, append_total_cost_compare);
1243 9656 : list_sort(partial_subpaths, append_startup_cost_compare);
1244 : }
1245 28238 : pathnode->first_partial_path = list_length(subpaths);
1246 28238 : pathnode->subpaths = list_concat(subpaths, partial_subpaths);
1247 :
1248 : /*
1249 : * Apply query-wide LIMIT if known and path is for sole base relation.
1250 : * (Handling this at this low level is a bit klugy.)
1251 : */
1252 28238 : if (root != NULL && bms_equal(rel->relids, root->all_baserels))
1253 13730 : pathnode->limit_tuples = root->limit_tuples;
1254 : else
1255 14508 : pathnode->limit_tuples = -1.0;
1256 :
1257 103336 : foreach(l, pathnode->subpaths)
1258 : {
1259 75098 : Path *subpath = (Path *) lfirst(l);
1260 :
1261 136202 : pathnode->path.parallel_safe = pathnode->path.parallel_safe &&
1262 61104 : subpath->parallel_safe;
1263 :
1264 : /* All child paths must have same parameterization */
1265 : Assert(bms_equal(PATH_REQ_OUTER(subpath), required_outer));
1266 : }
1267 :
1268 : Assert(!parallel_aware || pathnode->path.parallel_safe);
1269 :
1270 : /*
1271 : * If there's exactly one child path, the Append is a no-op and will be
1272 : * discarded later (in setrefs.c); therefore, we can inherit the child's
1273 : * size and cost, as well as its pathkeys if any (overriding whatever the
1274 : * caller might've said). Otherwise, we must do the normal costsize
1275 : * calculation.
1276 : */
1277 28238 : if (list_length(pathnode->subpaths) == 1)
1278 : {
1279 4778 : Path *child = (Path *) linitial(pathnode->subpaths);
1280 :
1281 4778 : pathnode->path.rows = child->rows;
1282 4778 : pathnode->path.startup_cost = child->startup_cost;
1283 4778 : pathnode->path.total_cost = child->total_cost;
1284 4778 : pathnode->path.pathkeys = child->pathkeys;
1285 : }
1286 : else
1287 23460 : cost_append(pathnode);
1288 :
1289 : /* If the caller provided a row estimate, override the computed value. */
1290 28238 : if (rows >= 0)
1291 244 : pathnode->path.rows = rows;
1292 :
1293 28238 : return pathnode;
1294 : }
1295 :
1296 : /*
1297 : * append_total_cost_compare
1298 : * list_sort comparator for sorting append child paths
1299 : * by total_cost descending
1300 : *
1301 : * For equal total costs, we fall back to comparing startup costs; if those
1302 : * are equal too, break ties using bms_compare on the paths' relids.
1303 : * (This is to avoid getting unpredictable results from list_sort.)
1304 : */
1305 : static int
1306 2068 : append_total_cost_compare(const ListCell *a, const ListCell *b)
1307 : {
1308 2068 : Path *path1 = (Path *) lfirst(a);
1309 2068 : Path *path2 = (Path *) lfirst(b);
1310 : int cmp;
1311 :
1312 2068 : cmp = compare_path_costs(path1, path2, TOTAL_COST);
1313 2068 : if (cmp != 0)
1314 1660 : return -cmp;
1315 408 : return bms_compare(path1->parent->relids, path2->parent->relids);
1316 : }
1317 :
1318 : /*
1319 : * append_startup_cost_compare
1320 : * list_sort comparator for sorting append child paths
1321 : * by startup_cost descending
1322 : *
1323 : * For equal startup costs, we fall back to comparing total costs; if those
1324 : * are equal too, break ties using bms_compare on the paths' relids.
1325 : * (This is to avoid getting unpredictable results from list_sort.)
1326 : */
1327 : static int
1328 17646 : append_startup_cost_compare(const ListCell *a, const ListCell *b)
1329 : {
1330 17646 : Path *path1 = (Path *) lfirst(a);
1331 17646 : Path *path2 = (Path *) lfirst(b);
1332 : int cmp;
1333 :
1334 17646 : cmp = compare_path_costs(path1, path2, STARTUP_COST);
1335 17646 : if (cmp != 0)
1336 4026 : return -cmp;
1337 13620 : return bms_compare(path1->parent->relids, path2->parent->relids);
1338 : }
1339 :
1340 : /*
1341 : * create_merge_append_path
1342 : * Creates a path corresponding to a MergeAppend plan, returning the
1343 : * pathnode.
1344 : */
1345 : MergeAppendPath *
1346 1344 : create_merge_append_path(PlannerInfo *root,
1347 : RelOptInfo *rel,
1348 : List *subpaths,
1349 : List *pathkeys,
1350 : Relids required_outer,
1351 : List *partitioned_rels)
1352 : {
1353 1344 : MergeAppendPath *pathnode = makeNode(MergeAppendPath);
1354 : Cost input_startup_cost;
1355 : Cost input_total_cost;
1356 : ListCell *l;
1357 :
1358 1344 : pathnode->path.pathtype = T_MergeAppend;
1359 1344 : pathnode->path.parent = rel;
1360 1344 : pathnode->path.pathtarget = rel->reltarget;
1361 1344 : pathnode->path.param_info = get_appendrel_parampathinfo(rel,
1362 : required_outer);
1363 1344 : pathnode->path.parallel_aware = false;
1364 1344 : pathnode->path.parallel_safe = rel->consider_parallel;
1365 1344 : pathnode->path.parallel_workers = 0;
1366 1344 : pathnode->path.pathkeys = pathkeys;
1367 1344 : pathnode->partitioned_rels = list_copy(partitioned_rels);
1368 1344 : pathnode->subpaths = subpaths;
1369 :
1370 : /*
1371 : * Apply query-wide LIMIT if known and path is for sole base relation.
1372 : * (Handling this at this low level is a bit klugy.)
1373 : */
1374 1344 : if (bms_equal(rel->relids, root->all_baserels))
1375 488 : pathnode->limit_tuples = root->limit_tuples;
1376 : else
1377 856 : pathnode->limit_tuples = -1.0;
1378 :
1379 : /*
1380 : * Add up the sizes and costs of the input paths.
1381 : */
1382 1344 : pathnode->path.rows = 0;
1383 1344 : input_startup_cost = 0;
1384 1344 : input_total_cost = 0;
1385 5158 : foreach(l, subpaths)
1386 : {
1387 3814 : Path *subpath = (Path *) lfirst(l);
1388 :
1389 3814 : pathnode->path.rows += subpath->rows;
1390 6378 : pathnode->path.parallel_safe = pathnode->path.parallel_safe &&
1391 2564 : subpath->parallel_safe;
1392 :
1393 3814 : if (pathkeys_contained_in(pathkeys, subpath->pathkeys))
1394 : {
1395 : /* Subpath is adequately ordered, we won't need to sort it */
1396 3674 : input_startup_cost += subpath->startup_cost;
1397 3674 : input_total_cost += subpath->total_cost;
1398 : }
1399 : else
1400 : {
1401 : /* We'll need to insert a Sort node, so include cost for that */
1402 : Path sort_path; /* dummy for result of cost_sort */
1403 :
1404 420 : cost_sort(&sort_path,
1405 : root,
1406 : pathkeys,
1407 : subpath->total_cost,
1408 140 : subpath->parent->tuples,
1409 140 : subpath->pathtarget->width,
1410 : 0.0,
1411 : work_mem,
1412 : pathnode->limit_tuples);
1413 140 : input_startup_cost += sort_path.startup_cost;
1414 140 : input_total_cost += sort_path.total_cost;
1415 : }
1416 :
1417 : /* All child paths must have same parameterization */
1418 : Assert(bms_equal(PATH_REQ_OUTER(subpath), required_outer));
1419 : }
1420 :
1421 : /*
1422 : * Now we can compute total costs of the MergeAppend. If there's exactly
1423 : * one child path, the MergeAppend is a no-op and will be discarded later
1424 : * (in setrefs.c); otherwise we do the normal cost calculation.
1425 : */
1426 1344 : if (list_length(subpaths) == 1)
1427 : {
1428 38 : pathnode->path.startup_cost = input_startup_cost;
1429 38 : pathnode->path.total_cost = input_total_cost;
1430 : }
1431 : else
1432 1306 : cost_merge_append(&pathnode->path, root,
1433 : pathkeys, list_length(subpaths),
1434 : input_startup_cost, input_total_cost,
1435 : pathnode->path.rows);
1436 :
1437 1344 : return pathnode;
1438 : }
1439 :
1440 : /*
1441 : * create_group_result_path
1442 : * Creates a path representing a Result-and-nothing-else plan.
1443 : *
1444 : * This is only used for degenerate grouping cases, in which we know we
1445 : * need to produce one result row, possibly filtered by a HAVING qual.
1446 : */
1447 : GroupResultPath *
1448 110620 : create_group_result_path(PlannerInfo *root, RelOptInfo *rel,
1449 : PathTarget *target, List *havingqual)
1450 : {
1451 110620 : GroupResultPath *pathnode = makeNode(GroupResultPath);
1452 :
1453 110620 : pathnode->path.pathtype = T_Result;
1454 110620 : pathnode->path.parent = rel;
1455 110620 : pathnode->path.pathtarget = target;
1456 110620 : pathnode->path.param_info = NULL; /* there are no other rels... */
1457 110620 : pathnode->path.parallel_aware = false;
1458 110620 : pathnode->path.parallel_safe = rel->consider_parallel;
1459 110620 : pathnode->path.parallel_workers = 0;
1460 110620 : pathnode->path.pathkeys = NIL;
1461 110620 : pathnode->quals = havingqual;
1462 :
1463 : /*
1464 : * We can't quite use cost_resultscan() because the quals we want to
1465 : * account for are not baserestrict quals of the rel. Might as well just
1466 : * hack it here.
1467 : */
1468 110620 : pathnode->path.rows = 1;
1469 110620 : pathnode->path.startup_cost = target->cost.startup;
1470 221240 : pathnode->path.total_cost = target->cost.startup +
1471 110620 : cpu_tuple_cost + target->cost.per_tuple;
1472 :
1473 : /*
1474 : * Add cost of qual, if any --- but we ignore its selectivity, since our
1475 : * rowcount estimate should be 1 no matter what the qual is.
1476 : */
1477 110620 : if (havingqual)
1478 : {
1479 : QualCost qual_cost;
1480 :
1481 276 : cost_qual_eval(&qual_cost, havingqual, root);
1482 : /* havingqual is evaluated once at startup */
1483 276 : pathnode->path.startup_cost += qual_cost.startup + qual_cost.per_tuple;
1484 276 : pathnode->path.total_cost += qual_cost.startup + qual_cost.per_tuple;
1485 : }
1486 :
1487 110620 : return pathnode;
1488 : }
1489 :
1490 : /*
1491 : * create_material_path
1492 : * Creates a path corresponding to a Material plan, returning the
1493 : * pathnode.
1494 : */
1495 : MaterialPath *
1496 215092 : create_material_path(RelOptInfo *rel, Path *subpath)
1497 : {
1498 215092 : MaterialPath *pathnode = makeNode(MaterialPath);
1499 :
1500 : Assert(subpath->parent == rel);
1501 :
1502 215092 : pathnode->path.pathtype = T_Material;
1503 215092 : pathnode->path.parent = rel;
1504 215092 : pathnode->path.pathtarget = rel->reltarget;
1505 215092 : pathnode->path.param_info = subpath->param_info;
1506 215092 : pathnode->path.parallel_aware = false;
1507 405000 : pathnode->path.parallel_safe = rel->consider_parallel &&
1508 189908 : subpath->parallel_safe;
1509 215092 : pathnode->path.parallel_workers = subpath->parallel_workers;
1510 215092 : pathnode->path.pathkeys = subpath->pathkeys;
1511 :
1512 215092 : pathnode->subpath = subpath;
1513 :
1514 215092 : cost_material(&pathnode->path,
1515 : subpath->startup_cost,
1516 : subpath->total_cost,
1517 : subpath->rows,
1518 215092 : subpath->pathtarget->width);
1519 :
1520 215092 : return pathnode;
1521 : }
1522 :
1523 : /*
1524 : * create_unique_path
1525 : * Creates a path representing elimination of distinct rows from the
1526 : * input data. Distinct-ness is defined according to the needs of the
1527 : * semijoin represented by sjinfo. If it is not possible to identify
1528 : * how to make the data unique, NULL is returned.
1529 : *
1530 : * If used at all, this is likely to be called repeatedly on the same rel;
1531 : * and the input subpath should always be the same (the cheapest_total path
1532 : * for the rel). So we cache the result.
1533 : */
1534 : UniquePath *
1535 10862 : create_unique_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1536 : SpecialJoinInfo *sjinfo)
1537 : {
1538 : UniquePath *pathnode;
1539 : Path sort_path; /* dummy for result of cost_sort */
1540 : Path agg_path; /* dummy for result of cost_agg */
1541 : MemoryContext oldcontext;
1542 : int numCols;
1543 :
1544 : /* Caller made a mistake if subpath isn't cheapest_total ... */
1545 : Assert(subpath == rel->cheapest_total_path);
1546 : Assert(subpath->parent == rel);
1547 : /* ... or if SpecialJoinInfo is the wrong one */
1548 : Assert(sjinfo->jointype == JOIN_SEMI);
1549 : Assert(bms_equal(rel->relids, sjinfo->syn_righthand));
1550 :
1551 : /* If result already cached, return it */
1552 10862 : if (rel->cheapest_unique_path)
1553 9084 : return (UniquePath *) rel->cheapest_unique_path;
1554 :
1555 : /* If it's not possible to unique-ify, return NULL */
1556 1778 : if (!(sjinfo->semi_can_btree || sjinfo->semi_can_hash))
1557 62 : return NULL;
1558 :
1559 : /*
1560 : * When called during GEQO join planning, we are in a short-lived memory
1561 : * context. We must make sure that the path and any subsidiary data
1562 : * structures created for a baserel survive the GEQO cycle, else the
1563 : * baserel is trashed for future GEQO cycles. On the other hand, when we
1564 : * are creating those for a joinrel during GEQO, we don't want them to
1565 : * clutter the main planning context. Upshot is that the best solution is
1566 : * to explicitly allocate memory in the same context the given RelOptInfo
1567 : * is in.
1568 : */
1569 1716 : oldcontext = MemoryContextSwitchTo(GetMemoryChunkContext(rel));
1570 :
1571 1716 : pathnode = makeNode(UniquePath);
1572 :
1573 1716 : pathnode->path.pathtype = T_Unique;
1574 1716 : pathnode->path.parent = rel;
1575 1716 : pathnode->path.pathtarget = rel->reltarget;
1576 1716 : pathnode->path.param_info = subpath->param_info;
1577 1716 : pathnode->path.parallel_aware = false;
1578 3272 : pathnode->path.parallel_safe = rel->consider_parallel &&
1579 1556 : subpath->parallel_safe;
1580 1716 : pathnode->path.parallel_workers = subpath->parallel_workers;
1581 :
1582 : /*
1583 : * Assume the output is unsorted, since we don't necessarily have pathkeys
1584 : * to represent it. (This might get overridden below.)
1585 : */
1586 1716 : pathnode->path.pathkeys = NIL;
1587 :
1588 1716 : pathnode->subpath = subpath;
1589 1716 : pathnode->in_operators = sjinfo->semi_operators;
1590 1716 : pathnode->uniq_exprs = sjinfo->semi_rhs_exprs;
1591 :
1592 : /*
1593 : * If the input is a relation and it has a unique index that proves the
1594 : * semi_rhs_exprs are unique, then we don't need to do anything. Note
1595 : * that relation_has_unique_index_for automatically considers restriction
1596 : * clauses for the rel, as well.
1597 : */
1598 1988 : if (rel->rtekind == RTE_RELATION && sjinfo->semi_can_btree &&
1599 272 : relation_has_unique_index_for(root, rel, NIL,
1600 : sjinfo->semi_rhs_exprs,
1601 : sjinfo->semi_operators))
1602 : {
1603 0 : pathnode->umethod = UNIQUE_PATH_NOOP;
1604 0 : pathnode->path.rows = rel->rows;
1605 0 : pathnode->path.startup_cost = subpath->startup_cost;
1606 0 : pathnode->path.total_cost = subpath->total_cost;
1607 0 : pathnode->path.pathkeys = subpath->pathkeys;
1608 :
1609 0 : rel->cheapest_unique_path = (Path *) pathnode;
1610 :
1611 0 : MemoryContextSwitchTo(oldcontext);
1612 :
1613 0 : return pathnode;
1614 : }
1615 :
1616 : /*
1617 : * If the input is a subquery whose output must be unique already, then we
1618 : * don't need to do anything. The test for uniqueness has to consider
1619 : * exactly which columns we are extracting; for example "SELECT DISTINCT
1620 : * x,y" doesn't guarantee that x alone is distinct. So we cannot check for
1621 : * this optimization unless semi_rhs_exprs consists only of simple Vars
1622 : * referencing subquery outputs. (Possibly we could do something with
1623 : * expressions in the subquery outputs, too, but for now keep it simple.)
1624 : */
1625 1716 : if (rel->rtekind == RTE_SUBQUERY)
1626 : {
1627 264 : RangeTblEntry *rte = planner_rt_fetch(rel->relid, root);
1628 :
1629 264 : if (query_supports_distinctness(rte->subquery))
1630 : {
1631 : List *sub_tlist_colnos;
1632 :
1633 240 : sub_tlist_colnos = translate_sub_tlist(sjinfo->semi_rhs_exprs,
1634 240 : rel->relid);
1635 :
1636 280 : if (sub_tlist_colnos &&
1637 40 : query_is_distinct_for(rte->subquery,
1638 : sub_tlist_colnos,
1639 : sjinfo->semi_operators))
1640 : {
1641 0 : pathnode->umethod = UNIQUE_PATH_NOOP;
1642 0 : pathnode->path.rows = rel->rows;
1643 0 : pathnode->path.startup_cost = subpath->startup_cost;
1644 0 : pathnode->path.total_cost = subpath->total_cost;
1645 0 : pathnode->path.pathkeys = subpath->pathkeys;
1646 :
1647 0 : rel->cheapest_unique_path = (Path *) pathnode;
1648 :
1649 0 : MemoryContextSwitchTo(oldcontext);
1650 :
1651 0 : return pathnode;
1652 : }
1653 : }
1654 : }
1655 :
1656 : /* Estimate number of output rows */
1657 1716 : pathnode->path.rows = estimate_num_groups(root,
1658 : sjinfo->semi_rhs_exprs,
1659 : rel->rows,
1660 : NULL);
1661 1716 : numCols = list_length(sjinfo->semi_rhs_exprs);
1662 :
1663 1716 : if (sjinfo->semi_can_btree)
1664 : {
1665 : /*
1666 : * Estimate cost for sort+unique implementation
1667 : */
1668 3432 : cost_sort(&sort_path, root, NIL,
1669 : subpath->total_cost,
1670 : rel->rows,
1671 1716 : subpath->pathtarget->width,
1672 : 0.0,
1673 : work_mem,
1674 : -1.0);
1675 :
1676 : /*
1677 : * Charge one cpu_operator_cost per comparison per input tuple. We
1678 : * assume all columns get compared at most of the tuples. (XXX
1679 : * probably this is an overestimate.) This should agree with
1680 : * create_upper_unique_path.
1681 : */
1682 1716 : sort_path.total_cost += cpu_operator_cost * rel->rows * numCols;
1683 : }
1684 :
1685 1716 : if (sjinfo->semi_can_hash)
1686 : {
1687 : /*
1688 : * Estimate the overhead per hashtable entry at 64 bytes (same as in
1689 : * planner.c).
1690 : */
1691 1716 : int hashentrysize = subpath->pathtarget->width + 64;
1692 :
1693 1716 : if (hashentrysize * pathnode->path.rows > work_mem * 1024L)
1694 : {
1695 : /*
1696 : * We should not try to hash. Hack the SpecialJoinInfo to
1697 : * remember this, in case we come through here again.
1698 : */
1699 0 : sjinfo->semi_can_hash = false;
1700 : }
1701 : else
1702 1716 : cost_agg(&agg_path, root,
1703 : AGG_HASHED, NULL,
1704 : numCols, pathnode->path.rows,
1705 : NIL,
1706 : subpath->startup_cost,
1707 : subpath->total_cost,
1708 : rel->rows);
1709 : }
1710 :
1711 1716 : if (sjinfo->semi_can_btree && sjinfo->semi_can_hash)
1712 : {
1713 3432 : if (agg_path.total_cost < sort_path.total_cost)
1714 1666 : pathnode->umethod = UNIQUE_PATH_HASH;
1715 : else
1716 50 : pathnode->umethod = UNIQUE_PATH_SORT;
1717 : }
1718 0 : else if (sjinfo->semi_can_btree)
1719 0 : pathnode->umethod = UNIQUE_PATH_SORT;
1720 0 : else if (sjinfo->semi_can_hash)
1721 0 : pathnode->umethod = UNIQUE_PATH_HASH;
1722 : else
1723 : {
1724 : /* we can get here only if we abandoned hashing above */
1725 0 : MemoryContextSwitchTo(oldcontext);
1726 0 : return NULL;
1727 : }
1728 :
1729 1716 : if (pathnode->umethod == UNIQUE_PATH_HASH)
1730 : {
1731 1666 : pathnode->path.startup_cost = agg_path.startup_cost;
1732 1666 : pathnode->path.total_cost = agg_path.total_cost;
1733 : }
1734 : else
1735 : {
1736 50 : pathnode->path.startup_cost = sort_path.startup_cost;
1737 50 : pathnode->path.total_cost = sort_path.total_cost;
1738 : }
1739 :
1740 1716 : rel->cheapest_unique_path = (Path *) pathnode;
1741 :
1742 1716 : MemoryContextSwitchTo(oldcontext);
1743 :
1744 1716 : return pathnode;
1745 : }
1746 :
1747 : /*
1748 : * create_gather_merge_path
1749 : *
1750 : * Creates a path corresponding to a gather merge scan, returning
1751 : * the pathnode.
1752 : */
1753 : GatherMergePath *
1754 2868 : create_gather_merge_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1755 : PathTarget *target, List *pathkeys,
1756 : Relids required_outer, double *rows)
1757 : {
1758 2868 : GatherMergePath *pathnode = makeNode(GatherMergePath);
1759 2868 : Cost input_startup_cost = 0;
1760 2868 : Cost input_total_cost = 0;
1761 :
1762 : Assert(subpath->parallel_safe);
1763 : Assert(pathkeys);
1764 :
1765 2868 : pathnode->path.pathtype = T_GatherMerge;
1766 2868 : pathnode->path.parent = rel;
1767 2868 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1768 : required_outer);
1769 2868 : pathnode->path.parallel_aware = false;
1770 :
1771 2868 : pathnode->subpath = subpath;
1772 2868 : pathnode->num_workers = subpath->parallel_workers;
1773 2868 : pathnode->path.pathkeys = pathkeys;
1774 2868 : pathnode->path.pathtarget = target ? target : rel->reltarget;
1775 2868 : pathnode->path.rows += subpath->rows;
1776 :
1777 2868 : if (pathkeys_contained_in(pathkeys, subpath->pathkeys))
1778 : {
1779 : /* Subpath is adequately ordered, we won't need to sort it */
1780 2868 : input_startup_cost += subpath->startup_cost;
1781 2868 : input_total_cost += subpath->total_cost;
1782 : }
1783 : else
1784 : {
1785 : /* We'll need to insert a Sort node, so include cost for that */
1786 : Path sort_path; /* dummy for result of cost_sort */
1787 :
1788 0 : cost_sort(&sort_path,
1789 : root,
1790 : pathkeys,
1791 : subpath->total_cost,
1792 : subpath->rows,
1793 0 : subpath->pathtarget->width,
1794 : 0.0,
1795 : work_mem,
1796 : -1);
1797 0 : input_startup_cost += sort_path.startup_cost;
1798 0 : input_total_cost += sort_path.total_cost;
1799 : }
1800 :
1801 2868 : cost_gather_merge(pathnode, root, rel, pathnode->path.param_info,
1802 : input_startup_cost, input_total_cost, rows);
1803 :
1804 2868 : return pathnode;
1805 : }
1806 :
1807 : /*
1808 : * translate_sub_tlist - get subquery column numbers represented by tlist
1809 : *
1810 : * The given targetlist usually contains only Vars referencing the given relid.
1811 : * Extract their varattnos (ie, the column numbers of the subquery) and return
1812 : * as an integer List.
1813 : *
1814 : * If any of the tlist items is not a simple Var, we cannot determine whether
1815 : * the subquery's uniqueness condition (if any) matches ours, so punt and
1816 : * return NIL.
1817 : */
1818 : static List *
1819 240 : translate_sub_tlist(List *tlist, int relid)
1820 : {
1821 240 : List *result = NIL;
1822 : ListCell *l;
1823 :
1824 280 : foreach(l, tlist)
1825 : {
1826 240 : Var *var = (Var *) lfirst(l);
1827 :
1828 280 : if (!var || !IsA(var, Var) ||
1829 40 : var->varno != relid)
1830 200 : return NIL; /* punt */
1831 :
1832 40 : result = lappend_int(result, var->varattno);
1833 : }
1834 40 : return result;
1835 : }
1836 :
1837 : /*
1838 : * create_gather_path
1839 : * Creates a path corresponding to a gather scan, returning the
1840 : * pathnode.
1841 : *
1842 : * 'rows' may optionally be set to override row estimates from other sources.
1843 : */
1844 : GatherPath *
1845 7766 : create_gather_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1846 : PathTarget *target, Relids required_outer, double *rows)
1847 : {
1848 7766 : GatherPath *pathnode = makeNode(GatherPath);
1849 :
1850 : Assert(subpath->parallel_safe);
1851 :
1852 7766 : pathnode->path.pathtype = T_Gather;
1853 7766 : pathnode->path.parent = rel;
1854 7766 : pathnode->path.pathtarget = target;
1855 7766 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1856 : required_outer);
1857 7766 : pathnode->path.parallel_aware = false;
1858 7766 : pathnode->path.parallel_safe = false;
1859 7766 : pathnode->path.parallel_workers = 0;
1860 7766 : pathnode->path.pathkeys = NIL; /* Gather has unordered result */
1861 :
1862 7766 : pathnode->subpath = subpath;
1863 7766 : pathnode->num_workers = subpath->parallel_workers;
1864 7766 : pathnode->single_copy = false;
1865 :
1866 7766 : if (pathnode->num_workers == 0)
1867 : {
1868 0 : pathnode->path.pathkeys = subpath->pathkeys;
1869 0 : pathnode->num_workers = 1;
1870 0 : pathnode->single_copy = true;
1871 : }
1872 :
1873 7766 : cost_gather(pathnode, root, rel, pathnode->path.param_info, rows);
1874 :
1875 7766 : return pathnode;
1876 : }
1877 :
1878 : /*
1879 : * create_subqueryscan_path
1880 : * Creates a path corresponding to a scan of a subquery,
1881 : * returning the pathnode.
1882 : */
1883 : SubqueryScanPath *
1884 8448 : create_subqueryscan_path(PlannerInfo *root, RelOptInfo *rel, Path *subpath,
1885 : List *pathkeys, Relids required_outer)
1886 : {
1887 8448 : SubqueryScanPath *pathnode = makeNode(SubqueryScanPath);
1888 :
1889 8448 : pathnode->path.pathtype = T_SubqueryScan;
1890 8448 : pathnode->path.parent = rel;
1891 8448 : pathnode->path.pathtarget = rel->reltarget;
1892 8448 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
1893 : required_outer);
1894 8448 : pathnode->path.parallel_aware = false;
1895 13076 : pathnode->path.parallel_safe = rel->consider_parallel &&
1896 4628 : subpath->parallel_safe;
1897 8448 : pathnode->path.parallel_workers = subpath->parallel_workers;
1898 8448 : pathnode->path.pathkeys = pathkeys;
1899 8448 : pathnode->subpath = subpath;
1900 :
1901 8448 : cost_subqueryscan(pathnode, root, rel, pathnode->path.param_info);
1902 :
1903 8448 : return pathnode;
1904 : }
1905 :
1906 : /*
1907 : * create_functionscan_path
1908 : * Creates a path corresponding to a sequential scan of a function,
1909 : * returning the pathnode.
1910 : */
1911 : Path *
1912 28482 : create_functionscan_path(PlannerInfo *root, RelOptInfo *rel,
1913 : List *pathkeys, Relids required_outer)
1914 : {
1915 28482 : Path *pathnode = makeNode(Path);
1916 :
1917 28482 : pathnode->pathtype = T_FunctionScan;
1918 28482 : pathnode->parent = rel;
1919 28482 : pathnode->pathtarget = rel->reltarget;
1920 28482 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1921 : required_outer);
1922 28482 : pathnode->parallel_aware = false;
1923 28482 : pathnode->parallel_safe = rel->consider_parallel;
1924 28482 : pathnode->parallel_workers = 0;
1925 28482 : pathnode->pathkeys = pathkeys;
1926 :
1927 28482 : cost_functionscan(pathnode, root, rel, pathnode->param_info);
1928 :
1929 28482 : return pathnode;
1930 : }
1931 :
1932 : /*
1933 : * create_tablefuncscan_path
1934 : * Creates a path corresponding to a sequential scan of a table function,
1935 : * returning the pathnode.
1936 : */
1937 : Path *
1938 140 : create_tablefuncscan_path(PlannerInfo *root, RelOptInfo *rel,
1939 : Relids required_outer)
1940 : {
1941 140 : Path *pathnode = makeNode(Path);
1942 :
1943 140 : pathnode->pathtype = T_TableFuncScan;
1944 140 : pathnode->parent = rel;
1945 140 : pathnode->pathtarget = rel->reltarget;
1946 140 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1947 : required_outer);
1948 140 : pathnode->parallel_aware = false;
1949 140 : pathnode->parallel_safe = rel->consider_parallel;
1950 140 : pathnode->parallel_workers = 0;
1951 140 : pathnode->pathkeys = NIL; /* result is always unordered */
1952 :
1953 140 : cost_tablefuncscan(pathnode, root, rel, pathnode->param_info);
1954 :
1955 140 : return pathnode;
1956 : }
1957 :
1958 : /*
1959 : * create_valuesscan_path
1960 : * Creates a path corresponding to a scan of a VALUES list,
1961 : * returning the pathnode.
1962 : */
1963 : Path *
1964 3784 : create_valuesscan_path(PlannerInfo *root, RelOptInfo *rel,
1965 : Relids required_outer)
1966 : {
1967 3784 : Path *pathnode = makeNode(Path);
1968 :
1969 3784 : pathnode->pathtype = T_ValuesScan;
1970 3784 : pathnode->parent = rel;
1971 3784 : pathnode->pathtarget = rel->reltarget;
1972 3784 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1973 : required_outer);
1974 3784 : pathnode->parallel_aware = false;
1975 3784 : pathnode->parallel_safe = rel->consider_parallel;
1976 3784 : pathnode->parallel_workers = 0;
1977 3784 : pathnode->pathkeys = NIL; /* result is always unordered */
1978 :
1979 3784 : cost_valuesscan(pathnode, root, rel, pathnode->param_info);
1980 :
1981 3784 : return pathnode;
1982 : }
1983 :
1984 : /*
1985 : * create_ctescan_path
1986 : * Creates a path corresponding to a scan of a non-self-reference CTE,
1987 : * returning the pathnode.
1988 : */
1989 : Path *
1990 896 : create_ctescan_path(PlannerInfo *root, RelOptInfo *rel, Relids required_outer)
1991 : {
1992 896 : Path *pathnode = makeNode(Path);
1993 :
1994 896 : pathnode->pathtype = T_CteScan;
1995 896 : pathnode->parent = rel;
1996 896 : pathnode->pathtarget = rel->reltarget;
1997 896 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
1998 : required_outer);
1999 896 : pathnode->parallel_aware = false;
2000 896 : pathnode->parallel_safe = rel->consider_parallel;
2001 896 : pathnode->parallel_workers = 0;
2002 896 : pathnode->pathkeys = NIL; /* XXX for now, result is always unordered */
2003 :
2004 896 : cost_ctescan(pathnode, root, rel, pathnode->param_info);
2005 :
2006 896 : return pathnode;
2007 : }
2008 :
2009 : /*
2010 : * create_namedtuplestorescan_path
2011 : * Creates a path corresponding to a scan of a named tuplestore, returning
2012 : * the pathnode.
2013 : */
2014 : Path *
2015 256 : create_namedtuplestorescan_path(PlannerInfo *root, RelOptInfo *rel,
2016 : Relids required_outer)
2017 : {
2018 256 : Path *pathnode = makeNode(Path);
2019 :
2020 256 : pathnode->pathtype = T_NamedTuplestoreScan;
2021 256 : pathnode->parent = rel;
2022 256 : pathnode->pathtarget = rel->reltarget;
2023 256 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2024 : required_outer);
2025 256 : pathnode->parallel_aware = false;
2026 256 : pathnode->parallel_safe = rel->consider_parallel;
2027 256 : pathnode->parallel_workers = 0;
2028 256 : pathnode->pathkeys = NIL; /* result is always unordered */
2029 :
2030 256 : cost_namedtuplestorescan(pathnode, root, rel, pathnode->param_info);
2031 :
2032 256 : return pathnode;
2033 : }
2034 :
2035 : /*
2036 : * create_resultscan_path
2037 : * Creates a path corresponding to a scan of an RTE_RESULT relation,
2038 : * returning the pathnode.
2039 : */
2040 : Path *
2041 638 : create_resultscan_path(PlannerInfo *root, RelOptInfo *rel,
2042 : Relids required_outer)
2043 : {
2044 638 : Path *pathnode = makeNode(Path);
2045 :
2046 638 : pathnode->pathtype = T_Result;
2047 638 : pathnode->parent = rel;
2048 638 : pathnode->pathtarget = rel->reltarget;
2049 638 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2050 : required_outer);
2051 638 : pathnode->parallel_aware = false;
2052 638 : pathnode->parallel_safe = rel->consider_parallel;
2053 638 : pathnode->parallel_workers = 0;
2054 638 : pathnode->pathkeys = NIL; /* result is always unordered */
2055 :
2056 638 : cost_resultscan(pathnode, root, rel, pathnode->param_info);
2057 :
2058 638 : return pathnode;
2059 : }
2060 :
2061 : /*
2062 : * create_worktablescan_path
2063 : * Creates a path corresponding to a scan of a self-reference CTE,
2064 : * returning the pathnode.
2065 : */
2066 : Path *
2067 320 : create_worktablescan_path(PlannerInfo *root, RelOptInfo *rel,
2068 : Relids required_outer)
2069 : {
2070 320 : Path *pathnode = makeNode(Path);
2071 :
2072 320 : pathnode->pathtype = T_WorkTableScan;
2073 320 : pathnode->parent = rel;
2074 320 : pathnode->pathtarget = rel->reltarget;
2075 320 : pathnode->param_info = get_baserel_parampathinfo(root, rel,
2076 : required_outer);
2077 320 : pathnode->parallel_aware = false;
2078 320 : pathnode->parallel_safe = rel->consider_parallel;
2079 320 : pathnode->parallel_workers = 0;
2080 320 : pathnode->pathkeys = NIL; /* result is always unordered */
2081 :
2082 : /* Cost is the same as for a regular CTE scan */
2083 320 : cost_ctescan(pathnode, root, rel, pathnode->param_info);
2084 :
2085 320 : return pathnode;
2086 : }
2087 :
2088 : /*
2089 : * create_foreignscan_path
2090 : * Creates a path corresponding to a scan of a foreign base table,
2091 : * returning the pathnode.
2092 : *
2093 : * This function is never called from core Postgres; rather, it's expected
2094 : * to be called by the GetForeignPaths function of a foreign data wrapper.
2095 : * We make the FDW supply all fields of the path, since we do not have any way
2096 : * to calculate them in core. However, there is a usually-sane default for
2097 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2098 : */
2099 : ForeignPath *
2100 2718 : create_foreignscan_path(PlannerInfo *root, RelOptInfo *rel,
2101 : PathTarget *target,
2102 : double rows, Cost startup_cost, Cost total_cost,
2103 : List *pathkeys,
2104 : Relids required_outer,
2105 : Path *fdw_outerpath,
2106 : List *fdw_private)
2107 : {
2108 2718 : ForeignPath *pathnode = makeNode(ForeignPath);
2109 :
2110 : /* Historically some FDWs were confused about when to use this */
2111 : Assert(IS_SIMPLE_REL(rel));
2112 :
2113 2718 : pathnode->path.pathtype = T_ForeignScan;
2114 2718 : pathnode->path.parent = rel;
2115 2718 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2116 2718 : pathnode->path.param_info = get_baserel_parampathinfo(root, rel,
2117 : required_outer);
2118 2718 : pathnode->path.parallel_aware = false;
2119 2718 : pathnode->path.parallel_safe = rel->consider_parallel;
2120 2718 : pathnode->path.parallel_workers = 0;
2121 2718 : pathnode->path.rows = rows;
2122 2718 : pathnode->path.startup_cost = startup_cost;
2123 2718 : pathnode->path.total_cost = total_cost;
2124 2718 : pathnode->path.pathkeys = pathkeys;
2125 :
2126 2718 : pathnode->fdw_outerpath = fdw_outerpath;
2127 2718 : pathnode->fdw_private = fdw_private;
2128 :
2129 2718 : return pathnode;
2130 : }
2131 :
2132 : /*
2133 : * create_foreign_join_path
2134 : * Creates a path corresponding to a scan of a foreign join,
2135 : * returning the pathnode.
2136 : *
2137 : * This function is never called from core Postgres; rather, it's expected
2138 : * to be called by the GetForeignJoinPaths function of a foreign data wrapper.
2139 : * We make the FDW supply all fields of the path, since we do not have any way
2140 : * to calculate them in core. However, there is a usually-sane default for
2141 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2142 : */
2143 : ForeignPath *
2144 702 : create_foreign_join_path(PlannerInfo *root, RelOptInfo *rel,
2145 : PathTarget *target,
2146 : double rows, Cost startup_cost, Cost total_cost,
2147 : List *pathkeys,
2148 : Relids required_outer,
2149 : Path *fdw_outerpath,
2150 : List *fdw_private)
2151 : {
2152 702 : ForeignPath *pathnode = makeNode(ForeignPath);
2153 :
2154 : /*
2155 : * We should use get_joinrel_parampathinfo to handle parameterized paths,
2156 : * but the API of this function doesn't support it, and existing
2157 : * extensions aren't yet trying to build such paths anyway. For the
2158 : * moment just throw an error if someone tries it; eventually we should
2159 : * revisit this.
2160 : */
2161 702 : if (!bms_is_empty(required_outer) || !bms_is_empty(rel->lateral_relids))
2162 0 : elog(ERROR, "parameterized foreign joins are not supported yet");
2163 :
2164 702 : pathnode->path.pathtype = T_ForeignScan;
2165 702 : pathnode->path.parent = rel;
2166 702 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2167 702 : pathnode->path.param_info = NULL; /* XXX see above */
2168 702 : pathnode->path.parallel_aware = false;
2169 702 : pathnode->path.parallel_safe = rel->consider_parallel;
2170 702 : pathnode->path.parallel_workers = 0;
2171 702 : pathnode->path.rows = rows;
2172 702 : pathnode->path.startup_cost = startup_cost;
2173 702 : pathnode->path.total_cost = total_cost;
2174 702 : pathnode->path.pathkeys = pathkeys;
2175 :
2176 702 : pathnode->fdw_outerpath = fdw_outerpath;
2177 702 : pathnode->fdw_private = fdw_private;
2178 :
2179 702 : return pathnode;
2180 : }
2181 :
2182 : /*
2183 : * create_foreign_upper_path
2184 : * Creates a path corresponding to an upper relation that's computed
2185 : * directly by an FDW, returning the pathnode.
2186 : *
2187 : * This function is never called from core Postgres; rather, it's expected to
2188 : * be called by the GetForeignUpperPaths function of a foreign data wrapper.
2189 : * We make the FDW supply all fields of the path, since we do not have any way
2190 : * to calculate them in core. However, there is a usually-sane default for
2191 : * the pathtarget (rel->reltarget), so we let a NULL for "target" select that.
2192 : */
2193 : ForeignPath *
2194 448 : create_foreign_upper_path(PlannerInfo *root, RelOptInfo *rel,
2195 : PathTarget *target,
2196 : double rows, Cost startup_cost, Cost total_cost,
2197 : List *pathkeys,
2198 : Path *fdw_outerpath,
2199 : List *fdw_private)
2200 : {
2201 448 : ForeignPath *pathnode = makeNode(ForeignPath);
2202 :
2203 : /*
2204 : * Upper relations should never have any lateral references, since joining
2205 : * is complete.
2206 : */
2207 : Assert(bms_is_empty(rel->lateral_relids));
2208 :
2209 448 : pathnode->path.pathtype = T_ForeignScan;
2210 448 : pathnode->path.parent = rel;
2211 448 : pathnode->path.pathtarget = target ? target : rel->reltarget;
2212 448 : pathnode->path.param_info = NULL;
2213 448 : pathnode->path.parallel_aware = false;
2214 448 : pathnode->path.parallel_safe = rel->consider_parallel;
2215 448 : pathnode->path.parallel_workers = 0;
2216 448 : pathnode->path.rows = rows;
2217 448 : pathnode->path.startup_cost = startup_cost;
2218 448 : pathnode->path.total_cost = total_cost;
2219 448 : pathnode->path.pathkeys = pathkeys;
2220 :
2221 448 : pathnode->fdw_outerpath = fdw_outerpath;
2222 448 : pathnode->fdw_private = fdw_private;
2223 :
2224 448 : return pathnode;
2225 : }
2226 :
2227 : /*
2228 : * calc_nestloop_required_outer
2229 : * Compute the required_outer set for a nestloop join path
2230 : *
2231 : * Note: result must not share storage with either input
2232 : */
2233 : Relids
2234 1092050 : calc_nestloop_required_outer(Relids outerrelids,
2235 : Relids outer_paramrels,
2236 : Relids innerrelids,
2237 : Relids inner_paramrels)
2238 : {
2239 : Relids required_outer;
2240 :
2241 : /* inner_path can require rels from outer path, but not vice versa */
2242 : Assert(!bms_overlap(outer_paramrels, innerrelids));
2243 : /* easy case if inner path is not parameterized */
2244 1092050 : if (!inner_paramrels)
2245 842016 : return bms_copy(outer_paramrels);
2246 : /* else, form the union ... */
2247 250034 : required_outer = bms_union(outer_paramrels, inner_paramrels);
2248 : /* ... and remove any mention of now-satisfied outer rels */
2249 250034 : required_outer = bms_del_members(required_outer,
2250 : outerrelids);
2251 : /* maintain invariant that required_outer is exactly NULL if empty */
2252 250034 : if (bms_is_empty(required_outer))
2253 : {
2254 205330 : bms_free(required_outer);
2255 205330 : required_outer = NULL;
2256 : }
2257 250034 : return required_outer;
2258 : }
2259 :
2260 : /*
2261 : * calc_non_nestloop_required_outer
2262 : * Compute the required_outer set for a merge or hash join path
2263 : *
2264 : * Note: result must not share storage with either input
2265 : */
2266 : Relids
2267 779884 : calc_non_nestloop_required_outer(Path *outer_path, Path *inner_path)
2268 : {
2269 779884 : Relids outer_paramrels = PATH_REQ_OUTER(outer_path);
2270 779884 : Relids inner_paramrels = PATH_REQ_OUTER(inner_path);
2271 : Relids required_outer;
2272 :
2273 : /* neither path can require rels from the other */
2274 : Assert(!bms_overlap(outer_paramrels, inner_path->parent->relids));
2275 : Assert(!bms_overlap(inner_paramrels, outer_path->parent->relids));
2276 : /* form the union ... */
2277 779884 : required_outer = bms_union(outer_paramrels, inner_paramrels);
2278 : /* we do not need an explicit test for empty; bms_union gets it right */
2279 779884 : return required_outer;
2280 : }
2281 :
2282 : /*
2283 : * create_nestloop_path
2284 : * Creates a pathnode corresponding to a nestloop join between two
2285 : * relations.
2286 : *
2287 : * 'joinrel' is the join relation.
2288 : * 'jointype' is the type of join required
2289 : * 'workspace' is the result from initial_cost_nestloop
2290 : * 'extra' contains various information about the join
2291 : * 'outer_path' is the outer path
2292 : * 'inner_path' is the inner path
2293 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2294 : * 'pathkeys' are the path keys of the new join path
2295 : * 'required_outer' is the set of required outer rels
2296 : *
2297 : * Returns the resulting path node.
2298 : */
2299 : NestPath *
2300 564640 : create_nestloop_path(PlannerInfo *root,
2301 : RelOptInfo *joinrel,
2302 : JoinType jointype,
2303 : JoinCostWorkspace *workspace,
2304 : JoinPathExtraData *extra,
2305 : Path *outer_path,
2306 : Path *inner_path,
2307 : List *restrict_clauses,
2308 : List *pathkeys,
2309 : Relids required_outer)
2310 : {
2311 564640 : NestPath *pathnode = makeNode(NestPath);
2312 564640 : Relids inner_req_outer = PATH_REQ_OUTER(inner_path);
2313 :
2314 : /*
2315 : * If the inner path is parameterized by the outer, we must drop any
2316 : * restrict_clauses that are due to be moved into the inner path. We have
2317 : * to do this now, rather than postpone the work till createplan time,
2318 : * because the restrict_clauses list can affect the size and cost
2319 : * estimates for this path.
2320 : */
2321 564640 : if (bms_overlap(inner_req_outer, outer_path->parent->relids))
2322 : {
2323 104696 : Relids inner_and_outer = bms_union(inner_path->parent->relids,
2324 : inner_req_outer);
2325 104696 : List *jclauses = NIL;
2326 : ListCell *lc;
2327 :
2328 218538 : foreach(lc, restrict_clauses)
2329 : {
2330 113842 : RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
2331 :
2332 113842 : if (!join_clause_is_movable_into(rinfo,
2333 113842 : inner_path->parent->relids,
2334 : inner_and_outer))
2335 7404 : jclauses = lappend(jclauses, rinfo);
2336 : }
2337 104696 : restrict_clauses = jclauses;
2338 : }
2339 :
2340 564640 : pathnode->path.pathtype = T_NestLoop;
2341 564640 : pathnode->path.parent = joinrel;
2342 564640 : pathnode->path.pathtarget = joinrel->reltarget;
2343 564640 : pathnode->path.param_info =
2344 564640 : get_joinrel_parampathinfo(root,
2345 : joinrel,
2346 : outer_path,
2347 : inner_path,
2348 : extra->sjinfo,
2349 : required_outer,
2350 : &restrict_clauses);
2351 564640 : pathnode->path.parallel_aware = false;
2352 1634740 : pathnode->path.parallel_safe = joinrel->consider_parallel &&
2353 1069492 : outer_path->parallel_safe && inner_path->parallel_safe;
2354 : /* This is a foolish way to estimate parallel_workers, but for now... */
2355 564640 : pathnode->path.parallel_workers = outer_path->parallel_workers;
2356 564640 : pathnode->path.pathkeys = pathkeys;
2357 564640 : pathnode->jointype = jointype;
2358 564640 : pathnode->inner_unique = extra->inner_unique;
2359 564640 : pathnode->outerjoinpath = outer_path;
2360 564640 : pathnode->innerjoinpath = inner_path;
2361 564640 : pathnode->joinrestrictinfo = restrict_clauses;
2362 :
2363 564640 : final_cost_nestloop(root, pathnode, workspace, extra);
2364 :
2365 564640 : return pathnode;
2366 : }
2367 :
2368 : /*
2369 : * create_mergejoin_path
2370 : * Creates a pathnode corresponding to a mergejoin join between
2371 : * two relations
2372 : *
2373 : * 'joinrel' is the join relation
2374 : * 'jointype' is the type of join required
2375 : * 'workspace' is the result from initial_cost_mergejoin
2376 : * 'extra' contains various information about the join
2377 : * 'outer_path' is the outer path
2378 : * 'inner_path' is the inner path
2379 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2380 : * 'pathkeys' are the path keys of the new join path
2381 : * 'required_outer' is the set of required outer rels
2382 : * 'mergeclauses' are the RestrictInfo nodes to use as merge clauses
2383 : * (this should be a subset of the restrict_clauses list)
2384 : * 'outersortkeys' are the sort varkeys for the outer relation
2385 : * 'innersortkeys' are the sort varkeys for the inner relation
2386 : */
2387 : MergePath *
2388 119548 : create_mergejoin_path(PlannerInfo *root,
2389 : RelOptInfo *joinrel,
2390 : JoinType jointype,
2391 : JoinCostWorkspace *workspace,
2392 : JoinPathExtraData *extra,
2393 : Path *outer_path,
2394 : Path *inner_path,
2395 : List *restrict_clauses,
2396 : List *pathkeys,
2397 : Relids required_outer,
2398 : List *mergeclauses,
2399 : List *outersortkeys,
2400 : List *innersortkeys)
2401 : {
2402 119548 : MergePath *pathnode = makeNode(MergePath);
2403 :
2404 119548 : pathnode->jpath.path.pathtype = T_MergeJoin;
2405 119548 : pathnode->jpath.path.parent = joinrel;
2406 119548 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2407 119548 : pathnode->jpath.path.param_info =
2408 119548 : get_joinrel_parampathinfo(root,
2409 : joinrel,
2410 : outer_path,
2411 : inner_path,
2412 : extra->sjinfo,
2413 : required_outer,
2414 : &restrict_clauses);
2415 119548 : pathnode->jpath.path.parallel_aware = false;
2416 338316 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2417 218652 : outer_path->parallel_safe && inner_path->parallel_safe;
2418 : /* This is a foolish way to estimate parallel_workers, but for now... */
2419 119548 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2420 119548 : pathnode->jpath.path.pathkeys = pathkeys;
2421 119548 : pathnode->jpath.jointype = jointype;
2422 119548 : pathnode->jpath.inner_unique = extra->inner_unique;
2423 119548 : pathnode->jpath.outerjoinpath = outer_path;
2424 119548 : pathnode->jpath.innerjoinpath = inner_path;
2425 119548 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2426 119548 : pathnode->path_mergeclauses = mergeclauses;
2427 119548 : pathnode->outersortkeys = outersortkeys;
2428 119548 : pathnode->innersortkeys = innersortkeys;
2429 : /* pathnode->skip_mark_restore will be set by final_cost_mergejoin */
2430 : /* pathnode->materialize_inner will be set by final_cost_mergejoin */
2431 :
2432 119548 : final_cost_mergejoin(root, pathnode, workspace, extra);
2433 :
2434 119548 : return pathnode;
2435 : }
2436 :
2437 : /*
2438 : * create_hashjoin_path
2439 : * Creates a pathnode corresponding to a hash join between two relations.
2440 : *
2441 : * 'joinrel' is the join relation
2442 : * 'jointype' is the type of join required
2443 : * 'workspace' is the result from initial_cost_hashjoin
2444 : * 'extra' contains various information about the join
2445 : * 'outer_path' is the cheapest outer path
2446 : * 'inner_path' is the cheapest inner path
2447 : * 'parallel_hash' to select Parallel Hash of inner path (shared hash table)
2448 : * 'restrict_clauses' are the RestrictInfo nodes to apply at the join
2449 : * 'required_outer' is the set of required outer rels
2450 : * 'hashclauses' are the RestrictInfo nodes to use as hash clauses
2451 : * (this should be a subset of the restrict_clauses list)
2452 : */
2453 : HashPath *
2454 135058 : create_hashjoin_path(PlannerInfo *root,
2455 : RelOptInfo *joinrel,
2456 : JoinType jointype,
2457 : JoinCostWorkspace *workspace,
2458 : JoinPathExtraData *extra,
2459 : Path *outer_path,
2460 : Path *inner_path,
2461 : bool parallel_hash,
2462 : List *restrict_clauses,
2463 : Relids required_outer,
2464 : List *hashclauses)
2465 : {
2466 135058 : HashPath *pathnode = makeNode(HashPath);
2467 :
2468 135058 : pathnode->jpath.path.pathtype = T_HashJoin;
2469 135058 : pathnode->jpath.path.parent = joinrel;
2470 135058 : pathnode->jpath.path.pathtarget = joinrel->reltarget;
2471 135058 : pathnode->jpath.path.param_info =
2472 135058 : get_joinrel_parampathinfo(root,
2473 : joinrel,
2474 : outer_path,
2475 : inner_path,
2476 : extra->sjinfo,
2477 : required_outer,
2478 : &restrict_clauses);
2479 135058 : pathnode->jpath.path.parallel_aware =
2480 135058 : joinrel->consider_parallel && parallel_hash;
2481 381218 : pathnode->jpath.path.parallel_safe = joinrel->consider_parallel &&
2482 245856 : outer_path->parallel_safe && inner_path->parallel_safe;
2483 : /* This is a foolish way to estimate parallel_workers, but for now... */
2484 135058 : pathnode->jpath.path.parallel_workers = outer_path->parallel_workers;
2485 :
2486 : /*
2487 : * A hashjoin never has pathkeys, since its output ordering is
2488 : * unpredictable due to possible batching. XXX If the inner relation is
2489 : * small enough, we could instruct the executor that it must not batch,
2490 : * and then we could assume that the output inherits the outer relation's
2491 : * ordering, which might save a sort step. However there is considerable
2492 : * downside if our estimate of the inner relation size is badly off. For
2493 : * the moment we don't risk it. (Note also that if we wanted to take this
2494 : * seriously, joinpath.c would have to consider many more paths for the
2495 : * outer rel than it does now.)
2496 : */
2497 135058 : pathnode->jpath.path.pathkeys = NIL;
2498 135058 : pathnode->jpath.jointype = jointype;
2499 135058 : pathnode->jpath.inner_unique = extra->inner_unique;
2500 135058 : pathnode->jpath.outerjoinpath = outer_path;
2501 135058 : pathnode->jpath.innerjoinpath = inner_path;
2502 135058 : pathnode->jpath.joinrestrictinfo = restrict_clauses;
2503 135058 : pathnode->path_hashclauses = hashclauses;
2504 : /* final_cost_hashjoin will fill in pathnode->num_batches */
2505 :
2506 135058 : final_cost_hashjoin(root, pathnode, workspace, extra);
2507 :
2508 135058 : return pathnode;
2509 : }
2510 :
2511 : /*
2512 : * create_projection_path
2513 : * Creates a pathnode that represents performing a projection.
2514 : *
2515 : * 'rel' is the parent relation associated with the result
2516 : * 'subpath' is the path representing the source of data
2517 : * 'target' is the PathTarget to be computed
2518 : */
2519 : ProjectionPath *
2520 226366 : create_projection_path(PlannerInfo *root,
2521 : RelOptInfo *rel,
2522 : Path *subpath,
2523 : PathTarget *target)
2524 : {
2525 226366 : ProjectionPath *pathnode = makeNode(ProjectionPath);
2526 226366 : PathTarget *oldtarget = subpath->pathtarget;
2527 :
2528 226366 : pathnode->path.pathtype = T_Result;
2529 226366 : pathnode->path.parent = rel;
2530 226366 : pathnode->path.pathtarget = target;
2531 : /* For now, assume we are above any joins, so no parameterization */
2532 226366 : pathnode->path.param_info = NULL;
2533 226366 : pathnode->path.parallel_aware = false;
2534 485458 : pathnode->path.parallel_safe = rel->consider_parallel &&
2535 258456 : subpath->parallel_safe &&
2536 32090 : is_parallel_safe(root, (Node *) target->exprs);
2537 226366 : pathnode->path.parallel_workers = subpath->parallel_workers;
2538 : /* Projection does not change the sort order */
2539 226366 : pathnode->path.pathkeys = subpath->pathkeys;
2540 :
2541 226366 : pathnode->subpath = subpath;
2542 :
2543 : /*
2544 : * We might not need a separate Result node. If the input plan node type
2545 : * can project, we can just tell it to project something else. Or, if it
2546 : * can't project but the desired target has the same expression list as
2547 : * what the input will produce anyway, we can still give it the desired
2548 : * tlist (possibly changing its ressortgroupref labels, but nothing else).
2549 : * Note: in the latter case, create_projection_plan has to recheck our
2550 : * conclusion; see comments therein.
2551 : */
2552 234606 : if (is_projection_capable_path(subpath) ||
2553 8240 : equal(oldtarget->exprs, target->exprs))
2554 : {
2555 : /* No separate Result node needed */
2556 225248 : pathnode->dummypp = true;
2557 :
2558 : /*
2559 : * Set cost of plan as subpath's cost, adjusted for tlist replacement.
2560 : */
2561 225248 : pathnode->path.rows = subpath->rows;
2562 450496 : pathnode->path.startup_cost = subpath->startup_cost +
2563 225248 : (target->cost.startup - oldtarget->cost.startup);
2564 675744 : pathnode->path.total_cost = subpath->total_cost +
2565 450496 : (target->cost.startup - oldtarget->cost.startup) +
2566 225248 : (target->cost.per_tuple - oldtarget->cost.per_tuple) * subpath->rows;
2567 : }
2568 : else
2569 : {
2570 : /* We really do need the Result node */
2571 1118 : pathnode->dummypp = false;
2572 :
2573 : /*
2574 : * The Result node's cost is cpu_tuple_cost per row, plus the cost of
2575 : * evaluating the tlist. There is no qual to worry about.
2576 : */
2577 1118 : pathnode->path.rows = subpath->rows;
2578 2236 : pathnode->path.startup_cost = subpath->startup_cost +
2579 1118 : target->cost.startup;
2580 3354 : pathnode->path.total_cost = subpath->total_cost +
2581 2236 : target->cost.startup +
2582 1118 : (cpu_tuple_cost + target->cost.per_tuple) * subpath->rows;
2583 : }
2584 :
2585 226366 : return pathnode;
2586 : }
2587 :
2588 : /*
2589 : * apply_projection_to_path
2590 : * Add a projection step, or just apply the target directly to given path.
2591 : *
2592 : * This has the same net effect as create_projection_path(), except that if
2593 : * a separate Result plan node isn't needed, we just replace the given path's
2594 : * pathtarget with the desired one. This must be used only when the caller
2595 : * knows that the given path isn't referenced elsewhere and so can be modified
2596 : * in-place.
2597 : *
2598 : * If the input path is a GatherPath or GatherMergePath, we try to push the
2599 : * new target down to its input as well; this is a yet more invasive
2600 : * modification of the input path, which create_projection_path() can't do.
2601 : *
2602 : * Note that we mustn't change the source path's parent link; so when it is
2603 : * add_path'd to "rel" things will be a bit inconsistent. So far that has
2604 : * not caused any trouble.
2605 : *
2606 : * 'rel' is the parent relation associated with the result
2607 : * 'path' is the path representing the source of data
2608 : * 'target' is the PathTarget to be computed
2609 : */
2610 : Path *
2611 14146 : apply_projection_to_path(PlannerInfo *root,
2612 : RelOptInfo *rel,
2613 : Path *path,
2614 : PathTarget *target)
2615 : {
2616 : QualCost oldcost;
2617 :
2618 : /*
2619 : * If given path can't project, we might need a Result node, so make a
2620 : * separate ProjectionPath.
2621 : */
2622 14146 : if (!is_projection_capable_path(path))
2623 6656 : return (Path *) create_projection_path(root, rel, path, target);
2624 :
2625 : /*
2626 : * We can just jam the desired tlist into the existing path, being sure to
2627 : * update its cost estimates appropriately.
2628 : */
2629 7490 : oldcost = path->pathtarget->cost;
2630 7490 : path->pathtarget = target;
2631 :
2632 7490 : path->startup_cost += target->cost.startup - oldcost.startup;
2633 14980 : path->total_cost += target->cost.startup - oldcost.startup +
2634 7490 : (target->cost.per_tuple - oldcost.per_tuple) * path->rows;
2635 :
2636 : /*
2637 : * If the path happens to be a Gather or GatherMerge path, we'd like to
2638 : * arrange for the subpath to return the required target list so that
2639 : * workers can help project. But if there is something that is not
2640 : * parallel-safe in the target expressions, then we can't.
2641 : */
2642 8692 : if ((IsA(path, GatherPath) ||IsA(path, GatherMergePath)) &&
2643 1202 : is_parallel_safe(root, (Node *) target->exprs))
2644 : {
2645 : /*
2646 : * We always use create_projection_path here, even if the subpath is
2647 : * projection-capable, so as to avoid modifying the subpath in place.
2648 : * It seems unlikely at present that there could be any other
2649 : * references to the subpath, but better safe than sorry.
2650 : *
2651 : * Note that we don't change the parallel path's cost estimates; it
2652 : * might be appropriate to do so, to reflect the fact that the bulk of
2653 : * the target evaluation will happen in workers.
2654 : */
2655 2404 : if (IsA(path, GatherPath))
2656 : {
2657 0 : GatherPath *gpath = (GatherPath *) path;
2658 :
2659 0 : gpath->subpath = (Path *)
2660 0 : create_projection_path(root,
2661 0 : gpath->subpath->parent,
2662 : gpath->subpath,
2663 : target);
2664 : }
2665 : else
2666 : {
2667 1202 : GatherMergePath *gmpath = (GatherMergePath *) path;
2668 :
2669 1202 : gmpath->subpath = (Path *)
2670 2404 : create_projection_path(root,
2671 1202 : gmpath->subpath->parent,
2672 : gmpath->subpath,
2673 : target);
2674 : }
2675 : }
2676 8904 : else if (path->parallel_safe &&
2677 2616 : !is_parallel_safe(root, (Node *) target->exprs))
2678 : {
2679 : /*
2680 : * We're inserting a parallel-restricted target list into a path
2681 : * currently marked parallel-safe, so we have to mark it as no longer
2682 : * safe.
2683 : */
2684 8 : path->parallel_safe = false;
2685 : }
2686 :
2687 7490 : return path;
2688 : }
2689 :
2690 : /*
2691 : * create_set_projection_path
2692 : * Creates a pathnode that represents performing a projection that
2693 : * includes set-returning functions.
2694 : *
2695 : * 'rel' is the parent relation associated with the result
2696 : * 'subpath' is the path representing the source of data
2697 : * 'target' is the PathTarget to be computed
2698 : */
2699 : ProjectSetPath *
2700 3376 : create_set_projection_path(PlannerInfo *root,
2701 : RelOptInfo *rel,
2702 : Path *subpath,
2703 : PathTarget *target)
2704 : {
2705 3376 : ProjectSetPath *pathnode = makeNode(ProjectSetPath);
2706 : double tlist_rows;
2707 : ListCell *lc;
2708 :
2709 3376 : pathnode->path.pathtype = T_ProjectSet;
2710 3376 : pathnode->path.parent = rel;
2711 3376 : pathnode->path.pathtarget = target;
2712 : /* For now, assume we are above any joins, so no parameterization */
2713 3376 : pathnode->path.param_info = NULL;
2714 3376 : pathnode->path.parallel_aware = false;
2715 7164 : pathnode->path.parallel_safe = rel->consider_parallel &&
2716 3768 : subpath->parallel_safe &&
2717 392 : is_parallel_safe(root, (Node *) target->exprs);
2718 3376 : pathnode->path.parallel_workers = subpath->parallel_workers;
2719 : /* Projection does not change the sort order XXX? */
2720 3376 : pathnode->path.pathkeys = subpath->pathkeys;
2721 :
2722 3376 : pathnode->subpath = subpath;
2723 :
2724 : /*
2725 : * Estimate number of rows produced by SRFs for each row of input; if
2726 : * there's more than one in this node, use the maximum.
2727 : */
2728 3376 : tlist_rows = 1;
2729 7902 : foreach(lc, target->exprs)
2730 : {
2731 4526 : Node *node = (Node *) lfirst(lc);
2732 : double itemrows;
2733 :
2734 4526 : itemrows = expression_returns_set_rows(root, node);
2735 4526 : if (tlist_rows < itemrows)
2736 3246 : tlist_rows = itemrows;
2737 : }
2738 :
2739 : /*
2740 : * In addition to the cost of evaluating the tlist, charge cpu_tuple_cost
2741 : * per input row, and half of cpu_tuple_cost for each added output row.
2742 : * This is slightly bizarre maybe, but it's what 9.6 did; we may revisit
2743 : * this estimate later.
2744 : */
2745 3376 : pathnode->path.rows = subpath->rows * tlist_rows;
2746 6752 : pathnode->path.startup_cost = subpath->startup_cost +
2747 3376 : target->cost.startup;
2748 10128 : pathnode->path.total_cost = subpath->total_cost +
2749 6752 : target->cost.startup +
2750 6752 : (cpu_tuple_cost + target->cost.per_tuple) * subpath->rows +
2751 3376 : (pathnode->path.rows - subpath->rows) * cpu_tuple_cost / 2;
2752 :
2753 3376 : return pathnode;
2754 : }
2755 :
2756 : /*
2757 : * create_sort_path
2758 : * Creates a pathnode that represents performing an explicit sort.
2759 : *
2760 : * 'rel' is the parent relation associated with the result
2761 : * 'subpath' is the path representing the source of data
2762 : * 'pathkeys' represents the desired sort order
2763 : * 'limit_tuples' is the estimated bound on the number of output tuples,
2764 : * or -1 if no LIMIT or couldn't estimate
2765 : */
2766 : SortPath *
2767 35778 : create_sort_path(PlannerInfo *root,
2768 : RelOptInfo *rel,
2769 : Path *subpath,
2770 : List *pathkeys,
2771 : double limit_tuples)
2772 : {
2773 35778 : SortPath *pathnode = makeNode(SortPath);
2774 :
2775 35778 : pathnode->path.pathtype = T_Sort;
2776 35778 : pathnode->path.parent = rel;
2777 : /* Sort doesn't project, so use source path's pathtarget */
2778 35778 : pathnode->path.pathtarget = subpath->pathtarget;
2779 : /* For now, assume we are above any joins, so no parameterization */
2780 35778 : pathnode->path.param_info = NULL;
2781 35778 : pathnode->path.parallel_aware = false;
2782 54256 : pathnode->path.parallel_safe = rel->consider_parallel &&
2783 18478 : subpath->parallel_safe;
2784 35778 : pathnode->path.parallel_workers = subpath->parallel_workers;
2785 35778 : pathnode->path.pathkeys = pathkeys;
2786 :
2787 35778 : pathnode->subpath = subpath;
2788 :
2789 71556 : cost_sort(&pathnode->path, root, pathkeys,
2790 : subpath->total_cost,
2791 : subpath->rows,
2792 35778 : subpath->pathtarget->width,
2793 : 0.0, /* XXX comparison_cost shouldn't be 0? */
2794 : work_mem, limit_tuples);
2795 :
2796 35778 : return pathnode;
2797 : }
2798 :
2799 : /*
2800 : * create_group_path
2801 : * Creates a pathnode that represents performing grouping of presorted input
2802 : *
2803 : * 'rel' is the parent relation associated with the result
2804 : * 'subpath' is the path representing the source of data
2805 : * 'target' is the PathTarget to be computed
2806 : * 'groupClause' is a list of SortGroupClause's representing the grouping
2807 : * 'qual' is the HAVING quals if any
2808 : * 'numGroups' is the estimated number of groups
2809 : */
2810 : GroupPath *
2811 452 : create_group_path(PlannerInfo *root,
2812 : RelOptInfo *rel,
2813 : Path *subpath,
2814 : List *groupClause,
2815 : List *qual,
2816 : double numGroups)
2817 : {
2818 452 : GroupPath *pathnode = makeNode(GroupPath);
2819 452 : PathTarget *target = rel->reltarget;
2820 :
2821 452 : pathnode->path.pathtype = T_Group;
2822 452 : pathnode->path.parent = rel;
2823 452 : pathnode->path.pathtarget = target;
2824 : /* For now, assume we are above any joins, so no parameterization */
2825 452 : pathnode->path.param_info = NULL;
2826 452 : pathnode->path.parallel_aware = false;
2827 674 : pathnode->path.parallel_safe = rel->consider_parallel &&
2828 222 : subpath->parallel_safe;
2829 452 : pathnode->path.parallel_workers = subpath->parallel_workers;
2830 : /* Group doesn't change sort ordering */
2831 452 : pathnode->path.pathkeys = subpath->pathkeys;
2832 :
2833 452 : pathnode->subpath = subpath;
2834 :
2835 452 : pathnode->groupClause = groupClause;
2836 452 : pathnode->qual = qual;
2837 :
2838 452 : cost_group(&pathnode->path, root,
2839 : list_length(groupClause),
2840 : numGroups,
2841 : qual,
2842 : subpath->startup_cost, subpath->total_cost,
2843 : subpath->rows);
2844 :
2845 : /* add tlist eval cost for each output row */
2846 452 : pathnode->path.startup_cost += target->cost.startup;
2847 904 : pathnode->path.total_cost += target->cost.startup +
2848 452 : target->cost.per_tuple * pathnode->path.rows;
2849 :
2850 452 : return pathnode;
2851 : }
2852 :
2853 : /*
2854 : * create_upper_unique_path
2855 : * Creates a pathnode that represents performing an explicit Unique step
2856 : * on presorted input.
2857 : *
2858 : * This produces a Unique plan node, but the use-case is so different from
2859 : * create_unique_path that it doesn't seem worth trying to merge the two.
2860 : *
2861 : * 'rel' is the parent relation associated with the result
2862 : * 'subpath' is the path representing the source of data
2863 : * 'numCols' is the number of grouping columns
2864 : * 'numGroups' is the estimated number of groups
2865 : *
2866 : * The input path must be sorted on the grouping columns, plus possibly
2867 : * additional columns; so the first numCols pathkeys are the grouping columns
2868 : */
2869 : UpperUniquePath *
2870 526 : create_upper_unique_path(PlannerInfo *root,
2871 : RelOptInfo *rel,
2872 : Path *subpath,
2873 : int numCols,
2874 : double numGroups)
2875 : {
2876 526 : UpperUniquePath *pathnode = makeNode(UpperUniquePath);
2877 :
2878 526 : pathnode->path.pathtype = T_Unique;
2879 526 : pathnode->path.parent = rel;
2880 : /* Unique doesn't project, so use source path's pathtarget */
2881 526 : pathnode->path.pathtarget = subpath->pathtarget;
2882 : /* For now, assume we are above any joins, so no parameterization */
2883 526 : pathnode->path.param_info = NULL;
2884 526 : pathnode->path.parallel_aware = false;
2885 878 : pathnode->path.parallel_safe = rel->consider_parallel &&
2886 352 : subpath->parallel_safe;
2887 526 : pathnode->path.parallel_workers = subpath->parallel_workers;
2888 : /* Unique doesn't change the input ordering */
2889 526 : pathnode->path.pathkeys = subpath->pathkeys;
2890 :
2891 526 : pathnode->subpath = subpath;
2892 526 : pathnode->numkeys = numCols;
2893 :
2894 : /*
2895 : * Charge one cpu_operator_cost per comparison per input tuple. We assume
2896 : * all columns get compared at most of the tuples. (XXX probably this is
2897 : * an overestimate.)
2898 : */
2899 526 : pathnode->path.startup_cost = subpath->startup_cost;
2900 1052 : pathnode->path.total_cost = subpath->total_cost +
2901 526 : cpu_operator_cost * subpath->rows * numCols;
2902 526 : pathnode->path.rows = numGroups;
2903 :
2904 526 : return pathnode;
2905 : }
2906 :
2907 : /*
2908 : * create_agg_path
2909 : * Creates a pathnode that represents performing aggregation/grouping
2910 : *
2911 : * 'rel' is the parent relation associated with the result
2912 : * 'subpath' is the path representing the source of data
2913 : * 'target' is the PathTarget to be computed
2914 : * 'aggstrategy' is the Agg node's basic implementation strategy
2915 : * 'aggsplit' is the Agg node's aggregate-splitting mode
2916 : * 'groupClause' is a list of SortGroupClause's representing the grouping
2917 : * 'qual' is the HAVING quals if any
2918 : * 'aggcosts' contains cost info about the aggregate functions to be computed
2919 : * 'numGroups' is the estimated number of groups (1 if not grouping)
2920 : */
2921 : AggPath *
2922 31598 : create_agg_path(PlannerInfo *root,
2923 : RelOptInfo *rel,
2924 : Path *subpath,
2925 : PathTarget *target,
2926 : AggStrategy aggstrategy,
2927 : AggSplit aggsplit,
2928 : List *groupClause,
2929 : List *qual,
2930 : const AggClauseCosts *aggcosts,
2931 : double numGroups)
2932 : {
2933 31598 : AggPath *pathnode = makeNode(AggPath);
2934 :
2935 31598 : pathnode->path.pathtype = T_Agg;
2936 31598 : pathnode->path.parent = rel;
2937 31598 : pathnode->path.pathtarget = target;
2938 : /* For now, assume we are above any joins, so no parameterization */
2939 31598 : pathnode->path.param_info = NULL;
2940 31598 : pathnode->path.parallel_aware = false;
2941 47604 : pathnode->path.parallel_safe = rel->consider_parallel &&
2942 16006 : subpath->parallel_safe;
2943 31598 : pathnode->path.parallel_workers = subpath->parallel_workers;
2944 31598 : if (aggstrategy == AGG_SORTED)
2945 4054 : pathnode->path.pathkeys = subpath->pathkeys; /* preserves order */
2946 : else
2947 27544 : pathnode->path.pathkeys = NIL; /* output is unordered */
2948 31598 : pathnode->subpath = subpath;
2949 :
2950 31598 : pathnode->aggstrategy = aggstrategy;
2951 31598 : pathnode->aggsplit = aggsplit;
2952 31598 : pathnode->numGroups = numGroups;
2953 31598 : pathnode->groupClause = groupClause;
2954 31598 : pathnode->qual = qual;
2955 :
2956 31598 : cost_agg(&pathnode->path, root,
2957 : aggstrategy, aggcosts,
2958 : list_length(groupClause), numGroups,
2959 : qual,
2960 : subpath->startup_cost, subpath->total_cost,
2961 : subpath->rows);
2962 :
2963 : /* add tlist eval cost for each output row */
2964 31598 : pathnode->path.startup_cost += target->cost.startup;
2965 63196 : pathnode->path.total_cost += target->cost.startup +
2966 31598 : target->cost.per_tuple * pathnode->path.rows;
2967 :
2968 31598 : return pathnode;
2969 : }
2970 :
2971 : /*
2972 : * create_groupingsets_path
2973 : * Creates a pathnode that represents performing GROUPING SETS aggregation
2974 : *
2975 : * GroupingSetsPath represents sorted grouping with one or more grouping sets.
2976 : * The input path's result must be sorted to match the last entry in
2977 : * rollup_groupclauses.
2978 : *
2979 : * 'rel' is the parent relation associated with the result
2980 : * 'subpath' is the path representing the source of data
2981 : * 'target' is the PathTarget to be computed
2982 : * 'having_qual' is the HAVING quals if any
2983 : * 'rollups' is a list of RollupData nodes
2984 : * 'agg_costs' contains cost info about the aggregate functions to be computed
2985 : * 'numGroups' is the estimated total number of groups
2986 : */
2987 : GroupingSetsPath *
2988 1020 : create_groupingsets_path(PlannerInfo *root,
2989 : RelOptInfo *rel,
2990 : Path *subpath,
2991 : List *having_qual,
2992 : AggStrategy aggstrategy,
2993 : List *rollups,
2994 : const AggClauseCosts *agg_costs,
2995 : double numGroups)
2996 : {
2997 1020 : GroupingSetsPath *pathnode = makeNode(GroupingSetsPath);
2998 1020 : PathTarget *target = rel->reltarget;
2999 : ListCell *lc;
3000 1020 : bool is_first = true;
3001 1020 : bool is_first_sort = true;
3002 :
3003 : /* The topmost generated Plan node will be an Agg */
3004 1020 : pathnode->path.pathtype = T_Agg;
3005 1020 : pathnode->path.parent = rel;
3006 1020 : pathnode->path.pathtarget = target;
3007 1020 : pathnode->path.param_info = subpath->param_info;
3008 1020 : pathnode->path.parallel_aware = false;
3009 1440 : pathnode->path.parallel_safe = rel->consider_parallel &&
3010 420 : subpath->parallel_safe;
3011 1020 : pathnode->path.parallel_workers = subpath->parallel_workers;
3012 1020 : pathnode->subpath = subpath;
3013 :
3014 : /*
3015 : * Simplify callers by downgrading AGG_SORTED to AGG_PLAIN, and AGG_MIXED
3016 : * to AGG_HASHED, here if possible.
3017 : */
3018 1464 : if (aggstrategy == AGG_SORTED &&
3019 676 : list_length(rollups) == 1 &&
3020 232 : ((RollupData *) linitial(rollups))->groupClause == NIL)
3021 28 : aggstrategy = AGG_PLAIN;
3022 :
3023 1472 : if (aggstrategy == AGG_MIXED &&
3024 452 : list_length(rollups) == 1)
3025 0 : aggstrategy = AGG_HASHED;
3026 :
3027 : /*
3028 : * Output will be in sorted order by group_pathkeys if, and only if, there
3029 : * is a single rollup operation on a non-empty list of grouping
3030 : * expressions.
3031 : */
3032 1020 : if (aggstrategy == AGG_SORTED && list_length(rollups) == 1)
3033 204 : pathnode->path.pathkeys = root->group_pathkeys;
3034 : else
3035 816 : pathnode->path.pathkeys = NIL;
3036 :
3037 1020 : pathnode->aggstrategy = aggstrategy;
3038 1020 : pathnode->rollups = rollups;
3039 1020 : pathnode->qual = having_qual;
3040 :
3041 : Assert(rollups != NIL);
3042 : Assert(aggstrategy != AGG_PLAIN || list_length(rollups) == 1);
3043 : Assert(aggstrategy != AGG_MIXED || list_length(rollups) > 1);
3044 :
3045 3456 : foreach(lc, rollups)
3046 : {
3047 2436 : RollupData *rollup = lfirst(lc);
3048 2436 : List *gsets = rollup->gsets;
3049 2436 : int numGroupCols = list_length(linitial(gsets));
3050 :
3051 : /*
3052 : * In AGG_SORTED or AGG_PLAIN mode, the first rollup takes the
3053 : * (already-sorted) input, and following ones do their own sort.
3054 : *
3055 : * In AGG_HASHED mode, there is one rollup for each grouping set.
3056 : *
3057 : * In AGG_MIXED mode, the first rollups are hashed, the first
3058 : * non-hashed one takes the (already-sorted) input, and following ones
3059 : * do their own sort.
3060 : */
3061 2436 : if (is_first)
3062 : {
3063 1020 : cost_agg(&pathnode->path, root,
3064 : aggstrategy,
3065 : agg_costs,
3066 : numGroupCols,
3067 : rollup->numGroups,
3068 : having_qual,
3069 : subpath->startup_cost,
3070 : subpath->total_cost,
3071 : subpath->rows);
3072 1020 : is_first = false;
3073 1020 : if (!rollup->is_hashed)
3074 444 : is_first_sort = false;
3075 : }
3076 : else
3077 : {
3078 : Path sort_path; /* dummy for result of cost_sort */
3079 : Path agg_path; /* dummy for result of cost_agg */
3080 :
3081 1416 : if (rollup->is_hashed || is_first_sort)
3082 : {
3083 : /*
3084 : * Account for cost of aggregation, but don't charge input
3085 : * cost again
3086 : */
3087 2104 : cost_agg(&agg_path, root,
3088 1052 : rollup->is_hashed ? AGG_HASHED : AGG_SORTED,
3089 : agg_costs,
3090 : numGroupCols,
3091 : rollup->numGroups,
3092 : having_qual,
3093 : 0.0, 0.0,
3094 : subpath->rows);
3095 2104 : if (!rollup->is_hashed)
3096 452 : is_first_sort = false;
3097 : }
3098 : else
3099 : {
3100 : /* Account for cost of sort, but don't charge input cost again */
3101 728 : cost_sort(&sort_path, root, NIL,
3102 : 0.0,
3103 : subpath->rows,
3104 364 : subpath->pathtarget->width,
3105 : 0.0,
3106 : work_mem,
3107 : -1.0);
3108 :
3109 : /* Account for cost of aggregation */
3110 :
3111 364 : cost_agg(&agg_path, root,
3112 : AGG_SORTED,
3113 : agg_costs,
3114 : numGroupCols,
3115 : rollup->numGroups,
3116 : having_qual,
3117 : sort_path.startup_cost,
3118 : sort_path.total_cost,
3119 : sort_path.rows);
3120 : }
3121 :
3122 1416 : pathnode->path.total_cost += agg_path.total_cost;
3123 1416 : pathnode->path.rows += agg_path.rows;
3124 : }
3125 : }
3126 :
3127 : /* add tlist eval cost for each output row */
3128 1020 : pathnode->path.startup_cost += target->cost.startup;
3129 2040 : pathnode->path.total_cost += target->cost.startup +
3130 1020 : target->cost.per_tuple * pathnode->path.rows;
3131 :
3132 1020 : return pathnode;
3133 : }
3134 :
3135 : /*
3136 : * create_minmaxagg_path
3137 : * Creates a pathnode that represents computation of MIN/MAX aggregates
3138 : *
3139 : * 'rel' is the parent relation associated with the result
3140 : * 'target' is the PathTarget to be computed
3141 : * 'mmaggregates' is a list of MinMaxAggInfo structs
3142 : * 'quals' is the HAVING quals if any
3143 : */
3144 : MinMaxAggPath *
3145 398 : create_minmaxagg_path(PlannerInfo *root,
3146 : RelOptInfo *rel,
3147 : PathTarget *target,
3148 : List *mmaggregates,
3149 : List *quals)
3150 : {
3151 398 : MinMaxAggPath *pathnode = makeNode(MinMaxAggPath);
3152 : Cost initplan_cost;
3153 : ListCell *lc;
3154 :
3155 : /* The topmost generated Plan node will be a Result */
3156 398 : pathnode->path.pathtype = T_Result;
3157 398 : pathnode->path.parent = rel;
3158 398 : pathnode->path.pathtarget = target;
3159 : /* For now, assume we are above any joins, so no parameterization */
3160 398 : pathnode->path.param_info = NULL;
3161 398 : pathnode->path.parallel_aware = false;
3162 : /* A MinMaxAggPath implies use of subplans, so cannot be parallel-safe */
3163 398 : pathnode->path.parallel_safe = false;
3164 398 : pathnode->path.parallel_workers = 0;
3165 : /* Result is one unordered row */
3166 398 : pathnode->path.rows = 1;
3167 398 : pathnode->path.pathkeys = NIL;
3168 :
3169 398 : pathnode->mmaggregates = mmaggregates;
3170 398 : pathnode->quals = quals;
3171 :
3172 : /* Calculate cost of all the initplans ... */
3173 398 : initplan_cost = 0;
3174 820 : foreach(lc, mmaggregates)
3175 : {
3176 422 : MinMaxAggInfo *mminfo = (MinMaxAggInfo *) lfirst(lc);
3177 :
3178 422 : initplan_cost += mminfo->pathcost;
3179 : }
3180 :
3181 : /* add tlist eval cost for each output row, plus cpu_tuple_cost */
3182 398 : pathnode->path.startup_cost = initplan_cost + target->cost.startup;
3183 1194 : pathnode->path.total_cost = initplan_cost + target->cost.startup +
3184 796 : target->cost.per_tuple + cpu_tuple_cost;
3185 :
3186 : /*
3187 : * Add cost of qual, if any --- but we ignore its selectivity, since our
3188 : * rowcount estimate should be 1 no matter what the qual is.
3189 : */
3190 398 : if (quals)
3191 : {
3192 : QualCost qual_cost;
3193 :
3194 0 : cost_qual_eval(&qual_cost, quals, root);
3195 0 : pathnode->path.startup_cost += qual_cost.startup;
3196 0 : pathnode->path.total_cost += qual_cost.startup + qual_cost.per_tuple;
3197 : }
3198 :
3199 398 : return pathnode;
3200 : }
3201 :
3202 : /*
3203 : * create_windowagg_path
3204 : * Creates a pathnode that represents computation of window functions
3205 : *
3206 : * 'rel' is the parent relation associated with the result
3207 : * 'subpath' is the path representing the source of data
3208 : * 'target' is the PathTarget to be computed
3209 : * 'windowFuncs' is a list of WindowFunc structs
3210 : * 'winclause' is a WindowClause that is common to all the WindowFuncs
3211 : *
3212 : * The input must be sorted according to the WindowClause's PARTITION keys
3213 : * plus ORDER BY keys.
3214 : */
3215 : WindowAggPath *
3216 1192 : create_windowagg_path(PlannerInfo *root,
3217 : RelOptInfo *rel,
3218 : Path *subpath,
3219 : PathTarget *target,
3220 : List *windowFuncs,
3221 : WindowClause *winclause)
3222 : {
3223 1192 : WindowAggPath *pathnode = makeNode(WindowAggPath);
3224 :
3225 1192 : pathnode->path.pathtype = T_WindowAgg;
3226 1192 : pathnode->path.parent = rel;
3227 1192 : pathnode->path.pathtarget = target;
3228 : /* For now, assume we are above any joins, so no parameterization */
3229 1192 : pathnode->path.param_info = NULL;
3230 1192 : pathnode->path.parallel_aware = false;
3231 1192 : pathnode->path.parallel_safe = rel->consider_parallel &&
3232 0 : subpath->parallel_safe;
3233 1192 : pathnode->path.parallel_workers = subpath->parallel_workers;
3234 : /* WindowAgg preserves the input sort order */
3235 1192 : pathnode->path.pathkeys = subpath->pathkeys;
3236 :
3237 1192 : pathnode->subpath = subpath;
3238 1192 : pathnode->winclause = winclause;
3239 :
3240 : /*
3241 : * For costing purposes, assume that there are no redundant partitioning
3242 : * or ordering columns; it's not worth the trouble to deal with that
3243 : * corner case here. So we just pass the unmodified list lengths to
3244 : * cost_windowagg.
3245 : */
3246 3576 : cost_windowagg(&pathnode->path, root,
3247 : windowFuncs,
3248 1192 : list_length(winclause->partitionClause),
3249 1192 : list_length(winclause->orderClause),
3250 : subpath->startup_cost,
3251 : subpath->total_cost,
3252 : subpath->rows);
3253 :
3254 : /* add tlist eval cost for each output row */
3255 1192 : pathnode->path.startup_cost += target->cost.startup;
3256 2384 : pathnode->path.total_cost += target->cost.startup +
3257 1192 : target->cost.per_tuple * pathnode->path.rows;
3258 :
3259 1192 : return pathnode;
3260 : }
3261 :
3262 : /*
3263 : * create_setop_path
3264 : * Creates a pathnode that represents computation of INTERSECT or EXCEPT
3265 : *
3266 : * 'rel' is the parent relation associated with the result
3267 : * 'subpath' is the path representing the source of data
3268 : * 'cmd' is the specific semantics (INTERSECT or EXCEPT, with/without ALL)
3269 : * 'strategy' is the implementation strategy (sorted or hashed)
3270 : * 'distinctList' is a list of SortGroupClause's representing the grouping
3271 : * 'flagColIdx' is the column number where the flag column will be, if any
3272 : * 'firstFlag' is the flag value for the first input relation when hashing;
3273 : * or -1 when sorting
3274 : * 'numGroups' is the estimated number of distinct groups
3275 : * 'outputRows' is the estimated number of output rows
3276 : */
3277 : SetOpPath *
3278 208 : create_setop_path(PlannerInfo *root,
3279 : RelOptInfo *rel,
3280 : Path *subpath,
3281 : SetOpCmd cmd,
3282 : SetOpStrategy strategy,
3283 : List *distinctList,
3284 : AttrNumber flagColIdx,
3285 : int firstFlag,
3286 : double numGroups,
3287 : double outputRows)
3288 : {
3289 208 : SetOpPath *pathnode = makeNode(SetOpPath);
3290 :
3291 208 : pathnode->path.pathtype = T_SetOp;
3292 208 : pathnode->path.parent = rel;
3293 : /* SetOp doesn't project, so use source path's pathtarget */
3294 208 : pathnode->path.pathtarget = subpath->pathtarget;
3295 : /* For now, assume we are above any joins, so no parameterization */
3296 208 : pathnode->path.param_info = NULL;
3297 208 : pathnode->path.parallel_aware = false;
3298 208 : pathnode->path.parallel_safe = rel->consider_parallel &&
3299 0 : subpath->parallel_safe;
3300 208 : pathnode->path.parallel_workers = subpath->parallel_workers;
3301 : /* SetOp preserves the input sort order if in sort mode */
3302 208 : pathnode->path.pathkeys =
3303 208 : (strategy == SETOP_SORTED) ? subpath->pathkeys : NIL;
3304 :
3305 208 : pathnode->subpath = subpath;
3306 208 : pathnode->cmd = cmd;
3307 208 : pathnode->strategy = strategy;
3308 208 : pathnode->distinctList = distinctList;
3309 208 : pathnode->flagColIdx = flagColIdx;
3310 208 : pathnode->firstFlag = firstFlag;
3311 208 : pathnode->numGroups = numGroups;
3312 :
3313 : /*
3314 : * Charge one cpu_operator_cost per comparison per input tuple. We assume
3315 : * all columns get compared at most of the tuples.
3316 : */
3317 208 : pathnode->path.startup_cost = subpath->startup_cost;
3318 416 : pathnode->path.total_cost = subpath->total_cost +
3319 208 : cpu_operator_cost * subpath->rows * list_length(distinctList);
3320 208 : pathnode->path.rows = outputRows;
3321 :
3322 208 : return pathnode;
3323 : }
3324 :
3325 : /*
3326 : * create_recursiveunion_path
3327 : * Creates a pathnode that represents a recursive UNION node
3328 : *
3329 : * 'rel' is the parent relation associated with the result
3330 : * 'leftpath' is the source of data for the non-recursive term
3331 : * 'rightpath' is the source of data for the recursive term
3332 : * 'target' is the PathTarget to be computed
3333 : * 'distinctList' is a list of SortGroupClause's representing the grouping
3334 : * 'wtParam' is the ID of Param representing work table
3335 : * 'numGroups' is the estimated number of groups
3336 : *
3337 : * For recursive UNION ALL, distinctList is empty and numGroups is zero
3338 : */
3339 : RecursiveUnionPath *
3340 320 : create_recursiveunion_path(PlannerInfo *root,
3341 : RelOptInfo *rel,
3342 : Path *leftpath,
3343 : Path *rightpath,
3344 : PathTarget *target,
3345 : List *distinctList,
3346 : int wtParam,
3347 : double numGroups)
3348 : {
3349 320 : RecursiveUnionPath *pathnode = makeNode(RecursiveUnionPath);
3350 :
3351 320 : pathnode->path.pathtype = T_RecursiveUnion;
3352 320 : pathnode->path.parent = rel;
3353 320 : pathnode->path.pathtarget = target;
3354 : /* For now, assume we are above any joins, so no parameterization */
3355 320 : pathnode->path.param_info = NULL;
3356 320 : pathnode->path.parallel_aware = false;
3357 640 : pathnode->path.parallel_safe = rel->consider_parallel &&
3358 320 : leftpath->parallel_safe && rightpath->parallel_safe;
3359 : /* Foolish, but we'll do it like joins for now: */
3360 320 : pathnode->path.parallel_workers = leftpath->parallel_workers;
3361 : /* RecursiveUnion result is always unsorted */
3362 320 : pathnode->path.pathkeys = NIL;
3363 :
3364 320 : pathnode->leftpath = leftpath;
3365 320 : pathnode->rightpath = rightpath;
3366 320 : pathnode->distinctList = distinctList;
3367 320 : pathnode->wtParam = wtParam;
3368 320 : pathnode->numGroups = numGroups;
3369 :
3370 320 : cost_recursive_union(&pathnode->path, leftpath, rightpath);
3371 :
3372 320 : return pathnode;
3373 : }
3374 :
3375 : /*
3376 : * create_lockrows_path
3377 : * Creates a pathnode that represents acquiring row locks
3378 : *
3379 : * 'rel' is the parent relation associated with the result
3380 : * 'subpath' is the path representing the source of data
3381 : * 'rowMarks' is a list of PlanRowMark's
3382 : * 'epqParam' is the ID of Param for EvalPlanQual re-eval
3383 : */
3384 : LockRowsPath *
3385 5280 : create_lockrows_path(PlannerInfo *root, RelOptInfo *rel,
3386 : Path *subpath, List *rowMarks, int epqParam)
3387 : {
3388 5280 : LockRowsPath *pathnode = makeNode(LockRowsPath);
3389 :
3390 5280 : pathnode->path.pathtype = T_LockRows;
3391 5280 : pathnode->path.parent = rel;
3392 : /* LockRows doesn't project, so use source path's pathtarget */
3393 5280 : pathnode->path.pathtarget = subpath->pathtarget;
3394 : /* For now, assume we are above any joins, so no parameterization */
3395 5280 : pathnode->path.param_info = NULL;
3396 5280 : pathnode->path.parallel_aware = false;
3397 5280 : pathnode->path.parallel_safe = false;
3398 5280 : pathnode->path.parallel_workers = 0;
3399 5280 : pathnode->path.rows = subpath->rows;
3400 :
3401 : /*
3402 : * The result cannot be assumed sorted, since locking might cause the sort
3403 : * key columns to be replaced with new values.
3404 : */
3405 5280 : pathnode->path.pathkeys = NIL;
3406 :
3407 5280 : pathnode->subpath = subpath;
3408 5280 : pathnode->rowMarks = rowMarks;
3409 5280 : pathnode->epqParam = epqParam;
3410 :
3411 : /*
3412 : * We should charge something extra for the costs of row locking and
3413 : * possible refetches, but it's hard to say how much. For now, use
3414 : * cpu_tuple_cost per row.
3415 : */
3416 5280 : pathnode->path.startup_cost = subpath->startup_cost;
3417 10560 : pathnode->path.total_cost = subpath->total_cost +
3418 5280 : cpu_tuple_cost * subpath->rows;
3419 :
3420 5280 : return pathnode;
3421 : }
3422 :
3423 : /*
3424 : * create_modifytable_path
3425 : * Creates a pathnode that represents performing INSERT/UPDATE/DELETE mods
3426 : *
3427 : * 'rel' is the parent relation associated with the result
3428 : * 'operation' is the operation type
3429 : * 'canSetTag' is true if we set the command tag/es_processed
3430 : * 'nominalRelation' is the parent RT index for use of EXPLAIN
3431 : * 'rootRelation' is the partitioned table root RT index, or 0 if none
3432 : * 'partColsUpdated' is true if any partitioning columns are being updated,
3433 : * either from the target relation or a descendent partitioned table.
3434 : * 'resultRelations' is an integer list of actual RT indexes of target rel(s)
3435 : * 'subpaths' is a list of Path(s) producing source data (one per rel)
3436 : * 'subroots' is a list of PlannerInfo structs (one per rel)
3437 : * 'withCheckOptionLists' is a list of WCO lists (one per rel)
3438 : * 'returningLists' is a list of RETURNING tlists (one per rel)
3439 : * 'rowMarks' is a list of PlanRowMarks (non-locking only)
3440 : * 'onconflict' is the ON CONFLICT clause, or NULL
3441 : * 'epqParam' is the ID of Param for EvalPlanQual re-eval
3442 : */
3443 : ModifyTablePath *
3444 73348 : create_modifytable_path(PlannerInfo *root, RelOptInfo *rel,
3445 : CmdType operation, bool canSetTag,
3446 : Index nominalRelation, Index rootRelation,
3447 : bool partColsUpdated,
3448 : List *resultRelations, List *subpaths,
3449 : List *subroots,
3450 : List *withCheckOptionLists, List *returningLists,
3451 : List *rowMarks, OnConflictExpr *onconflict,
3452 : int epqParam)
3453 : {
3454 73348 : ModifyTablePath *pathnode = makeNode(ModifyTablePath);
3455 : double total_size;
3456 : ListCell *lc;
3457 :
3458 : Assert(list_length(resultRelations) == list_length(subpaths));
3459 : Assert(list_length(resultRelations) == list_length(subroots));
3460 : Assert(withCheckOptionLists == NIL ||
3461 : list_length(resultRelations) == list_length(withCheckOptionLists));
3462 : Assert(returningLists == NIL ||
3463 : list_length(resultRelations) == list_length(returningLists));
3464 :
3465 73348 : pathnode->path.pathtype = T_ModifyTable;
3466 73348 : pathnode->path.parent = rel;
3467 : /* pathtarget is not interesting, just make it minimally valid */
3468 73348 : pathnode->path.pathtarget = rel->reltarget;
3469 : /* For now, assume we are above any joins, so no parameterization */
3470 73348 : pathnode->path.param_info = NULL;
3471 73348 : pathnode->path.parallel_aware = false;
3472 73348 : pathnode->path.parallel_safe = false;
3473 73348 : pathnode->path.parallel_workers = 0;
3474 73348 : pathnode->path.pathkeys = NIL;
3475 :
3476 : /*
3477 : * Compute cost & rowcount as sum of subpath costs & rowcounts.
3478 : *
3479 : * Currently, we don't charge anything extra for the actual table
3480 : * modification work, nor for the WITH CHECK OPTIONS or RETURNING
3481 : * expressions if any. It would only be window dressing, since
3482 : * ModifyTable is always a top-level node and there is no way for the
3483 : * costs to change any higher-level planning choices. But we might want
3484 : * to make it look better sometime.
3485 : */
3486 73348 : pathnode->path.startup_cost = 0;
3487 73348 : pathnode->path.total_cost = 0;
3488 73348 : pathnode->path.rows = 0;
3489 73348 : total_size = 0;
3490 147862 : foreach(lc, subpaths)
3491 : {
3492 74514 : Path *subpath = (Path *) lfirst(lc);
3493 :
3494 74514 : if (lc == list_head(subpaths)) /* first node? */
3495 73348 : pathnode->path.startup_cost = subpath->startup_cost;
3496 74514 : pathnode->path.total_cost += subpath->total_cost;
3497 74514 : pathnode->path.rows += subpath->rows;
3498 74514 : total_size += subpath->pathtarget->width * subpath->rows;
3499 : }
3500 :
3501 : /*
3502 : * Set width to the average width of the subpath outputs. XXX this is
3503 : * totally wrong: we should report zero if no RETURNING, else an average
3504 : * of the RETURNING tlist widths. But it's what happened historically,
3505 : * and improving it is a task for another day.
3506 : */
3507 73348 : if (pathnode->path.rows > 0)
3508 73280 : total_size /= pathnode->path.rows;
3509 73348 : pathnode->path.pathtarget->width = rint(total_size);
3510 :
3511 73348 : pathnode->operation = operation;
3512 73348 : pathnode->canSetTag = canSetTag;
3513 73348 : pathnode->nominalRelation = nominalRelation;
3514 73348 : pathnode->rootRelation = rootRelation;
3515 73348 : pathnode->partColsUpdated = partColsUpdated;
3516 73348 : pathnode->resultRelations = resultRelations;
3517 73348 : pathnode->subpaths = subpaths;
3518 73348 : pathnode->subroots = subroots;
3519 73348 : pathnode->withCheckOptionLists = withCheckOptionLists;
3520 73348 : pathnode->returningLists = returningLists;
3521 73348 : pathnode->rowMarks = rowMarks;
3522 73348 : pathnode->onconflict = onconflict;
3523 73348 : pathnode->epqParam = epqParam;
3524 :
3525 73348 : return pathnode;
3526 : }
3527 :
3528 : /*
3529 : * create_limit_path
3530 : * Creates a pathnode that represents performing LIMIT/OFFSET
3531 : *
3532 : * In addition to providing the actual OFFSET and LIMIT expressions,
3533 : * the caller must provide estimates of their values for costing purposes.
3534 : * The estimates are as computed by preprocess_limit(), ie, 0 represents
3535 : * the clause not being present, and -1 means it's present but we could
3536 : * not estimate its value.
3537 : *
3538 : * 'rel' is the parent relation associated with the result
3539 : * 'subpath' is the path representing the source of data
3540 : * 'limitOffset' is the actual OFFSET expression, or NULL
3541 : * 'limitCount' is the actual LIMIT expression, or NULL
3542 : * 'offset_est' is the estimated value of the OFFSET expression
3543 : * 'count_est' is the estimated value of the LIMIT expression
3544 : */
3545 : LimitPath *
3546 3382 : create_limit_path(PlannerInfo *root, RelOptInfo *rel,
3547 : Path *subpath,
3548 : Node *limitOffset, Node *limitCount,
3549 : int64 offset_est, int64 count_est)
3550 : {
3551 3382 : LimitPath *pathnode = makeNode(LimitPath);
3552 :
3553 3382 : pathnode->path.pathtype = T_Limit;
3554 3382 : pathnode->path.parent = rel;
3555 : /* Limit doesn't project, so use source path's pathtarget */
3556 3382 : pathnode->path.pathtarget = subpath->pathtarget;
3557 : /* For now, assume we are above any joins, so no parameterization */
3558 3382 : pathnode->path.param_info = NULL;
3559 3382 : pathnode->path.parallel_aware = false;
3560 4482 : pathnode->path.parallel_safe = rel->consider_parallel &&
3561 1100 : subpath->parallel_safe;
3562 3382 : pathnode->path.parallel_workers = subpath->parallel_workers;
3563 3382 : pathnode->path.rows = subpath->rows;
3564 3382 : pathnode->path.startup_cost = subpath->startup_cost;
3565 3382 : pathnode->path.total_cost = subpath->total_cost;
3566 3382 : pathnode->path.pathkeys = subpath->pathkeys;
3567 3382 : pathnode->subpath = subpath;
3568 3382 : pathnode->limitOffset = limitOffset;
3569 3382 : pathnode->limitCount = limitCount;
3570 :
3571 : /*
3572 : * Adjust the output rows count and costs according to the offset/limit.
3573 : */
3574 3382 : adjust_limit_rows_costs(&pathnode->path.rows,
3575 : &pathnode->path.startup_cost,
3576 : &pathnode->path.total_cost,
3577 : offset_est, count_est);
3578 :
3579 3382 : return pathnode;
3580 : }
3581 :
3582 : /*
3583 : * adjust_limit_rows_costs
3584 : * Adjust the size and cost estimates for a LimitPath node according to the
3585 : * offset/limit.
3586 : *
3587 : * This is only a cosmetic issue if we are at top level, but if we are
3588 : * building a subquery then it's important to report correct info to the outer
3589 : * planner.
3590 : *
3591 : * When the offset or count couldn't be estimated, use 10% of the estimated
3592 : * number of rows emitted from the subpath.
3593 : *
3594 : * XXX we don't bother to add eval costs of the offset/limit expressions
3595 : * themselves to the path costs. In theory we should, but in most cases those
3596 : * expressions are trivial and it's just not worth the trouble.
3597 : */
3598 : void
3599 3508 : adjust_limit_rows_costs(double *rows, /* in/out parameter */
3600 : Cost *startup_cost, /* in/out parameter */
3601 : Cost *total_cost, /* in/out parameter */
3602 : int64 offset_est,
3603 : int64 count_est)
3604 : {
3605 3508 : double input_rows = *rows;
3606 3508 : Cost input_startup_cost = *startup_cost;
3607 3508 : Cost input_total_cost = *total_cost;
3608 :
3609 3508 : if (offset_est != 0)
3610 : {
3611 : double offset_rows;
3612 :
3613 564 : if (offset_est > 0)
3614 548 : offset_rows = (double) offset_est;
3615 : else
3616 16 : offset_rows = clamp_row_est(input_rows * 0.10);
3617 564 : if (offset_rows > *rows)
3618 20 : offset_rows = *rows;
3619 564 : if (input_rows > 0)
3620 1128 : *startup_cost +=
3621 564 : (input_total_cost - input_startup_cost)
3622 564 : * offset_rows / input_rows;
3623 564 : *rows -= offset_rows;
3624 564 : if (*rows < 1)
3625 20 : *rows = 1;
3626 : }
3627 :
3628 3508 : if (count_est != 0)
3629 : {
3630 : double count_rows;
3631 :
3632 3492 : if (count_est > 0)
3633 3488 : count_rows = (double) count_est;
3634 : else
3635 4 : count_rows = clamp_row_est(input_rows * 0.10);
3636 3492 : if (count_rows > *rows)
3637 42 : count_rows = *rows;
3638 3492 : if (input_rows > 0)
3639 6984 : *total_cost = *startup_cost +
3640 3492 : (input_total_cost - input_startup_cost)
3641 3492 : * count_rows / input_rows;
3642 3492 : *rows = count_rows;
3643 3492 : if (*rows < 1)
3644 0 : *rows = 1;
3645 : }
3646 3508 : }
3647 :
3648 :
3649 : /*
3650 : * reparameterize_path
3651 : * Attempt to modify a Path to have greater parameterization
3652 : *
3653 : * We use this to attempt to bring all child paths of an appendrel to the
3654 : * same parameterization level, ensuring that they all enforce the same set
3655 : * of join quals (and thus that that parameterization can be attributed to
3656 : * an append path built from such paths). Currently, only a few path types
3657 : * are supported here, though more could be added at need. We return NULL
3658 : * if we can't reparameterize the given path.
3659 : *
3660 : * Note: we intentionally do not pass created paths to add_path(); it would
3661 : * possibly try to delete them on the grounds of being cost-inferior to the
3662 : * paths they were made from, and we don't want that. Paths made here are
3663 : * not necessarily of general-purpose usefulness, but they can be useful
3664 : * as members of an append path.
3665 : */
3666 : Path *
3667 188 : reparameterize_path(PlannerInfo *root, Path *path,
3668 : Relids required_outer,
3669 : double loop_count)
3670 : {
3671 188 : RelOptInfo *rel = path->parent;
3672 :
3673 : /* Can only increase, not decrease, path's parameterization */
3674 188 : if (!bms_is_subset(PATH_REQ_OUTER(path), required_outer))
3675 0 : return NULL;
3676 188 : switch (path->pathtype)
3677 : {
3678 : case T_SeqScan:
3679 124 : return create_seqscan_path(root, rel, required_outer, 0);
3680 : case T_SampleScan:
3681 0 : return (Path *) create_samplescan_path(root, rel, required_outer);
3682 : case T_IndexScan:
3683 : case T_IndexOnlyScan:
3684 : {
3685 0 : IndexPath *ipath = (IndexPath *) path;
3686 0 : IndexPath *newpath = makeNode(IndexPath);
3687 :
3688 : /*
3689 : * We can't use create_index_path directly, and would not want
3690 : * to because it would re-compute the indexqual conditions
3691 : * which is wasted effort. Instead we hack things a bit:
3692 : * flat-copy the path node, revise its param_info, and redo
3693 : * the cost estimate.
3694 : */
3695 0 : memcpy(newpath, ipath, sizeof(IndexPath));
3696 0 : newpath->path.param_info =
3697 0 : get_baserel_parampathinfo(root, rel, required_outer);
3698 0 : cost_index(newpath, root, loop_count, false);
3699 0 : return (Path *) newpath;
3700 : }
3701 : case T_BitmapHeapScan:
3702 : {
3703 0 : BitmapHeapPath *bpath = (BitmapHeapPath *) path;
3704 :
3705 0 : return (Path *) create_bitmap_heap_path(root,
3706 : rel,
3707 : bpath->bitmapqual,
3708 : required_outer,
3709 : loop_count, 0);
3710 : }
3711 : case T_SubqueryScan:
3712 : {
3713 0 : SubqueryScanPath *spath = (SubqueryScanPath *) path;
3714 :
3715 0 : return (Path *) create_subqueryscan_path(root,
3716 : rel,
3717 : spath->subpath,
3718 : spath->path.pathkeys,
3719 : required_outer);
3720 : }
3721 : case T_Result:
3722 : /* Supported only for RTE_RESULT scan paths */
3723 32 : if (IsA(path, Path))
3724 32 : return create_resultscan_path(root, rel, required_outer);
3725 0 : break;
3726 : case T_Append:
3727 : {
3728 0 : AppendPath *apath = (AppendPath *) path;
3729 0 : List *childpaths = NIL;
3730 0 : List *partialpaths = NIL;
3731 : int i;
3732 : ListCell *lc;
3733 :
3734 : /* Reparameterize the children */
3735 0 : i = 0;
3736 0 : foreach(lc, apath->subpaths)
3737 : {
3738 0 : Path *spath = (Path *) lfirst(lc);
3739 :
3740 0 : spath = reparameterize_path(root, spath,
3741 : required_outer,
3742 : loop_count);
3743 0 : if (spath == NULL)
3744 0 : return NULL;
3745 : /* We have to re-split the regular and partial paths */
3746 0 : if (i < apath->first_partial_path)
3747 0 : childpaths = lappend(childpaths, spath);
3748 : else
3749 0 : partialpaths = lappend(partialpaths, spath);
3750 0 : i++;
3751 : }
3752 0 : return (Path *)
3753 0 : create_append_path(root, rel, childpaths, partialpaths,
3754 : apath->path.pathkeys, required_outer,
3755 : apath->path.parallel_workers,
3756 0 : apath->path.parallel_aware,
3757 : apath->partitioned_rels,
3758 : -1);
3759 : }
3760 : default:
3761 32 : break;
3762 : }
3763 32 : return NULL;
3764 : }
3765 :
3766 : /*
3767 : * reparameterize_path_by_child
3768 : * Given a path parameterized by the parent of the given child relation,
3769 : * translate the path to be parameterized by the given child relation.
3770 : *
3771 : * The function creates a new path of the same type as the given path, but
3772 : * parameterized by the given child relation. Most fields from the original
3773 : * path can simply be flat-copied, but any expressions must be adjusted to
3774 : * refer to the correct varnos, and any paths must be recursively
3775 : * reparameterized. Other fields that refer to specific relids also need
3776 : * adjustment.
3777 : *
3778 : * The cost, number of rows, width and parallel path properties depend upon
3779 : * path->parent, which does not change during the translation. Hence those
3780 : * members are copied as they are.
3781 : *
3782 : * If the given path can not be reparameterized, the function returns NULL.
3783 : */
3784 : Path *
3785 2164 : reparameterize_path_by_child(PlannerInfo *root, Path *path,
3786 : RelOptInfo *child_rel)
3787 : {
3788 :
3789 : #define FLAT_COPY_PATH(newnode, node, nodetype) \
3790 : ( (newnode) = makeNode(nodetype), \
3791 : memcpy((newnode), (node), sizeof(nodetype)) )
3792 :
3793 : #define ADJUST_CHILD_ATTRS(node) \
3794 : ((node) = \
3795 : (List *) adjust_appendrel_attrs_multilevel(root, (Node *) (node), \
3796 : child_rel->relids, \
3797 : child_rel->top_parent_relids))
3798 :
3799 : #define REPARAMETERIZE_CHILD_PATH(path) \
3800 : do { \
3801 : (path) = reparameterize_path_by_child(root, (path), child_rel); \
3802 : if ((path) == NULL) \
3803 : return NULL; \
3804 : } while(0);
3805 :
3806 : #define REPARAMETERIZE_CHILD_PATH_LIST(pathlist) \
3807 : do { \
3808 : if ((pathlist) != NIL) \
3809 : { \
3810 : (pathlist) = reparameterize_pathlist_by_child(root, (pathlist), \
3811 : child_rel); \
3812 : if ((pathlist) == NIL) \
3813 : return NULL; \
3814 : } \
3815 : } while(0);
3816 :
3817 : Path *new_path;
3818 : ParamPathInfo *new_ppi;
3819 : ParamPathInfo *old_ppi;
3820 : Relids required_outer;
3821 :
3822 : /*
3823 : * If the path is not parameterized by parent of the given relation, it
3824 : * doesn't need reparameterization.
3825 : */
3826 4304 : if (!path->param_info ||
3827 2140 : !bms_overlap(PATH_REQ_OUTER(path), child_rel->top_parent_relids))
3828 128 : return path;
3829 :
3830 : /* Reparameterize a copy of given path. */
3831 2036 : switch (nodeTag(path))
3832 : {
3833 : case T_Path:
3834 288 : FLAT_COPY_PATH(new_path, path, Path);
3835 288 : break;
3836 :
3837 : case T_IndexPath:
3838 : {
3839 : IndexPath *ipath;
3840 :
3841 1280 : FLAT_COPY_PATH(ipath, path, IndexPath);
3842 1280 : ADJUST_CHILD_ATTRS(ipath->indexclauses);
3843 1280 : new_path = (Path *) ipath;
3844 : }
3845 1280 : break;
3846 :
3847 : case T_BitmapHeapPath:
3848 : {
3849 : BitmapHeapPath *bhpath;
3850 :
3851 0 : FLAT_COPY_PATH(bhpath, path, BitmapHeapPath);
3852 0 : REPARAMETERIZE_CHILD_PATH(bhpath->bitmapqual);
3853 0 : new_path = (Path *) bhpath;
3854 : }
3855 0 : break;
3856 :
3857 : case T_BitmapAndPath:
3858 : {
3859 : BitmapAndPath *bapath;
3860 :
3861 0 : FLAT_COPY_PATH(bapath, path, BitmapAndPath);
3862 0 : REPARAMETERIZE_CHILD_PATH_LIST(bapath->bitmapquals);
3863 0 : new_path = (Path *) bapath;
3864 : }
3865 0 : break;
3866 :
3867 : case T_BitmapOrPath:
3868 : {
3869 : BitmapOrPath *bopath;
3870 :
3871 0 : FLAT_COPY_PATH(bopath, path, BitmapOrPath);
3872 0 : REPARAMETERIZE_CHILD_PATH_LIST(bopath->bitmapquals);
3873 0 : new_path = (Path *) bopath;
3874 : }
3875 0 : break;
3876 :
3877 : case T_TidPath:
3878 : {
3879 : TidPath *tpath;
3880 :
3881 0 : FLAT_COPY_PATH(tpath, path, TidPath);
3882 0 : ADJUST_CHILD_ATTRS(tpath->tidquals);
3883 0 : new_path = (Path *) tpath;
3884 : }
3885 0 : break;
3886 :
3887 : case T_ForeignPath:
3888 : {
3889 : ForeignPath *fpath;
3890 : ReparameterizeForeignPathByChild_function rfpc_func;
3891 :
3892 52 : FLAT_COPY_PATH(fpath, path, ForeignPath);
3893 52 : if (fpath->fdw_outerpath)
3894 0 : REPARAMETERIZE_CHILD_PATH(fpath->fdw_outerpath);
3895 :
3896 : /* Hand over to FDW if needed. */
3897 52 : rfpc_func =
3898 52 : path->parent->fdwroutine->ReparameterizeForeignPathByChild;
3899 52 : if (rfpc_func)
3900 0 : fpath->fdw_private = rfpc_func(root, fpath->fdw_private,
3901 : child_rel);
3902 52 : new_path = (Path *) fpath;
3903 : }
3904 52 : break;
3905 :
3906 : case T_CustomPath:
3907 : {
3908 : CustomPath *cpath;
3909 :
3910 0 : FLAT_COPY_PATH(cpath, path, CustomPath);
3911 0 : REPARAMETERIZE_CHILD_PATH_LIST(cpath->custom_paths);
3912 0 : if (cpath->methods &&
3913 0 : cpath->methods->ReparameterizeCustomPathByChild)
3914 0 : cpath->custom_private =
3915 0 : cpath->methods->ReparameterizeCustomPathByChild(root,
3916 : cpath->custom_private,
3917 : child_rel);
3918 0 : new_path = (Path *) cpath;
3919 : }
3920 0 : break;
3921 :
3922 : case T_NestPath:
3923 : {
3924 : JoinPath *jpath;
3925 :
3926 200 : FLAT_COPY_PATH(jpath, path, NestPath);
3927 :
3928 200 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
3929 200 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
3930 200 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
3931 200 : new_path = (Path *) jpath;
3932 : }
3933 200 : break;
3934 :
3935 : case T_MergePath:
3936 : {
3937 : JoinPath *jpath;
3938 : MergePath *mpath;
3939 :
3940 24 : FLAT_COPY_PATH(mpath, path, MergePath);
3941 :
3942 24 : jpath = (JoinPath *) mpath;
3943 24 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
3944 24 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
3945 24 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
3946 24 : ADJUST_CHILD_ATTRS(mpath->path_mergeclauses);
3947 24 : new_path = (Path *) mpath;
3948 : }
3949 24 : break;
3950 :
3951 : case T_HashPath:
3952 : {
3953 : JoinPath *jpath;
3954 : HashPath *hpath;
3955 :
3956 112 : FLAT_COPY_PATH(hpath, path, HashPath);
3957 :
3958 112 : jpath = (JoinPath *) hpath;
3959 112 : REPARAMETERIZE_CHILD_PATH(jpath->outerjoinpath);
3960 112 : REPARAMETERIZE_CHILD_PATH(jpath->innerjoinpath);
3961 112 : ADJUST_CHILD_ATTRS(jpath->joinrestrictinfo);
3962 112 : ADJUST_CHILD_ATTRS(hpath->path_hashclauses);
3963 112 : new_path = (Path *) hpath;
3964 : }
3965 112 : break;
3966 :
3967 : case T_AppendPath:
3968 : {
3969 : AppendPath *apath;
3970 :
3971 80 : FLAT_COPY_PATH(apath, path, AppendPath);
3972 80 : REPARAMETERIZE_CHILD_PATH_LIST(apath->subpaths);
3973 80 : new_path = (Path *) apath;
3974 : }
3975 80 : break;
3976 :
3977 : case T_MergeAppendPath:
3978 : {
3979 : MergeAppendPath *mapath;
3980 :
3981 0 : FLAT_COPY_PATH(mapath, path, MergeAppendPath);
3982 0 : REPARAMETERIZE_CHILD_PATH_LIST(mapath->subpaths);
3983 0 : new_path = (Path *) mapath;
3984 : }
3985 0 : break;
3986 :
3987 : case T_MaterialPath:
3988 : {
3989 : MaterialPath *mpath;
3990 :
3991 0 : FLAT_COPY_PATH(mpath, path, MaterialPath);
3992 0 : REPARAMETERIZE_CHILD_PATH(mpath->subpath);
3993 0 : new_path = (Path *) mpath;
3994 : }
3995 0 : break;
3996 :
3997 : case T_UniquePath:
3998 : {
3999 : UniquePath *upath;
4000 :
4001 0 : FLAT_COPY_PATH(upath, path, UniquePath);
4002 0 : REPARAMETERIZE_CHILD_PATH(upath->subpath);
4003 0 : ADJUST_CHILD_ATTRS(upath->uniq_exprs);
4004 0 : new_path = (Path *) upath;
4005 : }
4006 0 : break;
4007 :
4008 : case T_GatherPath:
4009 : {
4010 : GatherPath *gpath;
4011 :
4012 0 : FLAT_COPY_PATH(gpath, path, GatherPath);
4013 0 : REPARAMETERIZE_CHILD_PATH(gpath->subpath);
4014 0 : new_path = (Path *) gpath;
4015 : }
4016 0 : break;
4017 :
4018 : case T_GatherMergePath:
4019 : {
4020 : GatherMergePath *gmpath;
4021 :
4022 0 : FLAT_COPY_PATH(gmpath, path, GatherMergePath);
4023 0 : REPARAMETERIZE_CHILD_PATH(gmpath->subpath);
4024 0 : new_path = (Path *) gmpath;
4025 : }
4026 0 : break;
4027 :
4028 : default:
4029 :
4030 : /* We don't know how to reparameterize this path. */
4031 0 : return NULL;
4032 : }
4033 :
4034 : /*
4035 : * Adjust the parameterization information, which refers to the topmost
4036 : * parent. The topmost parent can be multiple levels away from the given
4037 : * child, hence use multi-level expression adjustment routines.
4038 : */
4039 2036 : old_ppi = new_path->param_info;
4040 2036 : required_outer =
4041 2036 : adjust_child_relids_multilevel(root, old_ppi->ppi_req_outer,
4042 : child_rel->relids,
4043 : child_rel->top_parent_relids);
4044 :
4045 : /* If we already have a PPI for this parameterization, just return it */
4046 2036 : new_ppi = find_param_path_info(new_path->parent, required_outer);
4047 :
4048 : /*
4049 : * If not, build a new one and link it to the list of PPIs. For the same
4050 : * reason as explained in mark_dummy_rel(), allocate new PPI in the same
4051 : * context the given RelOptInfo is in.
4052 : */
4053 2036 : if (new_ppi == NULL)
4054 : {
4055 : MemoryContext oldcontext;
4056 956 : RelOptInfo *rel = path->parent;
4057 :
4058 956 : oldcontext = MemoryContextSwitchTo(GetMemoryChunkContext(rel));
4059 :
4060 956 : new_ppi = makeNode(ParamPathInfo);
4061 956 : new_ppi->ppi_req_outer = bms_copy(required_outer);
4062 956 : new_ppi->ppi_rows = old_ppi->ppi_rows;
4063 956 : new_ppi->ppi_clauses = old_ppi->ppi_clauses;
4064 956 : ADJUST_CHILD_ATTRS(new_ppi->ppi_clauses);
4065 956 : rel->ppilist = lappend(rel->ppilist, new_ppi);
4066 :
4067 956 : MemoryContextSwitchTo(oldcontext);
4068 : }
4069 2036 : bms_free(required_outer);
4070 :
4071 2036 : new_path->param_info = new_ppi;
4072 :
4073 : /*
4074 : * Adjust the path target if the parent of the outer relation is
4075 : * referenced in the targetlist. This can happen when only the parent of
4076 : * outer relation is laterally referenced in this relation.
4077 : */
4078 2036 : if (bms_overlap(path->parent->lateral_relids,
4079 2036 : child_rel->top_parent_relids))
4080 : {
4081 768 : new_path->pathtarget = copy_pathtarget(new_path->pathtarget);
4082 768 : ADJUST_CHILD_ATTRS(new_path->pathtarget->exprs);
4083 : }
4084 :
4085 2036 : return new_path;
4086 : }
4087 :
4088 : /*
4089 : * reparameterize_pathlist_by_child
4090 : * Helper function to reparameterize a list of paths by given child rel.
4091 : */
4092 : static List *
4093 80 : reparameterize_pathlist_by_child(PlannerInfo *root,
4094 : List *pathlist,
4095 : RelOptInfo *child_rel)
4096 : {
4097 : ListCell *lc;
4098 80 : List *result = NIL;
4099 :
4100 240 : foreach(lc, pathlist)
4101 : {
4102 160 : Path *path = reparameterize_path_by_child(root, lfirst(lc),
4103 : child_rel);
4104 :
4105 160 : if (path == NULL)
4106 : {
4107 0 : list_free(result);
4108 0 : return NIL;
4109 : }
4110 :
4111 160 : result = lappend(result, path);
4112 : }
4113 :
4114 80 : return result;
4115 : }
|
