/src/postgres/src/backend/optimizer/path/indxpath.c
Line | Count | Source (jump to first uncovered line) |
1 | | /*------------------------------------------------------------------------- |
2 | | * |
3 | | * indxpath.c |
4 | | * Routines to determine which indexes are usable for scanning a |
5 | | * given relation, and create Paths accordingly. |
6 | | * |
7 | | * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
8 | | * Portions Copyright (c) 1994, Regents of the University of California |
9 | | * |
10 | | * |
11 | | * IDENTIFICATION |
12 | | * src/backend/optimizer/path/indxpath.c |
13 | | * |
14 | | *------------------------------------------------------------------------- |
15 | | */ |
16 | | #include "postgres.h" |
17 | | |
18 | | #include <math.h> |
19 | | |
20 | | #include "access/stratnum.h" |
21 | | #include "access/sysattr.h" |
22 | | #include "catalog/pg_am.h" |
23 | | #include "catalog/pg_amop.h" |
24 | | #include "catalog/pg_operator.h" |
25 | | #include "catalog/pg_opfamily.h" |
26 | | #include "catalog/pg_type.h" |
27 | | #include "nodes/makefuncs.h" |
28 | | #include "nodes/nodeFuncs.h" |
29 | | #include "nodes/supportnodes.h" |
30 | | #include "optimizer/cost.h" |
31 | | #include "optimizer/optimizer.h" |
32 | | #include "optimizer/pathnode.h" |
33 | | #include "optimizer/paths.h" |
34 | | #include "optimizer/prep.h" |
35 | | #include "optimizer/restrictinfo.h" |
36 | | #include "utils/lsyscache.h" |
37 | | #include "utils/selfuncs.h" |
38 | | |
39 | | |
40 | | /* XXX see PartCollMatchesExprColl */ |
41 | | #define IndexCollMatchesExprColl(idxcollation, exprcollation) \ |
42 | 0 | ((idxcollation) == InvalidOid || (idxcollation) == (exprcollation)) |
43 | | |
44 | | /* Whether we are looking for plain indexscan, bitmap scan, or either */ |
45 | | typedef enum |
46 | | { |
47 | | ST_INDEXSCAN, /* must support amgettuple */ |
48 | | ST_BITMAPSCAN, /* must support amgetbitmap */ |
49 | | ST_ANYSCAN, /* either is okay */ |
50 | | } ScanTypeControl; |
51 | | |
52 | | /* Data structure for collecting qual clauses that match an index */ |
53 | | typedef struct |
54 | | { |
55 | | bool nonempty; /* True if lists are not all empty */ |
56 | | /* Lists of IndexClause nodes, one list per index column */ |
57 | | List *indexclauses[INDEX_MAX_KEYS]; |
58 | | } IndexClauseSet; |
59 | | |
60 | | /* Per-path data used within choose_bitmap_and() */ |
61 | | typedef struct |
62 | | { |
63 | | Path *path; /* IndexPath, BitmapAndPath, or BitmapOrPath */ |
64 | | List *quals; /* the WHERE clauses it uses */ |
65 | | List *preds; /* predicates of its partial index(es) */ |
66 | | Bitmapset *clauseids; /* quals+preds represented as a bitmapset */ |
67 | | bool unclassifiable; /* has too many quals+preds to process? */ |
68 | | } PathClauseUsage; |
69 | | |
70 | | /* Callback argument for ec_member_matches_indexcol */ |
71 | | typedef struct |
72 | | { |
73 | | IndexOptInfo *index; /* index we're considering */ |
74 | | int indexcol; /* index column we want to match to */ |
75 | | } ec_member_matches_arg; |
76 | | |
77 | | |
78 | | static void consider_index_join_clauses(PlannerInfo *root, RelOptInfo *rel, |
79 | | IndexOptInfo *index, |
80 | | IndexClauseSet *rclauseset, |
81 | | IndexClauseSet *jclauseset, |
82 | | IndexClauseSet *eclauseset, |
83 | | List **bitindexpaths); |
84 | | static void consider_index_join_outer_rels(PlannerInfo *root, RelOptInfo *rel, |
85 | | IndexOptInfo *index, |
86 | | IndexClauseSet *rclauseset, |
87 | | IndexClauseSet *jclauseset, |
88 | | IndexClauseSet *eclauseset, |
89 | | List **bitindexpaths, |
90 | | List *indexjoinclauses, |
91 | | int considered_clauses, |
92 | | List **considered_relids); |
93 | | static void get_join_index_paths(PlannerInfo *root, RelOptInfo *rel, |
94 | | IndexOptInfo *index, |
95 | | IndexClauseSet *rclauseset, |
96 | | IndexClauseSet *jclauseset, |
97 | | IndexClauseSet *eclauseset, |
98 | | List **bitindexpaths, |
99 | | Relids relids, |
100 | | List **considered_relids); |
101 | | static bool eclass_already_used(EquivalenceClass *parent_ec, Relids oldrelids, |
102 | | List *indexjoinclauses); |
103 | | static void get_index_paths(PlannerInfo *root, RelOptInfo *rel, |
104 | | IndexOptInfo *index, IndexClauseSet *clauses, |
105 | | List **bitindexpaths); |
106 | | static List *build_index_paths(PlannerInfo *root, RelOptInfo *rel, |
107 | | IndexOptInfo *index, IndexClauseSet *clauses, |
108 | | bool useful_predicate, |
109 | | ScanTypeControl scantype, |
110 | | bool *skip_nonnative_saop); |
111 | | static List *build_paths_for_OR(PlannerInfo *root, RelOptInfo *rel, |
112 | | List *clauses, List *other_clauses); |
113 | | static List *generate_bitmap_or_paths(PlannerInfo *root, RelOptInfo *rel, |
114 | | List *clauses, List *other_clauses); |
115 | | static Path *choose_bitmap_and(PlannerInfo *root, RelOptInfo *rel, |
116 | | List *paths); |
117 | | static int path_usage_comparator(const void *a, const void *b); |
118 | | static Cost bitmap_scan_cost_est(PlannerInfo *root, RelOptInfo *rel, |
119 | | Path *ipath); |
120 | | static Cost bitmap_and_cost_est(PlannerInfo *root, RelOptInfo *rel, |
121 | | List *paths); |
122 | | static PathClauseUsage *classify_index_clause_usage(Path *path, |
123 | | List **clauselist); |
124 | | static void find_indexpath_quals(Path *bitmapqual, List **quals, List **preds); |
125 | | static int find_list_position(Node *node, List **nodelist); |
126 | | static bool check_index_only(RelOptInfo *rel, IndexOptInfo *index); |
127 | | static double get_loop_count(PlannerInfo *root, Index cur_relid, Relids outer_relids); |
128 | | static double adjust_rowcount_for_semijoins(PlannerInfo *root, |
129 | | Index cur_relid, |
130 | | Index outer_relid, |
131 | | double rowcount); |
132 | | static double approximate_joinrel_size(PlannerInfo *root, Relids relids); |
133 | | static void match_restriction_clauses_to_index(PlannerInfo *root, |
134 | | IndexOptInfo *index, |
135 | | IndexClauseSet *clauseset); |
136 | | static void match_join_clauses_to_index(PlannerInfo *root, |
137 | | RelOptInfo *rel, IndexOptInfo *index, |
138 | | IndexClauseSet *clauseset, |
139 | | List **joinorclauses); |
140 | | static void match_eclass_clauses_to_index(PlannerInfo *root, |
141 | | IndexOptInfo *index, |
142 | | IndexClauseSet *clauseset); |
143 | | static void match_clauses_to_index(PlannerInfo *root, |
144 | | List *clauses, |
145 | | IndexOptInfo *index, |
146 | | IndexClauseSet *clauseset); |
147 | | static void match_clause_to_index(PlannerInfo *root, |
148 | | RestrictInfo *rinfo, |
149 | | IndexOptInfo *index, |
150 | | IndexClauseSet *clauseset); |
151 | | static IndexClause *match_clause_to_indexcol(PlannerInfo *root, |
152 | | RestrictInfo *rinfo, |
153 | | int indexcol, |
154 | | IndexOptInfo *index); |
155 | | static bool IsBooleanOpfamily(Oid opfamily); |
156 | | static IndexClause *match_boolean_index_clause(PlannerInfo *root, |
157 | | RestrictInfo *rinfo, |
158 | | int indexcol, IndexOptInfo *index); |
159 | | static IndexClause *match_opclause_to_indexcol(PlannerInfo *root, |
160 | | RestrictInfo *rinfo, |
161 | | int indexcol, |
162 | | IndexOptInfo *index); |
163 | | static IndexClause *match_funcclause_to_indexcol(PlannerInfo *root, |
164 | | RestrictInfo *rinfo, |
165 | | int indexcol, |
166 | | IndexOptInfo *index); |
167 | | static IndexClause *get_index_clause_from_support(PlannerInfo *root, |
168 | | RestrictInfo *rinfo, |
169 | | Oid funcid, |
170 | | int indexarg, |
171 | | int indexcol, |
172 | | IndexOptInfo *index); |
173 | | static IndexClause *match_saopclause_to_indexcol(PlannerInfo *root, |
174 | | RestrictInfo *rinfo, |
175 | | int indexcol, |
176 | | IndexOptInfo *index); |
177 | | static IndexClause *match_rowcompare_to_indexcol(PlannerInfo *root, |
178 | | RestrictInfo *rinfo, |
179 | | int indexcol, |
180 | | IndexOptInfo *index); |
181 | | static IndexClause *match_orclause_to_indexcol(PlannerInfo *root, |
182 | | RestrictInfo *rinfo, |
183 | | int indexcol, |
184 | | IndexOptInfo *index); |
185 | | static IndexClause *expand_indexqual_rowcompare(PlannerInfo *root, |
186 | | RestrictInfo *rinfo, |
187 | | int indexcol, |
188 | | IndexOptInfo *index, |
189 | | Oid expr_op, |
190 | | bool var_on_left); |
191 | | static void match_pathkeys_to_index(IndexOptInfo *index, List *pathkeys, |
192 | | List **orderby_clauses_p, |
193 | | List **clause_columns_p); |
194 | | static Expr *match_clause_to_ordering_op(IndexOptInfo *index, |
195 | | int indexcol, Expr *clause, Oid pk_opfamily); |
196 | | static bool ec_member_matches_indexcol(PlannerInfo *root, RelOptInfo *rel, |
197 | | EquivalenceClass *ec, EquivalenceMember *em, |
198 | | void *arg); |
199 | | |
200 | | |
201 | | /* |
202 | | * create_index_paths() |
203 | | * Generate all interesting index paths for the given relation. |
204 | | * Candidate paths are added to the rel's pathlist (using add_path). |
205 | | * |
206 | | * To be considered for an index scan, an index must match one or more |
207 | | * restriction clauses or join clauses from the query's qual condition, |
208 | | * or match the query's ORDER BY condition, or have a predicate that |
209 | | * matches the query's qual condition. |
210 | | * |
211 | | * There are two basic kinds of index scans. A "plain" index scan uses |
212 | | * only restriction clauses (possibly none at all) in its indexqual, |
213 | | * so it can be applied in any context. A "parameterized" index scan uses |
214 | | * join clauses (plus restriction clauses, if available) in its indexqual. |
215 | | * When joining such a scan to one of the relations supplying the other |
216 | | * variables used in its indexqual, the parameterized scan must appear as |
217 | | * the inner relation of a nestloop join; it can't be used on the outer side, |
218 | | * nor in a merge or hash join. In that context, values for the other rels' |
219 | | * attributes are available and fixed during any one scan of the indexpath. |
220 | | * |
221 | | * An IndexPath is generated and submitted to add_path() for each plain or |
222 | | * parameterized index scan this routine deems potentially interesting for |
223 | | * the current query. |
224 | | * |
225 | | * 'rel' is the relation for which we want to generate index paths |
226 | | * |
227 | | * Note: check_index_predicates() must have been run previously for this rel. |
228 | | * |
229 | | * Note: in cases involving LATERAL references in the relation's tlist, it's |
230 | | * possible that rel->lateral_relids is nonempty. Currently, we include |
231 | | * lateral_relids into the parameterization reported for each path, but don't |
232 | | * take it into account otherwise. The fact that any such rels *must* be |
233 | | * available as parameter sources perhaps should influence our choices of |
234 | | * index quals ... but for now, it doesn't seem worth troubling over. |
235 | | * In particular, comments below about "unparameterized" paths should be read |
236 | | * as meaning "unparameterized so far as the indexquals are concerned". |
237 | | */ |
238 | | void |
239 | | create_index_paths(PlannerInfo *root, RelOptInfo *rel) |
240 | 0 | { |
241 | 0 | List *indexpaths; |
242 | 0 | List *bitindexpaths; |
243 | 0 | List *bitjoinpaths; |
244 | 0 | List *joinorclauses; |
245 | 0 | IndexClauseSet rclauseset; |
246 | 0 | IndexClauseSet jclauseset; |
247 | 0 | IndexClauseSet eclauseset; |
248 | 0 | ListCell *lc; |
249 | | |
250 | | /* Skip the whole mess if no indexes */ |
251 | 0 | if (rel->indexlist == NIL) |
252 | 0 | return; |
253 | | |
254 | | /* Bitmap paths are collected and then dealt with at the end */ |
255 | 0 | bitindexpaths = bitjoinpaths = joinorclauses = NIL; |
256 | | |
257 | | /* Examine each index in turn */ |
258 | 0 | foreach(lc, rel->indexlist) |
259 | 0 | { |
260 | 0 | IndexOptInfo *index = (IndexOptInfo *) lfirst(lc); |
261 | | |
262 | | /* Protect limited-size array in IndexClauseSets */ |
263 | 0 | Assert(index->nkeycolumns <= INDEX_MAX_KEYS); |
264 | | |
265 | | /* |
266 | | * Ignore partial indexes that do not match the query. |
267 | | * (generate_bitmap_or_paths() might be able to do something with |
268 | | * them, but that's of no concern here.) |
269 | | */ |
270 | 0 | if (index->indpred != NIL && !index->predOK) |
271 | 0 | continue; |
272 | | |
273 | | /* |
274 | | * Identify the restriction clauses that can match the index. |
275 | | */ |
276 | 0 | MemSet(&rclauseset, 0, sizeof(rclauseset)); |
277 | 0 | match_restriction_clauses_to_index(root, index, &rclauseset); |
278 | | |
279 | | /* |
280 | | * Build index paths from the restriction clauses. These will be |
281 | | * non-parameterized paths. Plain paths go directly to add_path(), |
282 | | * bitmap paths are added to bitindexpaths to be handled below. |
283 | | */ |
284 | 0 | get_index_paths(root, rel, index, &rclauseset, |
285 | 0 | &bitindexpaths); |
286 | | |
287 | | /* |
288 | | * Identify the join clauses that can match the index. For the moment |
289 | | * we keep them separate from the restriction clauses. Note that this |
290 | | * step finds only "loose" join clauses that have not been merged into |
291 | | * EquivalenceClasses. Also, collect join OR clauses for later. |
292 | | */ |
293 | 0 | MemSet(&jclauseset, 0, sizeof(jclauseset)); |
294 | 0 | match_join_clauses_to_index(root, rel, index, |
295 | 0 | &jclauseset, &joinorclauses); |
296 | | |
297 | | /* |
298 | | * Look for EquivalenceClasses that can generate joinclauses matching |
299 | | * the index. |
300 | | */ |
301 | 0 | MemSet(&eclauseset, 0, sizeof(eclauseset)); |
302 | 0 | match_eclass_clauses_to_index(root, index, |
303 | 0 | &eclauseset); |
304 | | |
305 | | /* |
306 | | * If we found any plain or eclass join clauses, build parameterized |
307 | | * index paths using them. |
308 | | */ |
309 | 0 | if (jclauseset.nonempty || eclauseset.nonempty) |
310 | 0 | consider_index_join_clauses(root, rel, index, |
311 | 0 | &rclauseset, |
312 | 0 | &jclauseset, |
313 | 0 | &eclauseset, |
314 | 0 | &bitjoinpaths); |
315 | 0 | } |
316 | | |
317 | | /* |
318 | | * Generate BitmapOrPaths for any suitable OR-clauses present in the |
319 | | * restriction list. Add these to bitindexpaths. |
320 | | */ |
321 | 0 | indexpaths = generate_bitmap_or_paths(root, rel, |
322 | 0 | rel->baserestrictinfo, NIL); |
323 | 0 | bitindexpaths = list_concat(bitindexpaths, indexpaths); |
324 | | |
325 | | /* |
326 | | * Likewise, generate BitmapOrPaths for any suitable OR-clauses present in |
327 | | * the joinclause list. Add these to bitjoinpaths. |
328 | | */ |
329 | 0 | indexpaths = generate_bitmap_or_paths(root, rel, |
330 | 0 | joinorclauses, rel->baserestrictinfo); |
331 | 0 | bitjoinpaths = list_concat(bitjoinpaths, indexpaths); |
332 | | |
333 | | /* |
334 | | * If we found anything usable, generate a BitmapHeapPath for the most |
335 | | * promising combination of restriction bitmap index paths. Note there |
336 | | * will be only one such path no matter how many indexes exist. This |
337 | | * should be sufficient since there's basically only one figure of merit |
338 | | * (total cost) for such a path. |
339 | | */ |
340 | 0 | if (bitindexpaths != NIL) |
341 | 0 | { |
342 | 0 | Path *bitmapqual; |
343 | 0 | BitmapHeapPath *bpath; |
344 | |
|
345 | 0 | bitmapqual = choose_bitmap_and(root, rel, bitindexpaths); |
346 | 0 | bpath = create_bitmap_heap_path(root, rel, bitmapqual, |
347 | 0 | rel->lateral_relids, 1.0, 0); |
348 | 0 | add_path(rel, (Path *) bpath); |
349 | | |
350 | | /* create a partial bitmap heap path */ |
351 | 0 | if (rel->consider_parallel && rel->lateral_relids == NULL) |
352 | 0 | create_partial_bitmap_paths(root, rel, bitmapqual); |
353 | 0 | } |
354 | | |
355 | | /* |
356 | | * Likewise, if we found anything usable, generate BitmapHeapPaths for the |
357 | | * most promising combinations of join bitmap index paths. Our strategy |
358 | | * is to generate one such path for each distinct parameterization seen |
359 | | * among the available bitmap index paths. This may look pretty |
360 | | * expensive, but usually there won't be very many distinct |
361 | | * parameterizations. (This logic is quite similar to that in |
362 | | * consider_index_join_clauses, but we're working with whole paths not |
363 | | * individual clauses.) |
364 | | */ |
365 | 0 | if (bitjoinpaths != NIL) |
366 | 0 | { |
367 | 0 | List *all_path_outers; |
368 | | |
369 | | /* Identify each distinct parameterization seen in bitjoinpaths */ |
370 | 0 | all_path_outers = NIL; |
371 | 0 | foreach(lc, bitjoinpaths) |
372 | 0 | { |
373 | 0 | Path *path = (Path *) lfirst(lc); |
374 | 0 | Relids required_outer = PATH_REQ_OUTER(path); |
375 | |
|
376 | 0 | all_path_outers = list_append_unique(all_path_outers, |
377 | 0 | required_outer); |
378 | 0 | } |
379 | | |
380 | | /* Now, for each distinct parameterization set ... */ |
381 | 0 | foreach(lc, all_path_outers) |
382 | 0 | { |
383 | 0 | Relids max_outers = (Relids) lfirst(lc); |
384 | 0 | List *this_path_set; |
385 | 0 | Path *bitmapqual; |
386 | 0 | Relids required_outer; |
387 | 0 | double loop_count; |
388 | 0 | BitmapHeapPath *bpath; |
389 | 0 | ListCell *lcp; |
390 | | |
391 | | /* Identify all the bitmap join paths needing no more than that */ |
392 | 0 | this_path_set = NIL; |
393 | 0 | foreach(lcp, bitjoinpaths) |
394 | 0 | { |
395 | 0 | Path *path = (Path *) lfirst(lcp); |
396 | |
|
397 | 0 | if (bms_is_subset(PATH_REQ_OUTER(path), max_outers)) |
398 | 0 | this_path_set = lappend(this_path_set, path); |
399 | 0 | } |
400 | | |
401 | | /* |
402 | | * Add in restriction bitmap paths, since they can be used |
403 | | * together with any join paths. |
404 | | */ |
405 | 0 | this_path_set = list_concat(this_path_set, bitindexpaths); |
406 | | |
407 | | /* Select best AND combination for this parameterization */ |
408 | 0 | bitmapqual = choose_bitmap_and(root, rel, this_path_set); |
409 | | |
410 | | /* And push that path into the mix */ |
411 | 0 | required_outer = PATH_REQ_OUTER(bitmapqual); |
412 | 0 | loop_count = get_loop_count(root, rel->relid, required_outer); |
413 | 0 | bpath = create_bitmap_heap_path(root, rel, bitmapqual, |
414 | 0 | required_outer, loop_count, 0); |
415 | 0 | add_path(rel, (Path *) bpath); |
416 | 0 | } |
417 | 0 | } |
418 | 0 | } |
419 | | |
420 | | /* |
421 | | * consider_index_join_clauses |
422 | | * Given sets of join clauses for an index, decide which parameterized |
423 | | * index paths to build. |
424 | | * |
425 | | * Plain indexpaths are sent directly to add_path, while potential |
426 | | * bitmap indexpaths are added to *bitindexpaths for later processing. |
427 | | * |
428 | | * 'rel' is the index's heap relation |
429 | | * 'index' is the index for which we want to generate paths |
430 | | * 'rclauseset' is the collection of indexable restriction clauses |
431 | | * 'jclauseset' is the collection of indexable simple join clauses |
432 | | * 'eclauseset' is the collection of indexable clauses from EquivalenceClasses |
433 | | * '*bitindexpaths' is the list to add bitmap paths to |
434 | | */ |
435 | | static void |
436 | | consider_index_join_clauses(PlannerInfo *root, RelOptInfo *rel, |
437 | | IndexOptInfo *index, |
438 | | IndexClauseSet *rclauseset, |
439 | | IndexClauseSet *jclauseset, |
440 | | IndexClauseSet *eclauseset, |
441 | | List **bitindexpaths) |
442 | 0 | { |
443 | 0 | int considered_clauses = 0; |
444 | 0 | List *considered_relids = NIL; |
445 | 0 | int indexcol; |
446 | | |
447 | | /* |
448 | | * The strategy here is to identify every potentially useful set of outer |
449 | | * rels that can provide indexable join clauses. For each such set, |
450 | | * select all the join clauses available from those outer rels, add on all |
451 | | * the indexable restriction clauses, and generate plain and/or bitmap |
452 | | * index paths for that set of clauses. This is based on the assumption |
453 | | * that it's always better to apply a clause as an indexqual than as a |
454 | | * filter (qpqual); which is where an available clause would end up being |
455 | | * applied if we omit it from the indexquals. |
456 | | * |
457 | | * This looks expensive, but in most practical cases there won't be very |
458 | | * many distinct sets of outer rels to consider. As a safety valve when |
459 | | * that's not true, we use a heuristic: limit the number of outer rel sets |
460 | | * considered to a multiple of the number of clauses considered. (We'll |
461 | | * always consider using each individual join clause, though.) |
462 | | * |
463 | | * For simplicity in selecting relevant clauses, we represent each set of |
464 | | * outer rels as a maximum set of clause_relids --- that is, the indexed |
465 | | * relation itself is also included in the relids set. considered_relids |
466 | | * lists all relids sets we've already tried. |
467 | | */ |
468 | 0 | for (indexcol = 0; indexcol < index->nkeycolumns; indexcol++) |
469 | 0 | { |
470 | | /* Consider each applicable simple join clause */ |
471 | 0 | considered_clauses += list_length(jclauseset->indexclauses[indexcol]); |
472 | 0 | consider_index_join_outer_rels(root, rel, index, |
473 | 0 | rclauseset, jclauseset, eclauseset, |
474 | 0 | bitindexpaths, |
475 | 0 | jclauseset->indexclauses[indexcol], |
476 | 0 | considered_clauses, |
477 | 0 | &considered_relids); |
478 | | /* Consider each applicable eclass join clause */ |
479 | 0 | considered_clauses += list_length(eclauseset->indexclauses[indexcol]); |
480 | 0 | consider_index_join_outer_rels(root, rel, index, |
481 | 0 | rclauseset, jclauseset, eclauseset, |
482 | 0 | bitindexpaths, |
483 | 0 | eclauseset->indexclauses[indexcol], |
484 | 0 | considered_clauses, |
485 | 0 | &considered_relids); |
486 | 0 | } |
487 | 0 | } |
488 | | |
489 | | /* |
490 | | * consider_index_join_outer_rels |
491 | | * Generate parameterized paths based on clause relids in the clause list. |
492 | | * |
493 | | * Workhorse for consider_index_join_clauses; see notes therein for rationale. |
494 | | * |
495 | | * 'rel', 'index', 'rclauseset', 'jclauseset', 'eclauseset', and |
496 | | * 'bitindexpaths' as above |
497 | | * 'indexjoinclauses' is a list of IndexClauses for join clauses |
498 | | * 'considered_clauses' is the total number of clauses considered (so far) |
499 | | * '*considered_relids' is a list of all relids sets already considered |
500 | | */ |
501 | | static void |
502 | | consider_index_join_outer_rels(PlannerInfo *root, RelOptInfo *rel, |
503 | | IndexOptInfo *index, |
504 | | IndexClauseSet *rclauseset, |
505 | | IndexClauseSet *jclauseset, |
506 | | IndexClauseSet *eclauseset, |
507 | | List **bitindexpaths, |
508 | | List *indexjoinclauses, |
509 | | int considered_clauses, |
510 | | List **considered_relids) |
511 | 0 | { |
512 | 0 | ListCell *lc; |
513 | | |
514 | | /* Examine relids of each joinclause in the given list */ |
515 | 0 | foreach(lc, indexjoinclauses) |
516 | 0 | { |
517 | 0 | IndexClause *iclause = (IndexClause *) lfirst(lc); |
518 | 0 | Relids clause_relids = iclause->rinfo->clause_relids; |
519 | 0 | EquivalenceClass *parent_ec = iclause->rinfo->parent_ec; |
520 | 0 | int num_considered_relids; |
521 | | |
522 | | /* If we already tried its relids set, no need to do so again */ |
523 | 0 | if (list_member(*considered_relids, clause_relids)) |
524 | 0 | continue; |
525 | | |
526 | | /* |
527 | | * Generate the union of this clause's relids set with each |
528 | | * previously-tried set. This ensures we try this clause along with |
529 | | * every interesting subset of previous clauses. However, to avoid |
530 | | * exponential growth of planning time when there are many clauses, |
531 | | * limit the number of relid sets accepted to 10 * considered_clauses. |
532 | | * |
533 | | * Note: get_join_index_paths appends entries to *considered_relids, |
534 | | * but we do not need to visit such newly-added entries within this |
535 | | * loop, so we don't use foreach() here. No real harm would be done |
536 | | * if we did visit them, since the subset check would reject them; but |
537 | | * it would waste some cycles. |
538 | | */ |
539 | 0 | num_considered_relids = list_length(*considered_relids); |
540 | 0 | for (int pos = 0; pos < num_considered_relids; pos++) |
541 | 0 | { |
542 | 0 | Relids oldrelids = (Relids) list_nth(*considered_relids, pos); |
543 | | |
544 | | /* |
545 | | * If either is a subset of the other, no new set is possible. |
546 | | * This isn't a complete test for redundancy, but it's easy and |
547 | | * cheap. get_join_index_paths will check more carefully if we |
548 | | * already generated the same relids set. |
549 | | */ |
550 | 0 | if (bms_subset_compare(clause_relids, oldrelids) != BMS_DIFFERENT) |
551 | 0 | continue; |
552 | | |
553 | | /* |
554 | | * If this clause was derived from an equivalence class, the |
555 | | * clause list may contain other clauses derived from the same |
556 | | * eclass. We should not consider that combining this clause with |
557 | | * one of those clauses generates a usefully different |
558 | | * parameterization; so skip if any clause derived from the same |
559 | | * eclass would already have been included when using oldrelids. |
560 | | */ |
561 | 0 | if (parent_ec && |
562 | 0 | eclass_already_used(parent_ec, oldrelids, |
563 | 0 | indexjoinclauses)) |
564 | 0 | continue; |
565 | | |
566 | | /* |
567 | | * If the number of relid sets considered exceeds our heuristic |
568 | | * limit, stop considering combinations of clauses. We'll still |
569 | | * consider the current clause alone, though (below this loop). |
570 | | */ |
571 | 0 | if (list_length(*considered_relids) >= 10 * considered_clauses) |
572 | 0 | break; |
573 | | |
574 | | /* OK, try the union set */ |
575 | 0 | get_join_index_paths(root, rel, index, |
576 | 0 | rclauseset, jclauseset, eclauseset, |
577 | 0 | bitindexpaths, |
578 | 0 | bms_union(clause_relids, oldrelids), |
579 | 0 | considered_relids); |
580 | 0 | } |
581 | | |
582 | | /* Also try this set of relids by itself */ |
583 | 0 | get_join_index_paths(root, rel, index, |
584 | 0 | rclauseset, jclauseset, eclauseset, |
585 | 0 | bitindexpaths, |
586 | 0 | clause_relids, |
587 | 0 | considered_relids); |
588 | 0 | } |
589 | 0 | } |
590 | | |
591 | | /* |
592 | | * get_join_index_paths |
593 | | * Generate index paths using clauses from the specified outer relations. |
594 | | * In addition to generating paths, relids is added to *considered_relids |
595 | | * if not already present. |
596 | | * |
597 | | * Workhorse for consider_index_join_clauses; see notes therein for rationale. |
598 | | * |
599 | | * 'rel', 'index', 'rclauseset', 'jclauseset', 'eclauseset', |
600 | | * 'bitindexpaths', 'considered_relids' as above |
601 | | * 'relids' is the current set of relids to consider (the target rel plus |
602 | | * one or more outer rels) |
603 | | */ |
604 | | static void |
605 | | get_join_index_paths(PlannerInfo *root, RelOptInfo *rel, |
606 | | IndexOptInfo *index, |
607 | | IndexClauseSet *rclauseset, |
608 | | IndexClauseSet *jclauseset, |
609 | | IndexClauseSet *eclauseset, |
610 | | List **bitindexpaths, |
611 | | Relids relids, |
612 | | List **considered_relids) |
613 | 0 | { |
614 | 0 | IndexClauseSet clauseset; |
615 | 0 | int indexcol; |
616 | | |
617 | | /* If we already considered this relids set, don't repeat the work */ |
618 | 0 | if (list_member(*considered_relids, relids)) |
619 | 0 | return; |
620 | | |
621 | | /* Identify indexclauses usable with this relids set */ |
622 | 0 | MemSet(&clauseset, 0, sizeof(clauseset)); |
623 | |
|
624 | 0 | for (indexcol = 0; indexcol < index->nkeycolumns; indexcol++) |
625 | 0 | { |
626 | 0 | ListCell *lc; |
627 | | |
628 | | /* First find applicable simple join clauses */ |
629 | 0 | foreach(lc, jclauseset->indexclauses[indexcol]) |
630 | 0 | { |
631 | 0 | IndexClause *iclause = (IndexClause *) lfirst(lc); |
632 | |
|
633 | 0 | if (bms_is_subset(iclause->rinfo->clause_relids, relids)) |
634 | 0 | clauseset.indexclauses[indexcol] = |
635 | 0 | lappend(clauseset.indexclauses[indexcol], iclause); |
636 | 0 | } |
637 | | |
638 | | /* |
639 | | * Add applicable eclass join clauses. The clauses generated for each |
640 | | * column are redundant (cf generate_implied_equalities_for_column), |
641 | | * so we need at most one. This is the only exception to the general |
642 | | * rule of using all available index clauses. |
643 | | */ |
644 | 0 | foreach(lc, eclauseset->indexclauses[indexcol]) |
645 | 0 | { |
646 | 0 | IndexClause *iclause = (IndexClause *) lfirst(lc); |
647 | |
|
648 | 0 | if (bms_is_subset(iclause->rinfo->clause_relids, relids)) |
649 | 0 | { |
650 | 0 | clauseset.indexclauses[indexcol] = |
651 | 0 | lappend(clauseset.indexclauses[indexcol], iclause); |
652 | 0 | break; |
653 | 0 | } |
654 | 0 | } |
655 | | |
656 | | /* Add restriction clauses */ |
657 | 0 | clauseset.indexclauses[indexcol] = |
658 | 0 | list_concat(clauseset.indexclauses[indexcol], |
659 | 0 | rclauseset->indexclauses[indexcol]); |
660 | |
|
661 | 0 | if (clauseset.indexclauses[indexcol] != NIL) |
662 | 0 | clauseset.nonempty = true; |
663 | 0 | } |
664 | | |
665 | | /* We should have found something, else caller passed silly relids */ |
666 | 0 | Assert(clauseset.nonempty); |
667 | | |
668 | | /* Build index path(s) using the collected set of clauses */ |
669 | 0 | get_index_paths(root, rel, index, &clauseset, bitindexpaths); |
670 | | |
671 | | /* |
672 | | * Remember we considered paths for this set of relids. |
673 | | */ |
674 | 0 | *considered_relids = lappend(*considered_relids, relids); |
675 | 0 | } |
676 | | |
677 | | /* |
678 | | * eclass_already_used |
679 | | * True if any join clause usable with oldrelids was generated from |
680 | | * the specified equivalence class. |
681 | | */ |
682 | | static bool |
683 | | eclass_already_used(EquivalenceClass *parent_ec, Relids oldrelids, |
684 | | List *indexjoinclauses) |
685 | 0 | { |
686 | 0 | ListCell *lc; |
687 | |
|
688 | 0 | foreach(lc, indexjoinclauses) |
689 | 0 | { |
690 | 0 | IndexClause *iclause = (IndexClause *) lfirst(lc); |
691 | 0 | RestrictInfo *rinfo = iclause->rinfo; |
692 | |
|
693 | 0 | if (rinfo->parent_ec == parent_ec && |
694 | 0 | bms_is_subset(rinfo->clause_relids, oldrelids)) |
695 | 0 | return true; |
696 | 0 | } |
697 | 0 | return false; |
698 | 0 | } |
699 | | |
700 | | |
701 | | /* |
702 | | * get_index_paths |
703 | | * Given an index and a set of index clauses for it, construct IndexPaths. |
704 | | * |
705 | | * Plain indexpaths are sent directly to add_path, while potential |
706 | | * bitmap indexpaths are added to *bitindexpaths for later processing. |
707 | | * |
708 | | * This is a fairly simple frontend to build_index_paths(). Its reason for |
709 | | * existence is mainly to handle ScalarArrayOpExpr quals properly. If the |
710 | | * index AM supports them natively, we should just include them in simple |
711 | | * index paths. If not, we should exclude them while building simple index |
712 | | * paths, and then make a separate attempt to include them in bitmap paths. |
713 | | */ |
714 | | static void |
715 | | get_index_paths(PlannerInfo *root, RelOptInfo *rel, |
716 | | IndexOptInfo *index, IndexClauseSet *clauses, |
717 | | List **bitindexpaths) |
718 | 0 | { |
719 | 0 | List *indexpaths; |
720 | 0 | bool skip_nonnative_saop = false; |
721 | 0 | ListCell *lc; |
722 | | |
723 | | /* |
724 | | * Build simple index paths using the clauses. Allow ScalarArrayOpExpr |
725 | | * clauses only if the index AM supports them natively. |
726 | | */ |
727 | 0 | indexpaths = build_index_paths(root, rel, |
728 | 0 | index, clauses, |
729 | 0 | index->predOK, |
730 | 0 | ST_ANYSCAN, |
731 | 0 | &skip_nonnative_saop); |
732 | | |
733 | | /* |
734 | | * Submit all the ones that can form plain IndexScan plans to add_path. (A |
735 | | * plain IndexPath can represent either a plain IndexScan or an |
736 | | * IndexOnlyScan, but for our purposes here that distinction does not |
737 | | * matter. However, some of the indexes might support only bitmap scans, |
738 | | * and those we mustn't submit to add_path here.) |
739 | | * |
740 | | * Also, pick out the ones that are usable as bitmap scans. For that, we |
741 | | * must discard indexes that don't support bitmap scans, and we also are |
742 | | * only interested in paths that have some selectivity; we should discard |
743 | | * anything that was generated solely for ordering purposes. |
744 | | */ |
745 | 0 | foreach(lc, indexpaths) |
746 | 0 | { |
747 | 0 | IndexPath *ipath = (IndexPath *) lfirst(lc); |
748 | |
|
749 | 0 | if (index->amhasgettuple) |
750 | 0 | add_path(rel, (Path *) ipath); |
751 | |
|
752 | 0 | if (index->amhasgetbitmap && |
753 | 0 | (ipath->path.pathkeys == NIL || |
754 | 0 | ipath->indexselectivity < 1.0)) |
755 | 0 | *bitindexpaths = lappend(*bitindexpaths, ipath); |
756 | 0 | } |
757 | | |
758 | | /* |
759 | | * If there were ScalarArrayOpExpr clauses that the index can't handle |
760 | | * natively, generate bitmap scan paths relying on executor-managed |
761 | | * ScalarArrayOpExpr. |
762 | | */ |
763 | 0 | if (skip_nonnative_saop) |
764 | 0 | { |
765 | 0 | indexpaths = build_index_paths(root, rel, |
766 | 0 | index, clauses, |
767 | 0 | false, |
768 | 0 | ST_BITMAPSCAN, |
769 | 0 | NULL); |
770 | 0 | *bitindexpaths = list_concat(*bitindexpaths, indexpaths); |
771 | 0 | } |
772 | 0 | } |
773 | | |
774 | | /* |
775 | | * build_index_paths |
776 | | * Given an index and a set of index clauses for it, construct zero |
777 | | * or more IndexPaths. It also constructs zero or more partial IndexPaths. |
778 | | * |
779 | | * We return a list of paths because (1) this routine checks some cases |
780 | | * that should cause us to not generate any IndexPath, and (2) in some |
781 | | * cases we want to consider both a forward and a backward scan, so as |
782 | | * to obtain both sort orders. Note that the paths are just returned |
783 | | * to the caller and not immediately fed to add_path(). |
784 | | * |
785 | | * At top level, useful_predicate should be exactly the index's predOK flag |
786 | | * (ie, true if it has a predicate that was proven from the restriction |
787 | | * clauses). When working on an arm of an OR clause, useful_predicate |
788 | | * should be true if the predicate required the current OR list to be proven. |
789 | | * Note that this routine should never be called at all if the index has an |
790 | | * unprovable predicate. |
791 | | * |
792 | | * scantype indicates whether we want to create plain indexscans, bitmap |
793 | | * indexscans, or both. When it's ST_BITMAPSCAN, we will not consider |
794 | | * index ordering while deciding if a Path is worth generating. |
795 | | * |
796 | | * If skip_nonnative_saop is non-NULL, we ignore ScalarArrayOpExpr clauses |
797 | | * unless the index AM supports them directly, and we set *skip_nonnative_saop |
798 | | * to true if we found any such clauses (caller must initialize the variable |
799 | | * to false). If it's NULL, we do not ignore ScalarArrayOpExpr clauses. |
800 | | * |
801 | | * 'rel' is the index's heap relation |
802 | | * 'index' is the index for which we want to generate paths |
803 | | * 'clauses' is the collection of indexable clauses (IndexClause nodes) |
804 | | * 'useful_predicate' indicates whether the index has a useful predicate |
805 | | * 'scantype' indicates whether we need plain or bitmap scan support |
806 | | * 'skip_nonnative_saop' indicates whether to accept SAOP if index AM doesn't |
807 | | */ |
808 | | static List * |
809 | | build_index_paths(PlannerInfo *root, RelOptInfo *rel, |
810 | | IndexOptInfo *index, IndexClauseSet *clauses, |
811 | | bool useful_predicate, |
812 | | ScanTypeControl scantype, |
813 | | bool *skip_nonnative_saop) |
814 | 0 | { |
815 | 0 | List *result = NIL; |
816 | 0 | IndexPath *ipath; |
817 | 0 | List *index_clauses; |
818 | 0 | Relids outer_relids; |
819 | 0 | double loop_count; |
820 | 0 | List *orderbyclauses; |
821 | 0 | List *orderbyclausecols; |
822 | 0 | List *index_pathkeys; |
823 | 0 | List *useful_pathkeys; |
824 | 0 | bool pathkeys_possibly_useful; |
825 | 0 | bool index_is_ordered; |
826 | 0 | bool index_only_scan; |
827 | 0 | int indexcol; |
828 | |
|
829 | 0 | Assert(skip_nonnative_saop != NULL || scantype == ST_BITMAPSCAN); |
830 | | |
831 | | /* |
832 | | * Check that index supports the desired scan type(s) |
833 | | */ |
834 | 0 | switch (scantype) |
835 | 0 | { |
836 | 0 | case ST_INDEXSCAN: |
837 | 0 | if (!index->amhasgettuple) |
838 | 0 | return NIL; |
839 | 0 | break; |
840 | 0 | case ST_BITMAPSCAN: |
841 | 0 | if (!index->amhasgetbitmap) |
842 | 0 | return NIL; |
843 | 0 | break; |
844 | 0 | case ST_ANYSCAN: |
845 | | /* either or both are OK */ |
846 | 0 | break; |
847 | 0 | } |
848 | | |
849 | | /* |
850 | | * 1. Combine the per-column IndexClause lists into an overall list. |
851 | | * |
852 | | * In the resulting list, clauses are ordered by index key, so that the |
853 | | * column numbers form a nondecreasing sequence. (This order is depended |
854 | | * on by btree and possibly other places.) The list can be empty, if the |
855 | | * index AM allows that. |
856 | | * |
857 | | * We also build a Relids set showing which outer rels are required by the |
858 | | * selected clauses. Any lateral_relids are included in that, but not |
859 | | * otherwise accounted for. |
860 | | */ |
861 | 0 | index_clauses = NIL; |
862 | 0 | outer_relids = bms_copy(rel->lateral_relids); |
863 | 0 | for (indexcol = 0; indexcol < index->nkeycolumns; indexcol++) |
864 | 0 | { |
865 | 0 | ListCell *lc; |
866 | |
|
867 | 0 | foreach(lc, clauses->indexclauses[indexcol]) |
868 | 0 | { |
869 | 0 | IndexClause *iclause = (IndexClause *) lfirst(lc); |
870 | 0 | RestrictInfo *rinfo = iclause->rinfo; |
871 | |
|
872 | 0 | if (skip_nonnative_saop && !index->amsearcharray && |
873 | 0 | IsA(rinfo->clause, ScalarArrayOpExpr)) |
874 | 0 | { |
875 | | /* |
876 | | * Caller asked us to generate IndexPaths that omit any |
877 | | * ScalarArrayOpExpr clauses when the underlying index AM |
878 | | * lacks native support. |
879 | | * |
880 | | * We must omit this clause (and tell caller about it). |
881 | | */ |
882 | 0 | *skip_nonnative_saop = true; |
883 | 0 | continue; |
884 | 0 | } |
885 | | |
886 | | /* OK to include this clause */ |
887 | 0 | index_clauses = lappend(index_clauses, iclause); |
888 | 0 | outer_relids = bms_add_members(outer_relids, |
889 | 0 | rinfo->clause_relids); |
890 | 0 | } |
891 | | |
892 | | /* |
893 | | * If no clauses match the first index column, check for amoptionalkey |
894 | | * restriction. We can't generate a scan over an index with |
895 | | * amoptionalkey = false unless there's at least one index clause. |
896 | | * (When working on columns after the first, this test cannot fail. It |
897 | | * is always okay for columns after the first to not have any |
898 | | * clauses.) |
899 | | */ |
900 | 0 | if (index_clauses == NIL && !index->amoptionalkey) |
901 | 0 | return NIL; |
902 | 0 | } |
903 | | |
904 | | /* We do not want the index's rel itself listed in outer_relids */ |
905 | 0 | outer_relids = bms_del_member(outer_relids, rel->relid); |
906 | | |
907 | | /* Compute loop_count for cost estimation purposes */ |
908 | 0 | loop_count = get_loop_count(root, rel->relid, outer_relids); |
909 | | |
910 | | /* |
911 | | * 2. Compute pathkeys describing index's ordering, if any, then see how |
912 | | * many of them are actually useful for this query. This is not relevant |
913 | | * if we are only trying to build bitmap indexscans. |
914 | | */ |
915 | 0 | pathkeys_possibly_useful = (scantype != ST_BITMAPSCAN && |
916 | 0 | has_useful_pathkeys(root, rel)); |
917 | 0 | index_is_ordered = (index->sortopfamily != NULL); |
918 | 0 | if (index_is_ordered && pathkeys_possibly_useful) |
919 | 0 | { |
920 | 0 | index_pathkeys = build_index_pathkeys(root, index, |
921 | 0 | ForwardScanDirection); |
922 | 0 | useful_pathkeys = truncate_useless_pathkeys(root, rel, |
923 | 0 | index_pathkeys); |
924 | 0 | orderbyclauses = NIL; |
925 | 0 | orderbyclausecols = NIL; |
926 | 0 | } |
927 | 0 | else if (index->amcanorderbyop && pathkeys_possibly_useful) |
928 | 0 | { |
929 | | /* |
930 | | * See if we can generate ordering operators for query_pathkeys or at |
931 | | * least some prefix thereof. Matching to just a prefix of the |
932 | | * query_pathkeys will allow an incremental sort to be considered on |
933 | | * the index's partially sorted results. |
934 | | */ |
935 | 0 | match_pathkeys_to_index(index, root->query_pathkeys, |
936 | 0 | &orderbyclauses, |
937 | 0 | &orderbyclausecols); |
938 | 0 | if (list_length(root->query_pathkeys) == list_length(orderbyclauses)) |
939 | 0 | useful_pathkeys = root->query_pathkeys; |
940 | 0 | else |
941 | 0 | useful_pathkeys = list_copy_head(root->query_pathkeys, |
942 | 0 | list_length(orderbyclauses)); |
943 | 0 | } |
944 | 0 | else |
945 | 0 | { |
946 | 0 | useful_pathkeys = NIL; |
947 | 0 | orderbyclauses = NIL; |
948 | 0 | orderbyclausecols = NIL; |
949 | 0 | } |
950 | | |
951 | | /* |
952 | | * 3. Check if an index-only scan is possible. If we're not building |
953 | | * plain indexscans, this isn't relevant since bitmap scans don't support |
954 | | * index data retrieval anyway. |
955 | | */ |
956 | 0 | index_only_scan = (scantype != ST_BITMAPSCAN && |
957 | 0 | check_index_only(rel, index)); |
958 | | |
959 | | /* |
960 | | * 4. Generate an indexscan path if there are relevant restriction clauses |
961 | | * in the current clauses, OR the index ordering is potentially useful for |
962 | | * later merging or final output ordering, OR the index has a useful |
963 | | * predicate, OR an index-only scan is possible. |
964 | | */ |
965 | 0 | if (index_clauses != NIL || useful_pathkeys != NIL || useful_predicate || |
966 | 0 | index_only_scan) |
967 | 0 | { |
968 | 0 | ipath = create_index_path(root, index, |
969 | 0 | index_clauses, |
970 | 0 | orderbyclauses, |
971 | 0 | orderbyclausecols, |
972 | 0 | useful_pathkeys, |
973 | 0 | ForwardScanDirection, |
974 | 0 | index_only_scan, |
975 | 0 | outer_relids, |
976 | 0 | loop_count, |
977 | 0 | false); |
978 | 0 | result = lappend(result, ipath); |
979 | | |
980 | | /* |
981 | | * If appropriate, consider parallel index scan. We don't allow |
982 | | * parallel index scan for bitmap index scans. |
983 | | */ |
984 | 0 | if (index->amcanparallel && |
985 | 0 | rel->consider_parallel && outer_relids == NULL && |
986 | 0 | scantype != ST_BITMAPSCAN) |
987 | 0 | { |
988 | 0 | ipath = create_index_path(root, index, |
989 | 0 | index_clauses, |
990 | 0 | orderbyclauses, |
991 | 0 | orderbyclausecols, |
992 | 0 | useful_pathkeys, |
993 | 0 | ForwardScanDirection, |
994 | 0 | index_only_scan, |
995 | 0 | outer_relids, |
996 | 0 | loop_count, |
997 | 0 | true); |
998 | | |
999 | | /* |
1000 | | * if, after costing the path, we find that it's not worth using |
1001 | | * parallel workers, just free it. |
1002 | | */ |
1003 | 0 | if (ipath->path.parallel_workers > 0) |
1004 | 0 | add_partial_path(rel, (Path *) ipath); |
1005 | 0 | else |
1006 | 0 | pfree(ipath); |
1007 | 0 | } |
1008 | 0 | } |
1009 | | |
1010 | | /* |
1011 | | * 5. If the index is ordered, a backwards scan might be interesting. |
1012 | | */ |
1013 | 0 | if (index_is_ordered && pathkeys_possibly_useful) |
1014 | 0 | { |
1015 | 0 | index_pathkeys = build_index_pathkeys(root, index, |
1016 | 0 | BackwardScanDirection); |
1017 | 0 | useful_pathkeys = truncate_useless_pathkeys(root, rel, |
1018 | 0 | index_pathkeys); |
1019 | 0 | if (useful_pathkeys != NIL) |
1020 | 0 | { |
1021 | 0 | ipath = create_index_path(root, index, |
1022 | 0 | index_clauses, |
1023 | 0 | NIL, |
1024 | 0 | NIL, |
1025 | 0 | useful_pathkeys, |
1026 | 0 | BackwardScanDirection, |
1027 | 0 | index_only_scan, |
1028 | 0 | outer_relids, |
1029 | 0 | loop_count, |
1030 | 0 | false); |
1031 | 0 | result = lappend(result, ipath); |
1032 | | |
1033 | | /* If appropriate, consider parallel index scan */ |
1034 | 0 | if (index->amcanparallel && |
1035 | 0 | rel->consider_parallel && outer_relids == NULL && |
1036 | 0 | scantype != ST_BITMAPSCAN) |
1037 | 0 | { |
1038 | 0 | ipath = create_index_path(root, index, |
1039 | 0 | index_clauses, |
1040 | 0 | NIL, |
1041 | 0 | NIL, |
1042 | 0 | useful_pathkeys, |
1043 | 0 | BackwardScanDirection, |
1044 | 0 | index_only_scan, |
1045 | 0 | outer_relids, |
1046 | 0 | loop_count, |
1047 | 0 | true); |
1048 | | |
1049 | | /* |
1050 | | * if, after costing the path, we find that it's not worth |
1051 | | * using parallel workers, just free it. |
1052 | | */ |
1053 | 0 | if (ipath->path.parallel_workers > 0) |
1054 | 0 | add_partial_path(rel, (Path *) ipath); |
1055 | 0 | else |
1056 | 0 | pfree(ipath); |
1057 | 0 | } |
1058 | 0 | } |
1059 | 0 | } |
1060 | |
|
1061 | 0 | return result; |
1062 | 0 | } |
1063 | | |
1064 | | /* |
1065 | | * build_paths_for_OR |
1066 | | * Given a list of restriction clauses from one arm of an OR clause, |
1067 | | * construct all matching IndexPaths for the relation. |
1068 | | * |
1069 | | * Here we must scan all indexes of the relation, since a bitmap OR tree |
1070 | | * can use multiple indexes. |
1071 | | * |
1072 | | * The caller actually supplies two lists of restriction clauses: some |
1073 | | * "current" ones and some "other" ones. Both lists can be used freely |
1074 | | * to match keys of the index, but an index must use at least one of the |
1075 | | * "current" clauses to be considered usable. The motivation for this is |
1076 | | * examples like |
1077 | | * WHERE (x = 42) AND (... OR (y = 52 AND z = 77) OR ....) |
1078 | | * While we are considering the y/z subclause of the OR, we can use "x = 42" |
1079 | | * as one of the available index conditions; but we shouldn't match the |
1080 | | * subclause to any index on x alone, because such a Path would already have |
1081 | | * been generated at the upper level. So we could use an index on x,y,z |
1082 | | * or an index on x,y for the OR subclause, but not an index on just x. |
1083 | | * When dealing with a partial index, a match of the index predicate to |
1084 | | * one of the "current" clauses also makes the index usable. |
1085 | | * |
1086 | | * 'rel' is the relation for which we want to generate index paths |
1087 | | * 'clauses' is the current list of clauses (RestrictInfo nodes) |
1088 | | * 'other_clauses' is the list of additional upper-level clauses |
1089 | | */ |
1090 | | static List * |
1091 | | build_paths_for_OR(PlannerInfo *root, RelOptInfo *rel, |
1092 | | List *clauses, List *other_clauses) |
1093 | 0 | { |
1094 | 0 | List *result = NIL; |
1095 | 0 | List *all_clauses = NIL; /* not computed till needed */ |
1096 | 0 | ListCell *lc; |
1097 | |
|
1098 | 0 | foreach(lc, rel->indexlist) |
1099 | 0 | { |
1100 | 0 | IndexOptInfo *index = (IndexOptInfo *) lfirst(lc); |
1101 | 0 | IndexClauseSet clauseset; |
1102 | 0 | List *indexpaths; |
1103 | 0 | bool useful_predicate; |
1104 | | |
1105 | | /* Ignore index if it doesn't support bitmap scans */ |
1106 | 0 | if (!index->amhasgetbitmap) |
1107 | 0 | continue; |
1108 | | |
1109 | | /* |
1110 | | * Ignore partial indexes that do not match the query. If a partial |
1111 | | * index is marked predOK then we know it's OK. Otherwise, we have to |
1112 | | * test whether the added clauses are sufficient to imply the |
1113 | | * predicate. If so, we can use the index in the current context. |
1114 | | * |
1115 | | * We set useful_predicate to true iff the predicate was proven using |
1116 | | * the current set of clauses. This is needed to prevent matching a |
1117 | | * predOK index to an arm of an OR, which would be a legal but |
1118 | | * pointlessly inefficient plan. (A better plan will be generated by |
1119 | | * just scanning the predOK index alone, no OR.) |
1120 | | */ |
1121 | 0 | useful_predicate = false; |
1122 | 0 | if (index->indpred != NIL) |
1123 | 0 | { |
1124 | 0 | if (index->predOK) |
1125 | 0 | { |
1126 | | /* Usable, but don't set useful_predicate */ |
1127 | 0 | } |
1128 | 0 | else |
1129 | 0 | { |
1130 | | /* Form all_clauses if not done already */ |
1131 | 0 | if (all_clauses == NIL) |
1132 | 0 | all_clauses = list_concat_copy(clauses, other_clauses); |
1133 | |
|
1134 | 0 | if (!predicate_implied_by(index->indpred, all_clauses, false)) |
1135 | 0 | continue; /* can't use it at all */ |
1136 | | |
1137 | 0 | if (!predicate_implied_by(index->indpred, other_clauses, false)) |
1138 | 0 | useful_predicate = true; |
1139 | 0 | } |
1140 | 0 | } |
1141 | | |
1142 | | /* |
1143 | | * Identify the restriction clauses that can match the index. |
1144 | | */ |
1145 | 0 | MemSet(&clauseset, 0, sizeof(clauseset)); |
1146 | 0 | match_clauses_to_index(root, clauses, index, &clauseset); |
1147 | | |
1148 | | /* |
1149 | | * If no matches so far, and the index predicate isn't useful, we |
1150 | | * don't want it. |
1151 | | */ |
1152 | 0 | if (!clauseset.nonempty && !useful_predicate) |
1153 | 0 | continue; |
1154 | | |
1155 | | /* |
1156 | | * Add "other" restriction clauses to the clauseset. |
1157 | | */ |
1158 | 0 | match_clauses_to_index(root, other_clauses, index, &clauseset); |
1159 | | |
1160 | | /* |
1161 | | * Construct paths if possible. |
1162 | | */ |
1163 | 0 | indexpaths = build_index_paths(root, rel, |
1164 | 0 | index, &clauseset, |
1165 | 0 | useful_predicate, |
1166 | 0 | ST_BITMAPSCAN, |
1167 | 0 | NULL); |
1168 | 0 | result = list_concat(result, indexpaths); |
1169 | 0 | } |
1170 | |
|
1171 | 0 | return result; |
1172 | 0 | } |
1173 | | |
1174 | | /* |
1175 | | * Utility structure used to group similar OR-clause arguments in |
1176 | | * group_similar_or_args(). It represents information about the OR-clause |
1177 | | * argument and its matching index key. |
1178 | | */ |
1179 | | typedef struct |
1180 | | { |
1181 | | int indexnum; /* index of the matching index, or -1 if no |
1182 | | * matching index */ |
1183 | | int colnum; /* index of the matching column, or -1 if no |
1184 | | * matching index */ |
1185 | | Oid opno; /* OID of the OpClause operator, or InvalidOid |
1186 | | * if not an OpExpr */ |
1187 | | Oid inputcollid; /* OID of the OpClause input collation */ |
1188 | | int argindex; /* index of the clause in the list of |
1189 | | * arguments */ |
1190 | | int groupindex; /* value of argindex for the fist clause in |
1191 | | * the group of similar clauses */ |
1192 | | } OrArgIndexMatch; |
1193 | | |
1194 | | /* |
1195 | | * Comparison function for OrArgIndexMatch which provides sort order placing |
1196 | | * similar OR-clause arguments together. |
1197 | | */ |
1198 | | static int |
1199 | | or_arg_index_match_cmp(const void *a, const void *b) |
1200 | 0 | { |
1201 | 0 | const OrArgIndexMatch *match_a = (const OrArgIndexMatch *) a; |
1202 | 0 | const OrArgIndexMatch *match_b = (const OrArgIndexMatch *) b; |
1203 | |
|
1204 | 0 | if (match_a->indexnum < match_b->indexnum) |
1205 | 0 | return -1; |
1206 | 0 | else if (match_a->indexnum > match_b->indexnum) |
1207 | 0 | return 1; |
1208 | | |
1209 | 0 | if (match_a->colnum < match_b->colnum) |
1210 | 0 | return -1; |
1211 | 0 | else if (match_a->colnum > match_b->colnum) |
1212 | 0 | return 1; |
1213 | | |
1214 | 0 | if (match_a->opno < match_b->opno) |
1215 | 0 | return -1; |
1216 | 0 | else if (match_a->opno > match_b->opno) |
1217 | 0 | return 1; |
1218 | | |
1219 | 0 | if (match_a->inputcollid < match_b->inputcollid) |
1220 | 0 | return -1; |
1221 | 0 | else if (match_a->inputcollid > match_b->inputcollid) |
1222 | 0 | return 1; |
1223 | | |
1224 | 0 | if (match_a->argindex < match_b->argindex) |
1225 | 0 | return -1; |
1226 | 0 | else if (match_a->argindex > match_b->argindex) |
1227 | 0 | return 1; |
1228 | | |
1229 | 0 | return 0; |
1230 | 0 | } |
1231 | | |
1232 | | /* |
1233 | | * Another comparison function for OrArgIndexMatch. It sorts groups together |
1234 | | * using groupindex. The group items are then sorted by argindex. |
1235 | | */ |
1236 | | static int |
1237 | | or_arg_index_match_cmp_group(const void *a, const void *b) |
1238 | 0 | { |
1239 | 0 | const OrArgIndexMatch *match_a = (const OrArgIndexMatch *) a; |
1240 | 0 | const OrArgIndexMatch *match_b = (const OrArgIndexMatch *) b; |
1241 | |
|
1242 | 0 | if (match_a->groupindex < match_b->groupindex) |
1243 | 0 | return -1; |
1244 | 0 | else if (match_a->groupindex > match_b->groupindex) |
1245 | 0 | return 1; |
1246 | | |
1247 | 0 | if (match_a->argindex < match_b->argindex) |
1248 | 0 | return -1; |
1249 | 0 | else if (match_a->argindex > match_b->argindex) |
1250 | 0 | return 1; |
1251 | | |
1252 | 0 | return 0; |
1253 | 0 | } |
1254 | | |
1255 | | /* |
1256 | | * group_similar_or_args |
1257 | | * Transform incoming OR-restrictinfo into a list of sub-restrictinfos, |
1258 | | * each of them containing a subset of similar OR-clause arguments from |
1259 | | * the source rinfo. |
1260 | | * |
1261 | | * Similar OR-clause arguments are of the form "indexkey op constant" having |
1262 | | * the same indexkey, operator, and collation. Constant may comprise either |
1263 | | * Const or Param. It may be employed later, during the |
1264 | | * match_clause_to_indexcol() to transform the whole OR-sub-rinfo to an SAOP |
1265 | | * clause. |
1266 | | * |
1267 | | * Returns the processed list of OR-clause arguments. |
1268 | | */ |
1269 | | static List * |
1270 | | group_similar_or_args(PlannerInfo *root, RelOptInfo *rel, RestrictInfo *rinfo) |
1271 | 0 | { |
1272 | 0 | int n; |
1273 | 0 | int i; |
1274 | 0 | int group_start; |
1275 | 0 | OrArgIndexMatch *matches; |
1276 | 0 | bool matched = false; |
1277 | 0 | ListCell *lc; |
1278 | 0 | ListCell *lc2; |
1279 | 0 | List *orargs; |
1280 | 0 | List *result = NIL; |
1281 | 0 | Index relid = rel->relid; |
1282 | |
|
1283 | 0 | Assert(IsA(rinfo->orclause, BoolExpr)); |
1284 | 0 | orargs = ((BoolExpr *) rinfo->orclause)->args; |
1285 | 0 | n = list_length(orargs); |
1286 | | |
1287 | | /* |
1288 | | * To avoid N^2 behavior, take utility pass along the list of OR-clause |
1289 | | * arguments. For each argument, fill the OrArgIndexMatch structure, |
1290 | | * which will be used to sort these arguments at the next step. |
1291 | | */ |
1292 | 0 | i = -1; |
1293 | 0 | matches = (OrArgIndexMatch *) palloc(sizeof(OrArgIndexMatch) * n); |
1294 | 0 | foreach(lc, orargs) |
1295 | 0 | { |
1296 | 0 | Node *arg = lfirst(lc); |
1297 | 0 | RestrictInfo *argrinfo; |
1298 | 0 | OpExpr *clause; |
1299 | 0 | Oid opno; |
1300 | 0 | Node *leftop, |
1301 | 0 | *rightop; |
1302 | 0 | Node *nonConstExpr; |
1303 | 0 | int indexnum; |
1304 | 0 | int colnum; |
1305 | |
|
1306 | 0 | i++; |
1307 | 0 | matches[i].argindex = i; |
1308 | 0 | matches[i].groupindex = i; |
1309 | 0 | matches[i].indexnum = -1; |
1310 | 0 | matches[i].colnum = -1; |
1311 | 0 | matches[i].opno = InvalidOid; |
1312 | 0 | matches[i].inputcollid = InvalidOid; |
1313 | |
|
1314 | 0 | if (!IsA(arg, RestrictInfo)) |
1315 | 0 | continue; |
1316 | | |
1317 | 0 | argrinfo = castNode(RestrictInfo, arg); |
1318 | | |
1319 | | /* Only operator clauses can match */ |
1320 | 0 | if (!IsA(argrinfo->clause, OpExpr)) |
1321 | 0 | continue; |
1322 | | |
1323 | 0 | clause = (OpExpr *) argrinfo->clause; |
1324 | 0 | opno = clause->opno; |
1325 | | |
1326 | | /* Only binary operators can match */ |
1327 | 0 | if (list_length(clause->args) != 2) |
1328 | 0 | continue; |
1329 | | |
1330 | | /* |
1331 | | * Ignore any RelabelType node above the operands. This is needed to |
1332 | | * be able to apply indexscanning in binary-compatible-operator cases. |
1333 | | * Note: we can assume there is at most one RelabelType node; |
1334 | | * eval_const_expressions() will have simplified if more than one. |
1335 | | */ |
1336 | 0 | leftop = get_leftop(clause); |
1337 | 0 | if (IsA(leftop, RelabelType)) |
1338 | 0 | leftop = (Node *) ((RelabelType *) leftop)->arg; |
1339 | |
|
1340 | 0 | rightop = get_rightop(clause); |
1341 | 0 | if (IsA(rightop, RelabelType)) |
1342 | 0 | rightop = (Node *) ((RelabelType *) rightop)->arg; |
1343 | | |
1344 | | /* |
1345 | | * Check for clauses of the form: (indexkey operator constant) or |
1346 | | * (constant operator indexkey). But we don't know a particular index |
1347 | | * yet. Therefore, we try to distinguish the potential index key and |
1348 | | * constant first, then search for a matching index key among all |
1349 | | * indexes. |
1350 | | */ |
1351 | 0 | if (bms_is_member(relid, argrinfo->right_relids) && |
1352 | 0 | !bms_is_member(relid, argrinfo->left_relids) && |
1353 | 0 | !contain_volatile_functions(leftop)) |
1354 | 0 | { |
1355 | 0 | opno = get_commutator(opno); |
1356 | |
|
1357 | 0 | if (!OidIsValid(opno)) |
1358 | 0 | { |
1359 | | /* commutator doesn't exist, we can't reverse the order */ |
1360 | 0 | continue; |
1361 | 0 | } |
1362 | 0 | nonConstExpr = rightop; |
1363 | 0 | } |
1364 | 0 | else if (bms_is_member(relid, argrinfo->left_relids) && |
1365 | 0 | !bms_is_member(relid, argrinfo->right_relids) && |
1366 | 0 | !contain_volatile_functions(rightop)) |
1367 | 0 | { |
1368 | 0 | nonConstExpr = leftop; |
1369 | 0 | } |
1370 | 0 | else |
1371 | 0 | { |
1372 | 0 | continue; |
1373 | 0 | } |
1374 | | |
1375 | | /* |
1376 | | * Match non-constant part to the index key. It's possible that a |
1377 | | * single non-constant part matches multiple index keys. It's OK, we |
1378 | | * just stop with first matching index key. Given that this choice is |
1379 | | * determined the same for every clause, we will group similar clauses |
1380 | | * together anyway. |
1381 | | */ |
1382 | 0 | indexnum = 0; |
1383 | 0 | foreach(lc2, rel->indexlist) |
1384 | 0 | { |
1385 | 0 | IndexOptInfo *index = (IndexOptInfo *) lfirst(lc2); |
1386 | | |
1387 | | /* |
1388 | | * Ignore index if it doesn't support bitmap scans or SAOP |
1389 | | * clauses. |
1390 | | */ |
1391 | 0 | if (!index->amhasgetbitmap || !index->amsearcharray) |
1392 | 0 | continue; |
1393 | | |
1394 | 0 | for (colnum = 0; colnum < index->nkeycolumns; colnum++) |
1395 | 0 | { |
1396 | 0 | if (match_index_to_operand(nonConstExpr, colnum, index)) |
1397 | 0 | { |
1398 | 0 | matches[i].indexnum = indexnum; |
1399 | 0 | matches[i].colnum = colnum; |
1400 | 0 | matches[i].opno = opno; |
1401 | 0 | matches[i].inputcollid = clause->inputcollid; |
1402 | 0 | matched = true; |
1403 | 0 | break; |
1404 | 0 | } |
1405 | 0 | } |
1406 | | |
1407 | | /* |
1408 | | * Stop looping through the indexes, if we managed to match |
1409 | | * nonConstExpr to any index column. |
1410 | | */ |
1411 | 0 | if (matches[i].indexnum >= 0) |
1412 | 0 | break; |
1413 | 0 | indexnum++; |
1414 | 0 | } |
1415 | 0 | } |
1416 | | |
1417 | | /* |
1418 | | * Fast-path check: if no clause is matching to the index column, we can |
1419 | | * just give up at this stage and return the clause list as-is. |
1420 | | */ |
1421 | 0 | if (!matched) |
1422 | 0 | { |
1423 | 0 | pfree(matches); |
1424 | 0 | return orargs; |
1425 | 0 | } |
1426 | | |
1427 | | /* |
1428 | | * Sort clauses to make similar clauses go together. But at the same |
1429 | | * time, we would like to change the order of clauses as little as |
1430 | | * possible. To do so, we reorder each group of similar clauses so that |
1431 | | * the first item of the group stays in place, and all the other items are |
1432 | | * moved after it. So, if there are no similar clauses, the order of |
1433 | | * clauses stays the same. When there are some groups, required |
1434 | | * reordering happens while the rest of the clauses remain in their |
1435 | | * places. That is achieved by assigning a 'groupindex' to each clause: |
1436 | | * the number of the first item in the group in the original clause list. |
1437 | | */ |
1438 | 0 | qsort(matches, n, sizeof(OrArgIndexMatch), or_arg_index_match_cmp); |
1439 | | |
1440 | | /* Assign groupindex to the sorted clauses */ |
1441 | 0 | for (i = 1; i < n; i++) |
1442 | 0 | { |
1443 | | /* |
1444 | | * When two clauses are similar and should belong to the same group, |
1445 | | * copy the 'groupindex' from the previous clause. Given we are |
1446 | | * considering clauses in direct order, all the clauses would have a |
1447 | | * 'groupindex' equal to the 'groupindex' of the first clause in the |
1448 | | * group. |
1449 | | */ |
1450 | 0 | if (matches[i].indexnum == matches[i - 1].indexnum && |
1451 | 0 | matches[i].colnum == matches[i - 1].colnum && |
1452 | 0 | matches[i].opno == matches[i - 1].opno && |
1453 | 0 | matches[i].inputcollid == matches[i - 1].inputcollid && |
1454 | 0 | matches[i].indexnum != -1) |
1455 | 0 | matches[i].groupindex = matches[i - 1].groupindex; |
1456 | 0 | } |
1457 | | |
1458 | | /* Re-sort clauses first by groupindex then by argindex */ |
1459 | 0 | qsort(matches, n, sizeof(OrArgIndexMatch), or_arg_index_match_cmp_group); |
1460 | | |
1461 | | /* |
1462 | | * Group similar clauses into single sub-restrictinfo. Side effect: the |
1463 | | * resulting list of restrictions will be sorted by indexnum and colnum. |
1464 | | */ |
1465 | 0 | group_start = 0; |
1466 | 0 | for (i = 1; i <= n; i++) |
1467 | 0 | { |
1468 | | /* Check if it's a group boundary */ |
1469 | 0 | if (group_start >= 0 && |
1470 | 0 | (i == n || |
1471 | 0 | matches[i].indexnum != matches[group_start].indexnum || |
1472 | 0 | matches[i].colnum != matches[group_start].colnum || |
1473 | 0 | matches[i].opno != matches[group_start].opno || |
1474 | 0 | matches[i].inputcollid != matches[group_start].inputcollid || |
1475 | 0 | matches[i].indexnum == -1)) |
1476 | 0 | { |
1477 | | /* |
1478 | | * One clause in group: add it "as is" to the upper-level OR. |
1479 | | */ |
1480 | 0 | if (i - group_start == 1) |
1481 | 0 | { |
1482 | 0 | result = lappend(result, |
1483 | 0 | list_nth(orargs, |
1484 | 0 | matches[group_start].argindex)); |
1485 | 0 | } |
1486 | 0 | else |
1487 | 0 | { |
1488 | | /* |
1489 | | * Two or more clauses in a group: create a nested OR. |
1490 | | */ |
1491 | 0 | List *args = NIL; |
1492 | 0 | List *rargs = NIL; |
1493 | 0 | RestrictInfo *subrinfo; |
1494 | 0 | int j; |
1495 | |
|
1496 | 0 | Assert(i - group_start >= 2); |
1497 | | |
1498 | | /* Construct the list of nested OR arguments */ |
1499 | 0 | for (j = group_start; j < i; j++) |
1500 | 0 | { |
1501 | 0 | Node *arg = list_nth(orargs, matches[j].argindex); |
1502 | |
|
1503 | 0 | rargs = lappend(rargs, arg); |
1504 | 0 | if (IsA(arg, RestrictInfo)) |
1505 | 0 | args = lappend(args, ((RestrictInfo *) arg)->clause); |
1506 | 0 | else |
1507 | 0 | args = lappend(args, arg); |
1508 | 0 | } |
1509 | | |
1510 | | /* Construct the nested OR and wrap it with RestrictInfo */ |
1511 | 0 | subrinfo = make_plain_restrictinfo(root, |
1512 | 0 | make_orclause(args), |
1513 | 0 | make_orclause(rargs), |
1514 | 0 | rinfo->is_pushed_down, |
1515 | 0 | rinfo->has_clone, |
1516 | 0 | rinfo->is_clone, |
1517 | 0 | rinfo->pseudoconstant, |
1518 | 0 | rinfo->security_level, |
1519 | 0 | rinfo->required_relids, |
1520 | 0 | rinfo->incompatible_relids, |
1521 | 0 | rinfo->outer_relids); |
1522 | 0 | result = lappend(result, subrinfo); |
1523 | 0 | } |
1524 | |
|
1525 | 0 | group_start = i; |
1526 | 0 | } |
1527 | 0 | } |
1528 | 0 | pfree(matches); |
1529 | 0 | return result; |
1530 | 0 | } |
1531 | | |
1532 | | /* |
1533 | | * make_bitmap_paths_for_or_group |
1534 | | * Generate bitmap paths for a group of similar OR-clause arguments |
1535 | | * produced by group_similar_or_args(). |
1536 | | * |
1537 | | * This function considers two cases: (1) matching a group of clauses to |
1538 | | * the index as a whole, and (2) matching the individual clauses one-by-one. |
1539 | | * (1) typically comprises an optimal solution. If not, (2) typically |
1540 | | * comprises fair alternative. |
1541 | | * |
1542 | | * Ideally, we could consider all arbitrary splits of arguments into |
1543 | | * subgroups, but that could lead to unacceptable computational complexity. |
1544 | | * This is why we only consider two cases of above. |
1545 | | */ |
1546 | | static List * |
1547 | | make_bitmap_paths_for_or_group(PlannerInfo *root, RelOptInfo *rel, |
1548 | | RestrictInfo *ri, List *other_clauses) |
1549 | 0 | { |
1550 | 0 | List *jointlist = NIL; |
1551 | 0 | List *splitlist = NIL; |
1552 | 0 | ListCell *lc; |
1553 | 0 | List *orargs; |
1554 | 0 | List *args = ((BoolExpr *) ri->orclause)->args; |
1555 | 0 | Cost jointcost = 0.0, |
1556 | 0 | splitcost = 0.0; |
1557 | 0 | Path *bitmapqual; |
1558 | 0 | List *indlist; |
1559 | | |
1560 | | /* |
1561 | | * First, try to match the whole group to the one index. |
1562 | | */ |
1563 | 0 | orargs = list_make1(ri); |
1564 | 0 | indlist = build_paths_for_OR(root, rel, |
1565 | 0 | orargs, |
1566 | 0 | other_clauses); |
1567 | 0 | if (indlist != NIL) |
1568 | 0 | { |
1569 | 0 | bitmapqual = choose_bitmap_and(root, rel, indlist); |
1570 | 0 | jointcost = bitmapqual->total_cost; |
1571 | 0 | jointlist = list_make1(bitmapqual); |
1572 | 0 | } |
1573 | | |
1574 | | /* |
1575 | | * If we manage to find a bitmap scan, which uses the group of OR-clause |
1576 | | * arguments as a whole, we can skip matching OR-clause arguments |
1577 | | * one-by-one as long as there are no other clauses, which can bring more |
1578 | | * efficiency to one-by-one case. |
1579 | | */ |
1580 | 0 | if (jointlist != NIL && other_clauses == NIL) |
1581 | 0 | return jointlist; |
1582 | | |
1583 | | /* |
1584 | | * Also try to match all containing clauses one-by-one. |
1585 | | */ |
1586 | 0 | foreach(lc, args) |
1587 | 0 | { |
1588 | 0 | orargs = list_make1(lfirst(lc)); |
1589 | |
|
1590 | 0 | indlist = build_paths_for_OR(root, rel, |
1591 | 0 | orargs, |
1592 | 0 | other_clauses); |
1593 | |
|
1594 | 0 | if (indlist == NIL) |
1595 | 0 | { |
1596 | 0 | splitlist = NIL; |
1597 | 0 | break; |
1598 | 0 | } |
1599 | | |
1600 | 0 | bitmapqual = choose_bitmap_and(root, rel, indlist); |
1601 | 0 | splitcost += bitmapqual->total_cost; |
1602 | 0 | splitlist = lappend(splitlist, bitmapqual); |
1603 | 0 | } |
1604 | | |
1605 | | /* |
1606 | | * Pick the best option. |
1607 | | */ |
1608 | 0 | if (splitlist == NIL) |
1609 | 0 | return jointlist; |
1610 | 0 | else if (jointlist == NIL) |
1611 | 0 | return splitlist; |
1612 | 0 | else |
1613 | 0 | return (jointcost < splitcost) ? jointlist : splitlist; |
1614 | 0 | } |
1615 | | |
1616 | | |
1617 | | /* |
1618 | | * generate_bitmap_or_paths |
1619 | | * Look through the list of clauses to find OR clauses, and generate |
1620 | | * a BitmapOrPath for each one we can handle that way. Return a list |
1621 | | * of the generated BitmapOrPaths. |
1622 | | * |
1623 | | * other_clauses is a list of additional clauses that can be assumed true |
1624 | | * for the purpose of generating indexquals, but are not to be searched for |
1625 | | * ORs. (See build_paths_for_OR() for motivation.) |
1626 | | */ |
1627 | | static List * |
1628 | | generate_bitmap_or_paths(PlannerInfo *root, RelOptInfo *rel, |
1629 | | List *clauses, List *other_clauses) |
1630 | 0 | { |
1631 | 0 | List *result = NIL; |
1632 | 0 | List *all_clauses; |
1633 | 0 | ListCell *lc; |
1634 | | |
1635 | | /* |
1636 | | * We can use both the current and other clauses as context for |
1637 | | * build_paths_for_OR; no need to remove ORs from the lists. |
1638 | | */ |
1639 | 0 | all_clauses = list_concat_copy(clauses, other_clauses); |
1640 | |
|
1641 | 0 | foreach(lc, clauses) |
1642 | 0 | { |
1643 | 0 | RestrictInfo *rinfo = lfirst_node(RestrictInfo, lc); |
1644 | 0 | List *pathlist; |
1645 | 0 | Path *bitmapqual; |
1646 | 0 | ListCell *j; |
1647 | 0 | List *groupedArgs; |
1648 | 0 | List *inner_other_clauses = NIL; |
1649 | | |
1650 | | /* Ignore RestrictInfos that aren't ORs */ |
1651 | 0 | if (!restriction_is_or_clause(rinfo)) |
1652 | 0 | continue; |
1653 | | |
1654 | | /* |
1655 | | * We must be able to match at least one index to each of the arms of |
1656 | | * the OR, else we can't use it. |
1657 | | */ |
1658 | 0 | pathlist = NIL; |
1659 | | |
1660 | | /* |
1661 | | * Group the similar OR-clause arguments into dedicated RestrictInfos, |
1662 | | * because each of those RestrictInfos has a chance to match the index |
1663 | | * as a whole. |
1664 | | */ |
1665 | 0 | groupedArgs = group_similar_or_args(root, rel, rinfo); |
1666 | |
|
1667 | 0 | if (groupedArgs != ((BoolExpr *) rinfo->orclause)->args) |
1668 | 0 | { |
1669 | | /* |
1670 | | * Some parts of the rinfo were probably grouped. In this case, |
1671 | | * we have a set of sub-rinfos that together are an exact |
1672 | | * duplicate of rinfo. Thus, we need to remove the rinfo from |
1673 | | * other clauses. match_clauses_to_index detects duplicated |
1674 | | * iclauses by comparing pointers to original rinfos that would be |
1675 | | * different. So, we must delete rinfo to avoid de-facto |
1676 | | * duplicated clauses in the index clauses list. |
1677 | | */ |
1678 | 0 | inner_other_clauses = list_delete(list_copy(all_clauses), rinfo); |
1679 | 0 | } |
1680 | |
|
1681 | 0 | foreach(j, groupedArgs) |
1682 | 0 | { |
1683 | 0 | Node *orarg = (Node *) lfirst(j); |
1684 | 0 | List *indlist; |
1685 | | |
1686 | | /* OR arguments should be ANDs or sub-RestrictInfos */ |
1687 | 0 | if (is_andclause(orarg)) |
1688 | 0 | { |
1689 | 0 | List *andargs = ((BoolExpr *) orarg)->args; |
1690 | |
|
1691 | 0 | indlist = build_paths_for_OR(root, rel, |
1692 | 0 | andargs, |
1693 | 0 | all_clauses); |
1694 | | |
1695 | | /* Recurse in case there are sub-ORs */ |
1696 | 0 | indlist = list_concat(indlist, |
1697 | 0 | generate_bitmap_or_paths(root, rel, |
1698 | 0 | andargs, |
1699 | 0 | all_clauses)); |
1700 | 0 | } |
1701 | 0 | else if (restriction_is_or_clause(castNode(RestrictInfo, orarg))) |
1702 | 0 | { |
1703 | 0 | RestrictInfo *ri = castNode(RestrictInfo, orarg); |
1704 | | |
1705 | | /* |
1706 | | * Generate bitmap paths for the group of similar OR-clause |
1707 | | * arguments. |
1708 | | */ |
1709 | 0 | indlist = make_bitmap_paths_for_or_group(root, |
1710 | 0 | rel, ri, |
1711 | 0 | inner_other_clauses); |
1712 | |
|
1713 | 0 | if (indlist == NIL) |
1714 | 0 | { |
1715 | 0 | pathlist = NIL; |
1716 | 0 | break; |
1717 | 0 | } |
1718 | 0 | else |
1719 | 0 | { |
1720 | 0 | pathlist = list_concat(pathlist, indlist); |
1721 | 0 | continue; |
1722 | 0 | } |
1723 | 0 | } |
1724 | 0 | else |
1725 | 0 | { |
1726 | 0 | RestrictInfo *ri = castNode(RestrictInfo, orarg); |
1727 | 0 | List *orargs; |
1728 | |
|
1729 | 0 | orargs = list_make1(ri); |
1730 | |
|
1731 | 0 | indlist = build_paths_for_OR(root, rel, |
1732 | 0 | orargs, |
1733 | 0 | all_clauses); |
1734 | 0 | } |
1735 | | |
1736 | | /* |
1737 | | * If nothing matched this arm, we can't do anything with this OR |
1738 | | * clause. |
1739 | | */ |
1740 | 0 | if (indlist == NIL) |
1741 | 0 | { |
1742 | 0 | pathlist = NIL; |
1743 | 0 | break; |
1744 | 0 | } |
1745 | | |
1746 | | /* |
1747 | | * OK, pick the most promising AND combination, and add it to |
1748 | | * pathlist. |
1749 | | */ |
1750 | 0 | bitmapqual = choose_bitmap_and(root, rel, indlist); |
1751 | 0 | pathlist = lappend(pathlist, bitmapqual); |
1752 | 0 | } |
1753 | |
|
1754 | 0 | if (inner_other_clauses != NIL) |
1755 | 0 | list_free(inner_other_clauses); |
1756 | | |
1757 | | /* |
1758 | | * If we have a match for every arm, then turn them into a |
1759 | | * BitmapOrPath, and add to result list. |
1760 | | */ |
1761 | 0 | if (pathlist != NIL) |
1762 | 0 | { |
1763 | 0 | bitmapqual = (Path *) create_bitmap_or_path(root, rel, pathlist); |
1764 | 0 | result = lappend(result, bitmapqual); |
1765 | 0 | } |
1766 | 0 | } |
1767 | |
|
1768 | 0 | return result; |
1769 | 0 | } |
1770 | | |
1771 | | |
1772 | | /* |
1773 | | * choose_bitmap_and |
1774 | | * Given a nonempty list of bitmap paths, AND them into one path. |
1775 | | * |
1776 | | * This is a nontrivial decision since we can legally use any subset of the |
1777 | | * given path set. We want to choose a good tradeoff between selectivity |
1778 | | * and cost of computing the bitmap. |
1779 | | * |
1780 | | * The result is either a single one of the inputs, or a BitmapAndPath |
1781 | | * combining multiple inputs. |
1782 | | */ |
1783 | | static Path * |
1784 | | choose_bitmap_and(PlannerInfo *root, RelOptInfo *rel, List *paths) |
1785 | 0 | { |
1786 | 0 | int npaths = list_length(paths); |
1787 | 0 | PathClauseUsage **pathinfoarray; |
1788 | 0 | PathClauseUsage *pathinfo; |
1789 | 0 | List *clauselist; |
1790 | 0 | List *bestpaths = NIL; |
1791 | 0 | Cost bestcost = 0; |
1792 | 0 | int i, |
1793 | 0 | j; |
1794 | 0 | ListCell *l; |
1795 | |
|
1796 | 0 | Assert(npaths > 0); /* else caller error */ |
1797 | 0 | if (npaths == 1) |
1798 | 0 | return (Path *) linitial(paths); /* easy case */ |
1799 | | |
1800 | | /* |
1801 | | * In theory we should consider every nonempty subset of the given paths. |
1802 | | * In practice that seems like overkill, given the crude nature of the |
1803 | | * estimates, not to mention the possible effects of higher-level AND and |
1804 | | * OR clauses. Moreover, it's completely impractical if there are a large |
1805 | | * number of paths, since the work would grow as O(2^N). |
1806 | | * |
1807 | | * As a heuristic, we first check for paths using exactly the same sets of |
1808 | | * WHERE clauses + index predicate conditions, and reject all but the |
1809 | | * cheapest-to-scan in any such group. This primarily gets rid of indexes |
1810 | | * that include the interesting columns but also irrelevant columns. (In |
1811 | | * situations where the DBA has gone overboard on creating variant |
1812 | | * indexes, this can make for a very large reduction in the number of |
1813 | | * paths considered further.) |
1814 | | * |
1815 | | * We then sort the surviving paths with the cheapest-to-scan first, and |
1816 | | * for each path, consider using that path alone as the basis for a bitmap |
1817 | | * scan. Then we consider bitmap AND scans formed from that path plus |
1818 | | * each subsequent (higher-cost) path, adding on a subsequent path if it |
1819 | | * results in a reduction in the estimated total scan cost. This means we |
1820 | | * consider about O(N^2) rather than O(2^N) path combinations, which is |
1821 | | * quite tolerable, especially given than N is usually reasonably small |
1822 | | * because of the prefiltering step. The cheapest of these is returned. |
1823 | | * |
1824 | | * We will only consider AND combinations in which no two indexes use the |
1825 | | * same WHERE clause. This is a bit of a kluge: it's needed because |
1826 | | * costsize.c and clausesel.c aren't very smart about redundant clauses. |
1827 | | * They will usually double-count the redundant clauses, producing a |
1828 | | * too-small selectivity that makes a redundant AND step look like it |
1829 | | * reduces the total cost. Perhaps someday that code will be smarter and |
1830 | | * we can remove this limitation. (But note that this also defends |
1831 | | * against flat-out duplicate input paths, which can happen because |
1832 | | * match_join_clauses_to_index will find the same OR join clauses that |
1833 | | * extract_restriction_or_clauses has pulled OR restriction clauses out |
1834 | | * of.) |
1835 | | * |
1836 | | * For the same reason, we reject AND combinations in which an index |
1837 | | * predicate clause duplicates another clause. Here we find it necessary |
1838 | | * to be even stricter: we'll reject a partial index if any of its |
1839 | | * predicate clauses are implied by the set of WHERE clauses and predicate |
1840 | | * clauses used so far. This covers cases such as a condition "x = 42" |
1841 | | * used with a plain index, followed by a clauseless scan of a partial |
1842 | | * index "WHERE x >= 40 AND x < 50". The partial index has been accepted |
1843 | | * only because "x = 42" was present, and so allowing it would partially |
1844 | | * double-count selectivity. (We could use predicate_implied_by on |
1845 | | * regular qual clauses too, to have a more intelligent, but much more |
1846 | | * expensive, check for redundancy --- but in most cases simple equality |
1847 | | * seems to suffice.) |
1848 | | */ |
1849 | | |
1850 | | /* |
1851 | | * Extract clause usage info and detect any paths that use exactly the |
1852 | | * same set of clauses; keep only the cheapest-to-scan of any such groups. |
1853 | | * The surviving paths are put into an array for qsort'ing. |
1854 | | */ |
1855 | 0 | pathinfoarray = (PathClauseUsage **) |
1856 | 0 | palloc(npaths * sizeof(PathClauseUsage *)); |
1857 | 0 | clauselist = NIL; |
1858 | 0 | npaths = 0; |
1859 | 0 | foreach(l, paths) |
1860 | 0 | { |
1861 | 0 | Path *ipath = (Path *) lfirst(l); |
1862 | |
|
1863 | 0 | pathinfo = classify_index_clause_usage(ipath, &clauselist); |
1864 | | |
1865 | | /* If it's unclassifiable, treat it as distinct from all others */ |
1866 | 0 | if (pathinfo->unclassifiable) |
1867 | 0 | { |
1868 | 0 | pathinfoarray[npaths++] = pathinfo; |
1869 | 0 | continue; |
1870 | 0 | } |
1871 | | |
1872 | 0 | for (i = 0; i < npaths; i++) |
1873 | 0 | { |
1874 | 0 | if (!pathinfoarray[i]->unclassifiable && |
1875 | 0 | bms_equal(pathinfo->clauseids, pathinfoarray[i]->clauseids)) |
1876 | 0 | break; |
1877 | 0 | } |
1878 | 0 | if (i < npaths) |
1879 | 0 | { |
1880 | | /* duplicate clauseids, keep the cheaper one */ |
1881 | 0 | Cost ncost; |
1882 | 0 | Cost ocost; |
1883 | 0 | Selectivity nselec; |
1884 | 0 | Selectivity oselec; |
1885 | |
|
1886 | 0 | cost_bitmap_tree_node(pathinfo->path, &ncost, &nselec); |
1887 | 0 | cost_bitmap_tree_node(pathinfoarray[i]->path, &ocost, &oselec); |
1888 | 0 | if (ncost < ocost) |
1889 | 0 | pathinfoarray[i] = pathinfo; |
1890 | 0 | } |
1891 | 0 | else |
1892 | 0 | { |
1893 | | /* not duplicate clauseids, add to array */ |
1894 | 0 | pathinfoarray[npaths++] = pathinfo; |
1895 | 0 | } |
1896 | 0 | } |
1897 | | |
1898 | | /* If only one surviving path, we're done */ |
1899 | 0 | if (npaths == 1) |
1900 | 0 | return pathinfoarray[0]->path; |
1901 | | |
1902 | | /* Sort the surviving paths by index access cost */ |
1903 | 0 | qsort(pathinfoarray, npaths, sizeof(PathClauseUsage *), |
1904 | 0 | path_usage_comparator); |
1905 | | |
1906 | | /* |
1907 | | * For each surviving index, consider it as an "AND group leader", and see |
1908 | | * whether adding on any of the later indexes results in an AND path with |
1909 | | * cheaper total cost than before. Then take the cheapest AND group. |
1910 | | * |
1911 | | * Note: paths that are either clauseless or unclassifiable will have |
1912 | | * empty clauseids, so that they will not be rejected by the clauseids |
1913 | | * filter here, nor will they cause later paths to be rejected by it. |
1914 | | */ |
1915 | 0 | for (i = 0; i < npaths; i++) |
1916 | 0 | { |
1917 | 0 | Cost costsofar; |
1918 | 0 | List *qualsofar; |
1919 | 0 | Bitmapset *clauseidsofar; |
1920 | |
|
1921 | 0 | pathinfo = pathinfoarray[i]; |
1922 | 0 | paths = list_make1(pathinfo->path); |
1923 | 0 | costsofar = bitmap_scan_cost_est(root, rel, pathinfo->path); |
1924 | 0 | qualsofar = list_concat_copy(pathinfo->quals, pathinfo->preds); |
1925 | 0 | clauseidsofar = bms_copy(pathinfo->clauseids); |
1926 | |
|
1927 | 0 | for (j = i + 1; j < npaths; j++) |
1928 | 0 | { |
1929 | 0 | Cost newcost; |
1930 | |
|
1931 | 0 | pathinfo = pathinfoarray[j]; |
1932 | | /* Check for redundancy */ |
1933 | 0 | if (bms_overlap(pathinfo->clauseids, clauseidsofar)) |
1934 | 0 | continue; /* consider it redundant */ |
1935 | 0 | if (pathinfo->preds) |
1936 | 0 | { |
1937 | 0 | bool redundant = false; |
1938 | | |
1939 | | /* we check each predicate clause separately */ |
1940 | 0 | foreach(l, pathinfo->preds) |
1941 | 0 | { |
1942 | 0 | Node *np = (Node *) lfirst(l); |
1943 | |
|
1944 | 0 | if (predicate_implied_by(list_make1(np), qualsofar, false)) |
1945 | 0 | { |
1946 | 0 | redundant = true; |
1947 | 0 | break; /* out of inner foreach loop */ |
1948 | 0 | } |
1949 | 0 | } |
1950 | 0 | if (redundant) |
1951 | 0 | continue; |
1952 | 0 | } |
1953 | | /* tentatively add new path to paths, so we can estimate cost */ |
1954 | 0 | paths = lappend(paths, pathinfo->path); |
1955 | 0 | newcost = bitmap_and_cost_est(root, rel, paths); |
1956 | 0 | if (newcost < costsofar) |
1957 | 0 | { |
1958 | | /* keep new path in paths, update subsidiary variables */ |
1959 | 0 | costsofar = newcost; |
1960 | 0 | qualsofar = list_concat(qualsofar, pathinfo->quals); |
1961 | 0 | qualsofar = list_concat(qualsofar, pathinfo->preds); |
1962 | 0 | clauseidsofar = bms_add_members(clauseidsofar, |
1963 | 0 | pathinfo->clauseids); |
1964 | 0 | } |
1965 | 0 | else |
1966 | 0 | { |
1967 | | /* reject new path, remove it from paths list */ |
1968 | 0 | paths = list_truncate(paths, list_length(paths) - 1); |
1969 | 0 | } |
1970 | 0 | } |
1971 | | |
1972 | | /* Keep the cheapest AND-group (or singleton) */ |
1973 | 0 | if (i == 0 || costsofar < bestcost) |
1974 | 0 | { |
1975 | 0 | bestpaths = paths; |
1976 | 0 | bestcost = costsofar; |
1977 | 0 | } |
1978 | | |
1979 | | /* some easy cleanup (we don't try real hard though) */ |
1980 | 0 | list_free(qualsofar); |
1981 | 0 | } |
1982 | |
|
1983 | 0 | if (list_length(bestpaths) == 1) |
1984 | 0 | return (Path *) linitial(bestpaths); /* no need for AND */ |
1985 | 0 | return (Path *) create_bitmap_and_path(root, rel, bestpaths); |
1986 | 0 | } |
1987 | | |
1988 | | /* qsort comparator to sort in increasing index access cost order */ |
1989 | | static int |
1990 | | path_usage_comparator(const void *a, const void *b) |
1991 | 0 | { |
1992 | 0 | PathClauseUsage *pa = *(PathClauseUsage *const *) a; |
1993 | 0 | PathClauseUsage *pb = *(PathClauseUsage *const *) b; |
1994 | 0 | Cost acost; |
1995 | 0 | Cost bcost; |
1996 | 0 | Selectivity aselec; |
1997 | 0 | Selectivity bselec; |
1998 | |
|
1999 | 0 | cost_bitmap_tree_node(pa->path, &acost, &aselec); |
2000 | 0 | cost_bitmap_tree_node(pb->path, &bcost, &bselec); |
2001 | | |
2002 | | /* |
2003 | | * If costs are the same, sort by selectivity. |
2004 | | */ |
2005 | 0 | if (acost < bcost) |
2006 | 0 | return -1; |
2007 | 0 | if (acost > bcost) |
2008 | 0 | return 1; |
2009 | | |
2010 | 0 | if (aselec < bselec) |
2011 | 0 | return -1; |
2012 | 0 | if (aselec > bselec) |
2013 | 0 | return 1; |
2014 | | |
2015 | 0 | return 0; |
2016 | 0 | } |
2017 | | |
2018 | | /* |
2019 | | * Estimate the cost of actually executing a bitmap scan with a single |
2020 | | * index path (which could be a BitmapAnd or BitmapOr node). |
2021 | | */ |
2022 | | static Cost |
2023 | | bitmap_scan_cost_est(PlannerInfo *root, RelOptInfo *rel, Path *ipath) |
2024 | 0 | { |
2025 | 0 | BitmapHeapPath bpath; |
2026 | | |
2027 | | /* Set up a dummy BitmapHeapPath */ |
2028 | 0 | bpath.path.type = T_BitmapHeapPath; |
2029 | 0 | bpath.path.pathtype = T_BitmapHeapScan; |
2030 | 0 | bpath.path.parent = rel; |
2031 | 0 | bpath.path.pathtarget = rel->reltarget; |
2032 | 0 | bpath.path.param_info = ipath->param_info; |
2033 | 0 | bpath.path.pathkeys = NIL; |
2034 | 0 | bpath.bitmapqual = ipath; |
2035 | | |
2036 | | /* |
2037 | | * Check the cost of temporary path without considering parallelism. |
2038 | | * Parallel bitmap heap path will be considered at later stage. |
2039 | | */ |
2040 | 0 | bpath.path.parallel_workers = 0; |
2041 | | |
2042 | | /* Now we can do cost_bitmap_heap_scan */ |
2043 | 0 | cost_bitmap_heap_scan(&bpath.path, root, rel, |
2044 | 0 | bpath.path.param_info, |
2045 | 0 | ipath, |
2046 | 0 | get_loop_count(root, rel->relid, |
2047 | 0 | PATH_REQ_OUTER(ipath))); |
2048 | |
|
2049 | 0 | return bpath.path.total_cost; |
2050 | 0 | } |
2051 | | |
2052 | | /* |
2053 | | * Estimate the cost of actually executing a BitmapAnd scan with the given |
2054 | | * inputs. |
2055 | | */ |
2056 | | static Cost |
2057 | | bitmap_and_cost_est(PlannerInfo *root, RelOptInfo *rel, List *paths) |
2058 | 0 | { |
2059 | 0 | BitmapAndPath *apath; |
2060 | | |
2061 | | /* |
2062 | | * Might as well build a real BitmapAndPath here, as the work is slightly |
2063 | | * too complicated to be worth repeating just to save one palloc. |
2064 | | */ |
2065 | 0 | apath = create_bitmap_and_path(root, rel, paths); |
2066 | |
|
2067 | 0 | return bitmap_scan_cost_est(root, rel, (Path *) apath); |
2068 | 0 | } |
2069 | | |
2070 | | |
2071 | | /* |
2072 | | * classify_index_clause_usage |
2073 | | * Construct a PathClauseUsage struct describing the WHERE clauses and |
2074 | | * index predicate clauses used by the given indexscan path. |
2075 | | * We consider two clauses the same if they are equal(). |
2076 | | * |
2077 | | * At some point we might want to migrate this info into the Path data |
2078 | | * structure proper, but for the moment it's only needed within |
2079 | | * choose_bitmap_and(). |
2080 | | * |
2081 | | * *clauselist is used and expanded as needed to identify all the distinct |
2082 | | * clauses seen across successive calls. Caller must initialize it to NIL |
2083 | | * before first call of a set. |
2084 | | */ |
2085 | | static PathClauseUsage * |
2086 | | classify_index_clause_usage(Path *path, List **clauselist) |
2087 | 0 | { |
2088 | 0 | PathClauseUsage *result; |
2089 | 0 | Bitmapset *clauseids; |
2090 | 0 | ListCell *lc; |
2091 | |
|
2092 | 0 | result = (PathClauseUsage *) palloc(sizeof(PathClauseUsage)); |
2093 | 0 | result->path = path; |
2094 | | |
2095 | | /* Recursively find the quals and preds used by the path */ |
2096 | 0 | result->quals = NIL; |
2097 | 0 | result->preds = NIL; |
2098 | 0 | find_indexpath_quals(path, &result->quals, &result->preds); |
2099 | | |
2100 | | /* |
2101 | | * Some machine-generated queries have outlandish numbers of qual clauses. |
2102 | | * To avoid getting into O(N^2) behavior even in this preliminary |
2103 | | * classification step, we want to limit the number of entries we can |
2104 | | * accumulate in *clauselist. Treat any path with more than 100 quals + |
2105 | | * preds as unclassifiable, which will cause calling code to consider it |
2106 | | * distinct from all other paths. |
2107 | | */ |
2108 | 0 | if (list_length(result->quals) + list_length(result->preds) > 100) |
2109 | 0 | { |
2110 | 0 | result->clauseids = NULL; |
2111 | 0 | result->unclassifiable = true; |
2112 | 0 | return result; |
2113 | 0 | } |
2114 | | |
2115 | | /* Build up a bitmapset representing the quals and preds */ |
2116 | 0 | clauseids = NULL; |
2117 | 0 | foreach(lc, result->quals) |
2118 | 0 | { |
2119 | 0 | Node *node = (Node *) lfirst(lc); |
2120 | |
|
2121 | 0 | clauseids = bms_add_member(clauseids, |
2122 | 0 | find_list_position(node, clauselist)); |
2123 | 0 | } |
2124 | 0 | foreach(lc, result->preds) |
2125 | 0 | { |
2126 | 0 | Node *node = (Node *) lfirst(lc); |
2127 | |
|
2128 | 0 | clauseids = bms_add_member(clauseids, |
2129 | 0 | find_list_position(node, clauselist)); |
2130 | 0 | } |
2131 | 0 | result->clauseids = clauseids; |
2132 | 0 | result->unclassifiable = false; |
2133 | |
|
2134 | 0 | return result; |
2135 | 0 | } |
2136 | | |
2137 | | |
2138 | | /* |
2139 | | * find_indexpath_quals |
2140 | | * |
2141 | | * Given the Path structure for a plain or bitmap indexscan, extract lists |
2142 | | * of all the index clauses and index predicate conditions used in the Path. |
2143 | | * These are appended to the initial contents of *quals and *preds (hence |
2144 | | * caller should initialize those to NIL). |
2145 | | * |
2146 | | * Note we are not trying to produce an accurate representation of the AND/OR |
2147 | | * semantics of the Path, but just find out all the base conditions used. |
2148 | | * |
2149 | | * The result lists contain pointers to the expressions used in the Path, |
2150 | | * but all the list cells are freshly built, so it's safe to destructively |
2151 | | * modify the lists (eg, by concat'ing with other lists). |
2152 | | */ |
2153 | | static void |
2154 | | find_indexpath_quals(Path *bitmapqual, List **quals, List **preds) |
2155 | 0 | { |
2156 | 0 | if (IsA(bitmapqual, BitmapAndPath)) |
2157 | 0 | { |
2158 | 0 | BitmapAndPath *apath = (BitmapAndPath *) bitmapqual; |
2159 | 0 | ListCell *l; |
2160 | |
|
2161 | 0 | foreach(l, apath->bitmapquals) |
2162 | 0 | { |
2163 | 0 | find_indexpath_quals((Path *) lfirst(l), quals, preds); |
2164 | 0 | } |
2165 | 0 | } |
2166 | 0 | else if (IsA(bitmapqual, BitmapOrPath)) |
2167 | 0 | { |
2168 | 0 | BitmapOrPath *opath = (BitmapOrPath *) bitmapqual; |
2169 | 0 | ListCell *l; |
2170 | |
|
2171 | 0 | foreach(l, opath->bitmapquals) |
2172 | 0 | { |
2173 | 0 | find_indexpath_quals((Path *) lfirst(l), quals, preds); |
2174 | 0 | } |
2175 | 0 | } |
2176 | 0 | else if (IsA(bitmapqual, IndexPath)) |
2177 | 0 | { |
2178 | 0 | IndexPath *ipath = (IndexPath *) bitmapqual; |
2179 | 0 | ListCell *l; |
2180 | |
|
2181 | 0 | foreach(l, ipath->indexclauses) |
2182 | 0 | { |
2183 | 0 | IndexClause *iclause = (IndexClause *) lfirst(l); |
2184 | |
|
2185 | 0 | *quals = lappend(*quals, iclause->rinfo->clause); |
2186 | 0 | } |
2187 | 0 | *preds = list_concat(*preds, ipath->indexinfo->indpred); |
2188 | 0 | } |
2189 | 0 | else |
2190 | 0 | elog(ERROR, "unrecognized node type: %d", nodeTag(bitmapqual)); |
2191 | 0 | } |
2192 | | |
2193 | | |
2194 | | /* |
2195 | | * find_list_position |
2196 | | * Return the given node's position (counting from 0) in the given |
2197 | | * list of nodes. If it's not equal() to any existing list member, |
2198 | | * add it at the end, and return that position. |
2199 | | */ |
2200 | | static int |
2201 | | find_list_position(Node *node, List **nodelist) |
2202 | 0 | { |
2203 | 0 | int i; |
2204 | 0 | ListCell *lc; |
2205 | |
|
2206 | 0 | i = 0; |
2207 | 0 | foreach(lc, *nodelist) |
2208 | 0 | { |
2209 | 0 | Node *oldnode = (Node *) lfirst(lc); |
2210 | |
|
2211 | 0 | if (equal(node, oldnode)) |
2212 | 0 | return i; |
2213 | 0 | i++; |
2214 | 0 | } |
2215 | | |
2216 | 0 | *nodelist = lappend(*nodelist, node); |
2217 | |
|
2218 | 0 | return i; |
2219 | 0 | } |
2220 | | |
2221 | | |
2222 | | /* |
2223 | | * check_index_only |
2224 | | * Determine whether an index-only scan is possible for this index. |
2225 | | */ |
2226 | | static bool |
2227 | | check_index_only(RelOptInfo *rel, IndexOptInfo *index) |
2228 | 0 | { |
2229 | 0 | bool result; |
2230 | 0 | Bitmapset *attrs_used = NULL; |
2231 | 0 | Bitmapset *index_canreturn_attrs = NULL; |
2232 | 0 | ListCell *lc; |
2233 | 0 | int i; |
2234 | | |
2235 | | /* Index-only scans must be enabled */ |
2236 | 0 | if (!enable_indexonlyscan) |
2237 | 0 | return false; |
2238 | | |
2239 | | /* |
2240 | | * Check that all needed attributes of the relation are available from the |
2241 | | * index. |
2242 | | */ |
2243 | | |
2244 | | /* |
2245 | | * First, identify all the attributes needed for joins or final output. |
2246 | | * Note: we must look at rel's targetlist, not the attr_needed data, |
2247 | | * because attr_needed isn't computed for inheritance child rels. |
2248 | | */ |
2249 | 0 | pull_varattnos((Node *) rel->reltarget->exprs, rel->relid, &attrs_used); |
2250 | | |
2251 | | /* |
2252 | | * Add all the attributes used by restriction clauses; but consider only |
2253 | | * those clauses not implied by the index predicate, since ones that are |
2254 | | * so implied don't need to be checked explicitly in the plan. |
2255 | | * |
2256 | | * Note: attributes used only in index quals would not be needed at |
2257 | | * runtime either, if we are certain that the index is not lossy. However |
2258 | | * it'd be complicated to account for that accurately, and it doesn't |
2259 | | * matter in most cases, since we'd conclude that such attributes are |
2260 | | * available from the index anyway. |
2261 | | */ |
2262 | 0 | foreach(lc, index->indrestrictinfo) |
2263 | 0 | { |
2264 | 0 | RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc); |
2265 | |
|
2266 | 0 | pull_varattnos((Node *) rinfo->clause, rel->relid, &attrs_used); |
2267 | 0 | } |
2268 | | |
2269 | | /* |
2270 | | * Construct a bitmapset of columns that the index can return back in an |
2271 | | * index-only scan. |
2272 | | */ |
2273 | 0 | for (i = 0; i < index->ncolumns; i++) |
2274 | 0 | { |
2275 | 0 | int attno = index->indexkeys[i]; |
2276 | | |
2277 | | /* |
2278 | | * For the moment, we just ignore index expressions. It might be nice |
2279 | | * to do something with them, later. |
2280 | | */ |
2281 | 0 | if (attno == 0) |
2282 | 0 | continue; |
2283 | | |
2284 | 0 | if (index->canreturn[i]) |
2285 | 0 | index_canreturn_attrs = |
2286 | 0 | bms_add_member(index_canreturn_attrs, |
2287 | 0 | attno - FirstLowInvalidHeapAttributeNumber); |
2288 | 0 | } |
2289 | | |
2290 | | /* Do we have all the necessary attributes? */ |
2291 | 0 | result = bms_is_subset(attrs_used, index_canreturn_attrs); |
2292 | |
|
2293 | 0 | bms_free(attrs_used); |
2294 | 0 | bms_free(index_canreturn_attrs); |
2295 | |
|
2296 | 0 | return result; |
2297 | 0 | } |
2298 | | |
2299 | | /* |
2300 | | * get_loop_count |
2301 | | * Choose the loop count estimate to use for costing a parameterized path |
2302 | | * with the given set of outer relids. |
2303 | | * |
2304 | | * Since we produce parameterized paths before we've begun to generate join |
2305 | | * relations, it's impossible to predict exactly how many times a parameterized |
2306 | | * path will be iterated; we don't know the size of the relation that will be |
2307 | | * on the outside of the nestloop. However, we should try to account for |
2308 | | * multiple iterations somehow in costing the path. The heuristic embodied |
2309 | | * here is to use the rowcount of the smallest other base relation needed in |
2310 | | * the join clauses used by the path. (We could alternatively consider the |
2311 | | * largest one, but that seems too optimistic.) This is of course the right |
2312 | | * answer for single-other-relation cases, and it seems like a reasonable |
2313 | | * zero-order approximation for multiway-join cases. |
2314 | | * |
2315 | | * In addition, we check to see if the other side of each join clause is on |
2316 | | * the inside of some semijoin that the current relation is on the outside of. |
2317 | | * If so, the only way that a parameterized path could be used is if the |
2318 | | * semijoin RHS has been unique-ified, so we should use the number of unique |
2319 | | * RHS rows rather than using the relation's raw rowcount. |
2320 | | * |
2321 | | * Note: for this to work, allpaths.c must establish all baserel size |
2322 | | * estimates before it begins to compute paths, or at least before it |
2323 | | * calls create_index_paths(). |
2324 | | */ |
2325 | | static double |
2326 | | get_loop_count(PlannerInfo *root, Index cur_relid, Relids outer_relids) |
2327 | 0 | { |
2328 | 0 | double result; |
2329 | 0 | int outer_relid; |
2330 | | |
2331 | | /* For a non-parameterized path, just return 1.0 quickly */ |
2332 | 0 | if (outer_relids == NULL) |
2333 | 0 | return 1.0; |
2334 | | |
2335 | 0 | result = 0.0; |
2336 | 0 | outer_relid = -1; |
2337 | 0 | while ((outer_relid = bms_next_member(outer_relids, outer_relid)) >= 0) |
2338 | 0 | { |
2339 | 0 | RelOptInfo *outer_rel; |
2340 | 0 | double rowcount; |
2341 | | |
2342 | | /* Paranoia: ignore bogus relid indexes */ |
2343 | 0 | if (outer_relid >= root->simple_rel_array_size) |
2344 | 0 | continue; |
2345 | 0 | outer_rel = root->simple_rel_array[outer_relid]; |
2346 | 0 | if (outer_rel == NULL) |
2347 | 0 | continue; |
2348 | 0 | Assert(outer_rel->relid == outer_relid); /* sanity check on array */ |
2349 | | |
2350 | | /* Other relation could be proven empty, if so ignore */ |
2351 | 0 | if (IS_DUMMY_REL(outer_rel)) |
2352 | 0 | continue; |
2353 | | |
2354 | | /* Otherwise, rel's rows estimate should be valid by now */ |
2355 | 0 | Assert(outer_rel->rows > 0); |
2356 | | |
2357 | | /* Check to see if rel is on the inside of any semijoins */ |
2358 | 0 | rowcount = adjust_rowcount_for_semijoins(root, |
2359 | 0 | cur_relid, |
2360 | 0 | outer_relid, |
2361 | 0 | outer_rel->rows); |
2362 | | |
2363 | | /* Remember smallest row count estimate among the outer rels */ |
2364 | 0 | if (result == 0.0 || result > rowcount) |
2365 | 0 | result = rowcount; |
2366 | 0 | } |
2367 | | /* Return 1.0 if we found no valid relations (shouldn't happen) */ |
2368 | 0 | return (result > 0.0) ? result : 1.0; |
2369 | 0 | } |
2370 | | |
2371 | | /* |
2372 | | * Check to see if outer_relid is on the inside of any semijoin that cur_relid |
2373 | | * is on the outside of. If so, replace rowcount with the estimated number of |
2374 | | * unique rows from the semijoin RHS (assuming that's smaller, which it might |
2375 | | * not be). The estimate is crude but it's the best we can do at this stage |
2376 | | * of the proceedings. |
2377 | | */ |
2378 | | static double |
2379 | | adjust_rowcount_for_semijoins(PlannerInfo *root, |
2380 | | Index cur_relid, |
2381 | | Index outer_relid, |
2382 | | double rowcount) |
2383 | 0 | { |
2384 | 0 | ListCell *lc; |
2385 | |
|
2386 | 0 | foreach(lc, root->join_info_list) |
2387 | 0 | { |
2388 | 0 | SpecialJoinInfo *sjinfo = (SpecialJoinInfo *) lfirst(lc); |
2389 | |
|
2390 | 0 | if (sjinfo->jointype == JOIN_SEMI && |
2391 | 0 | bms_is_member(cur_relid, sjinfo->syn_lefthand) && |
2392 | 0 | bms_is_member(outer_relid, sjinfo->syn_righthand)) |
2393 | 0 | { |
2394 | | /* Estimate number of unique-ified rows */ |
2395 | 0 | double nraw; |
2396 | 0 | double nunique; |
2397 | |
|
2398 | 0 | nraw = approximate_joinrel_size(root, sjinfo->syn_righthand); |
2399 | 0 | nunique = estimate_num_groups(root, |
2400 | 0 | sjinfo->semi_rhs_exprs, |
2401 | 0 | nraw, |
2402 | 0 | NULL, |
2403 | 0 | NULL); |
2404 | 0 | if (rowcount > nunique) |
2405 | 0 | rowcount = nunique; |
2406 | 0 | } |
2407 | 0 | } |
2408 | 0 | return rowcount; |
2409 | 0 | } |
2410 | | |
2411 | | /* |
2412 | | * Make an approximate estimate of the size of a joinrel. |
2413 | | * |
2414 | | * We don't have enough info at this point to get a good estimate, so we |
2415 | | * just multiply the base relation sizes together. Fortunately, this is |
2416 | | * the right answer anyway for the most common case with a single relation |
2417 | | * on the RHS of a semijoin. Also, estimate_num_groups() has only a weak |
2418 | | * dependency on its input_rows argument (it basically uses it as a clamp). |
2419 | | * So we might be able to get a fairly decent end result even with a severe |
2420 | | * overestimate of the RHS's raw size. |
2421 | | */ |
2422 | | static double |
2423 | | approximate_joinrel_size(PlannerInfo *root, Relids relids) |
2424 | 0 | { |
2425 | 0 | double rowcount = 1.0; |
2426 | 0 | int relid; |
2427 | |
|
2428 | 0 | relid = -1; |
2429 | 0 | while ((relid = bms_next_member(relids, relid)) >= 0) |
2430 | 0 | { |
2431 | 0 | RelOptInfo *rel; |
2432 | | |
2433 | | /* Paranoia: ignore bogus relid indexes */ |
2434 | 0 | if (relid >= root->simple_rel_array_size) |
2435 | 0 | continue; |
2436 | 0 | rel = root->simple_rel_array[relid]; |
2437 | 0 | if (rel == NULL) |
2438 | 0 | continue; |
2439 | 0 | Assert(rel->relid == relid); /* sanity check on array */ |
2440 | | |
2441 | | /* Relation could be proven empty, if so ignore */ |
2442 | 0 | if (IS_DUMMY_REL(rel)) |
2443 | 0 | continue; |
2444 | | |
2445 | | /* Otherwise, rel's rows estimate should be valid by now */ |
2446 | 0 | Assert(rel->rows > 0); |
2447 | | |
2448 | | /* Accumulate product */ |
2449 | 0 | rowcount *= rel->rows; |
2450 | 0 | } |
2451 | 0 | return rowcount; |
2452 | 0 | } |
2453 | | |
2454 | | |
2455 | | /**************************************************************************** |
2456 | | * ---- ROUTINES TO CHECK QUERY CLAUSES ---- |
2457 | | ****************************************************************************/ |
2458 | | |
2459 | | /* |
2460 | | * match_restriction_clauses_to_index |
2461 | | * Identify restriction clauses for the rel that match the index. |
2462 | | * Matching clauses are added to *clauseset. |
2463 | | */ |
2464 | | static void |
2465 | | match_restriction_clauses_to_index(PlannerInfo *root, |
2466 | | IndexOptInfo *index, |
2467 | | IndexClauseSet *clauseset) |
2468 | 0 | { |
2469 | | /* We can ignore clauses that are implied by the index predicate */ |
2470 | 0 | match_clauses_to_index(root, index->indrestrictinfo, index, clauseset); |
2471 | 0 | } |
2472 | | |
2473 | | /* |
2474 | | * match_join_clauses_to_index |
2475 | | * Identify join clauses for the rel that match the index. |
2476 | | * Matching clauses are added to *clauseset. |
2477 | | * Also, add any potentially usable join OR clauses to *joinorclauses. |
2478 | | * They also might be processed by match_clause_to_index() as a whole. |
2479 | | */ |
2480 | | static void |
2481 | | match_join_clauses_to_index(PlannerInfo *root, |
2482 | | RelOptInfo *rel, IndexOptInfo *index, |
2483 | | IndexClauseSet *clauseset, |
2484 | | List **joinorclauses) |
2485 | 0 | { |
2486 | 0 | ListCell *lc; |
2487 | | |
2488 | | /* Scan the rel's join clauses */ |
2489 | 0 | foreach(lc, rel->joininfo) |
2490 | 0 | { |
2491 | 0 | RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc); |
2492 | | |
2493 | | /* Check if clause can be moved to this rel */ |
2494 | 0 | if (!join_clause_is_movable_to(rinfo, rel)) |
2495 | 0 | continue; |
2496 | | |
2497 | | /* |
2498 | | * Potentially usable, so see if it matches the index or is an OR. Use |
2499 | | * list_append_unique_ptr() here to avoid possible duplicates when |
2500 | | * processing the same clauses with different indexes. |
2501 | | */ |
2502 | 0 | if (restriction_is_or_clause(rinfo)) |
2503 | 0 | *joinorclauses = list_append_unique_ptr(*joinorclauses, rinfo); |
2504 | |
|
2505 | 0 | match_clause_to_index(root, rinfo, index, clauseset); |
2506 | 0 | } |
2507 | 0 | } |
2508 | | |
2509 | | /* |
2510 | | * match_eclass_clauses_to_index |
2511 | | * Identify EquivalenceClass join clauses for the rel that match the index. |
2512 | | * Matching clauses are added to *clauseset. |
2513 | | */ |
2514 | | static void |
2515 | | match_eclass_clauses_to_index(PlannerInfo *root, IndexOptInfo *index, |
2516 | | IndexClauseSet *clauseset) |
2517 | 0 | { |
2518 | 0 | int indexcol; |
2519 | | |
2520 | | /* No work if rel is not in any such ECs */ |
2521 | 0 | if (!index->rel->has_eclass_joins) |
2522 | 0 | return; |
2523 | | |
2524 | 0 | for (indexcol = 0; indexcol < index->nkeycolumns; indexcol++) |
2525 | 0 | { |
2526 | 0 | ec_member_matches_arg arg; |
2527 | 0 | List *clauses; |
2528 | | |
2529 | | /* Generate clauses, skipping any that join to lateral_referencers */ |
2530 | 0 | arg.index = index; |
2531 | 0 | arg.indexcol = indexcol; |
2532 | 0 | clauses = generate_implied_equalities_for_column(root, |
2533 | 0 | index->rel, |
2534 | 0 | ec_member_matches_indexcol, |
2535 | 0 | &arg, |
2536 | 0 | index->rel->lateral_referencers); |
2537 | | |
2538 | | /* |
2539 | | * We have to check whether the results actually do match the index, |
2540 | | * since for non-btree indexes the EC's equality operators might not |
2541 | | * be in the index opclass (cf ec_member_matches_indexcol). |
2542 | | */ |
2543 | 0 | match_clauses_to_index(root, clauses, index, clauseset); |
2544 | 0 | } |
2545 | 0 | } |
2546 | | |
2547 | | /* |
2548 | | * match_clauses_to_index |
2549 | | * Perform match_clause_to_index() for each clause in a list. |
2550 | | * Matching clauses are added to *clauseset. |
2551 | | */ |
2552 | | static void |
2553 | | match_clauses_to_index(PlannerInfo *root, |
2554 | | List *clauses, |
2555 | | IndexOptInfo *index, |
2556 | | IndexClauseSet *clauseset) |
2557 | 0 | { |
2558 | 0 | ListCell *lc; |
2559 | |
|
2560 | 0 | foreach(lc, clauses) |
2561 | 0 | { |
2562 | 0 | RestrictInfo *rinfo = lfirst_node(RestrictInfo, lc); |
2563 | |
|
2564 | 0 | match_clause_to_index(root, rinfo, index, clauseset); |
2565 | 0 | } |
2566 | 0 | } |
2567 | | |
2568 | | /* |
2569 | | * match_clause_to_index |
2570 | | * Test whether a qual clause can be used with an index. |
2571 | | * |
2572 | | * If the clause is usable, add an IndexClause entry for it to the appropriate |
2573 | | * list in *clauseset. (*clauseset must be initialized to zeroes before first |
2574 | | * call.) |
2575 | | * |
2576 | | * Note: in some circumstances we may find the same RestrictInfos coming from |
2577 | | * multiple places. Defend against redundant outputs by refusing to add a |
2578 | | * clause twice (pointer equality should be a good enough check for this). |
2579 | | * |
2580 | | * Note: it's possible that a badly-defined index could have multiple matching |
2581 | | * columns. We always select the first match if so; this avoids scenarios |
2582 | | * wherein we get an inflated idea of the index's selectivity by using the |
2583 | | * same clause multiple times with different index columns. |
2584 | | */ |
2585 | | static void |
2586 | | match_clause_to_index(PlannerInfo *root, |
2587 | | RestrictInfo *rinfo, |
2588 | | IndexOptInfo *index, |
2589 | | IndexClauseSet *clauseset) |
2590 | 0 | { |
2591 | 0 | int indexcol; |
2592 | | |
2593 | | /* |
2594 | | * Never match pseudoconstants to indexes. (Normally a match could not |
2595 | | * happen anyway, since a pseudoconstant clause couldn't contain a Var, |
2596 | | * but what if someone builds an expression index on a constant? It's not |
2597 | | * totally unreasonable to do so with a partial index, either.) |
2598 | | */ |
2599 | 0 | if (rinfo->pseudoconstant) |
2600 | 0 | return; |
2601 | | |
2602 | | /* |
2603 | | * If clause can't be used as an indexqual because it must wait till after |
2604 | | * some lower-security-level restriction clause, reject it. |
2605 | | */ |
2606 | 0 | if (!restriction_is_securely_promotable(rinfo, index->rel)) |
2607 | 0 | return; |
2608 | | |
2609 | | /* OK, check each index key column for a match */ |
2610 | 0 | for (indexcol = 0; indexcol < index->nkeycolumns; indexcol++) |
2611 | 0 | { |
2612 | 0 | IndexClause *iclause; |
2613 | 0 | ListCell *lc; |
2614 | | |
2615 | | /* Ignore duplicates */ |
2616 | 0 | foreach(lc, clauseset->indexclauses[indexcol]) |
2617 | 0 | { |
2618 | 0 | iclause = (IndexClause *) lfirst(lc); |
2619 | |
|
2620 | 0 | if (iclause->rinfo == rinfo) |
2621 | 0 | return; |
2622 | 0 | } |
2623 | | |
2624 | | /* OK, try to match the clause to the index column */ |
2625 | 0 | iclause = match_clause_to_indexcol(root, |
2626 | 0 | rinfo, |
2627 | 0 | indexcol, |
2628 | 0 | index); |
2629 | 0 | if (iclause) |
2630 | 0 | { |
2631 | | /* Success, so record it */ |
2632 | 0 | clauseset->indexclauses[indexcol] = |
2633 | 0 | lappend(clauseset->indexclauses[indexcol], iclause); |
2634 | 0 | clauseset->nonempty = true; |
2635 | 0 | return; |
2636 | 0 | } |
2637 | 0 | } |
2638 | 0 | } |
2639 | | |
2640 | | /* |
2641 | | * match_clause_to_indexcol() |
2642 | | * Determine whether a restriction clause matches a column of an index, |
2643 | | * and if so, build an IndexClause node describing the details. |
2644 | | * |
2645 | | * To match an index normally, an operator clause: |
2646 | | * |
2647 | | * (1) must be in the form (indexkey op const) or (const op indexkey); |
2648 | | * and |
2649 | | * (2) must contain an operator which is in the index's operator family |
2650 | | * for this column; and |
2651 | | * (3) must match the collation of the index, if collation is relevant. |
2652 | | * |
2653 | | * Our definition of "const" is exceedingly liberal: we allow anything that |
2654 | | * doesn't involve a volatile function or a Var of the index's relation. |
2655 | | * In particular, Vars belonging to other relations of the query are |
2656 | | * accepted here, since a clause of that form can be used in a |
2657 | | * parameterized indexscan. It's the responsibility of higher code levels |
2658 | | * to manage restriction and join clauses appropriately. |
2659 | | * |
2660 | | * Note: we do need to check for Vars of the index's relation on the |
2661 | | * "const" side of the clause, since clauses like (a.f1 OP (b.f2 OP a.f3)) |
2662 | | * are not processable by a parameterized indexscan on a.f1, whereas |
2663 | | * something like (a.f1 OP (b.f2 OP c.f3)) is. |
2664 | | * |
2665 | | * Presently, the executor can only deal with indexquals that have the |
2666 | | * indexkey on the left, so we can only use clauses that have the indexkey |
2667 | | * on the right if we can commute the clause to put the key on the left. |
2668 | | * We handle that by generating an IndexClause with the correctly-commuted |
2669 | | * opclause as a derived indexqual. |
2670 | | * |
2671 | | * If the index has a collation, the clause must have the same collation. |
2672 | | * For collation-less indexes, we assume it doesn't matter; this is |
2673 | | * necessary for cases like "hstore ? text", wherein hstore's operators |
2674 | | * don't care about collation but the clause will get marked with a |
2675 | | * collation anyway because of the text argument. (This logic is |
2676 | | * embodied in the macro IndexCollMatchesExprColl.) |
2677 | | * |
2678 | | * It is also possible to match RowCompareExpr clauses to indexes (but |
2679 | | * currently, only btree indexes handle this). |
2680 | | * |
2681 | | * It is also possible to match ScalarArrayOpExpr clauses to indexes, when |
2682 | | * the clause is of the form "indexkey op ANY (arrayconst)". |
2683 | | * |
2684 | | * It is also possible to match a list of OR clauses if it might be |
2685 | | * transformed into a single ScalarArrayOpExpr clause. On success, |
2686 | | * the returning index clause will contain a transformed clause. |
2687 | | * |
2688 | | * For boolean indexes, it is also possible to match the clause directly |
2689 | | * to the indexkey; or perhaps the clause is (NOT indexkey). |
2690 | | * |
2691 | | * And, last but not least, some operators and functions can be processed |
2692 | | * to derive (typically lossy) indexquals from a clause that isn't in |
2693 | | * itself indexable. If we see that any operand of an OpExpr or FuncExpr |
2694 | | * matches the index key, and the function has a planner support function |
2695 | | * attached to it, we'll invoke the support function to see if such an |
2696 | | * indexqual can be built. |
2697 | | * |
2698 | | * 'rinfo' is the clause to be tested (as a RestrictInfo node). |
2699 | | * 'indexcol' is a column number of 'index' (counting from 0). |
2700 | | * 'index' is the index of interest. |
2701 | | * |
2702 | | * Returns an IndexClause if the clause can be used with this index key, |
2703 | | * or NULL if not. |
2704 | | * |
2705 | | * NOTE: This routine always returns NULL if the clause is an AND clause. |
2706 | | * Higher-level routines deal with OR and AND clauses. OR clause can be |
2707 | | * matched as a whole by match_orclause_to_indexcol() though. |
2708 | | */ |
2709 | | static IndexClause * |
2710 | | match_clause_to_indexcol(PlannerInfo *root, |
2711 | | RestrictInfo *rinfo, |
2712 | | int indexcol, |
2713 | | IndexOptInfo *index) |
2714 | 0 | { |
2715 | 0 | IndexClause *iclause; |
2716 | 0 | Expr *clause = rinfo->clause; |
2717 | 0 | Oid opfamily; |
2718 | |
|
2719 | 0 | Assert(indexcol < index->nkeycolumns); |
2720 | | |
2721 | | /* |
2722 | | * Historically this code has coped with NULL clauses. That's probably |
2723 | | * not possible anymore, but we might as well continue to cope. |
2724 | | */ |
2725 | 0 | if (clause == NULL) |
2726 | 0 | return NULL; |
2727 | | |
2728 | | /* First check for boolean-index cases. */ |
2729 | 0 | opfamily = index->opfamily[indexcol]; |
2730 | 0 | if (IsBooleanOpfamily(opfamily)) |
2731 | 0 | { |
2732 | 0 | iclause = match_boolean_index_clause(root, rinfo, indexcol, index); |
2733 | 0 | if (iclause) |
2734 | 0 | return iclause; |
2735 | 0 | } |
2736 | | |
2737 | | /* |
2738 | | * Clause must be an opclause, funcclause, ScalarArrayOpExpr, |
2739 | | * RowCompareExpr, or OR-clause that could be converted to SAOP. Or, if |
2740 | | * the index supports it, we can handle IS NULL/NOT NULL clauses. |
2741 | | */ |
2742 | 0 | if (IsA(clause, OpExpr)) |
2743 | 0 | { |
2744 | 0 | return match_opclause_to_indexcol(root, rinfo, indexcol, index); |
2745 | 0 | } |
2746 | 0 | else if (IsA(clause, FuncExpr)) |
2747 | 0 | { |
2748 | 0 | return match_funcclause_to_indexcol(root, rinfo, indexcol, index); |
2749 | 0 | } |
2750 | 0 | else if (IsA(clause, ScalarArrayOpExpr)) |
2751 | 0 | { |
2752 | 0 | return match_saopclause_to_indexcol(root, rinfo, indexcol, index); |
2753 | 0 | } |
2754 | 0 | else if (IsA(clause, RowCompareExpr)) |
2755 | 0 | { |
2756 | 0 | return match_rowcompare_to_indexcol(root, rinfo, indexcol, index); |
2757 | 0 | } |
2758 | 0 | else if (restriction_is_or_clause(rinfo)) |
2759 | 0 | { |
2760 | 0 | return match_orclause_to_indexcol(root, rinfo, indexcol, index); |
2761 | 0 | } |
2762 | 0 | else if (index->amsearchnulls && IsA(clause, NullTest)) |
2763 | 0 | { |
2764 | 0 | NullTest *nt = (NullTest *) clause; |
2765 | |
|
2766 | 0 | if (!nt->argisrow && |
2767 | 0 | match_index_to_operand((Node *) nt->arg, indexcol, index)) |
2768 | 0 | { |
2769 | 0 | iclause = makeNode(IndexClause); |
2770 | 0 | iclause->rinfo = rinfo; |
2771 | 0 | iclause->indexquals = list_make1(rinfo); |
2772 | 0 | iclause->lossy = false; |
2773 | 0 | iclause->indexcol = indexcol; |
2774 | 0 | iclause->indexcols = NIL; |
2775 | 0 | return iclause; |
2776 | 0 | } |
2777 | 0 | } |
2778 | | |
2779 | 0 | return NULL; |
2780 | 0 | } |
2781 | | |
2782 | | /* |
2783 | | * IsBooleanOpfamily |
2784 | | * Detect whether an opfamily supports boolean equality as an operator. |
2785 | | * |
2786 | | * If the opfamily OID is in the range of built-in objects, we can rely |
2787 | | * on hard-wired knowledge of which built-in opfamilies support this. |
2788 | | * For extension opfamilies, there's no choice but to do a catcache lookup. |
2789 | | */ |
2790 | | static bool |
2791 | | IsBooleanOpfamily(Oid opfamily) |
2792 | 0 | { |
2793 | 0 | if (opfamily < FirstNormalObjectId) |
2794 | 0 | return IsBuiltinBooleanOpfamily(opfamily); |
2795 | 0 | else |
2796 | 0 | return op_in_opfamily(BooleanEqualOperator, opfamily); |
2797 | 0 | } |
2798 | | |
2799 | | /* |
2800 | | * match_boolean_index_clause |
2801 | | * Recognize restriction clauses that can be matched to a boolean index. |
2802 | | * |
2803 | | * The idea here is that, for an index on a boolean column that supports the |
2804 | | * BooleanEqualOperator, we can transform a plain reference to the indexkey |
2805 | | * into "indexkey = true", or "NOT indexkey" into "indexkey = false", etc, |
2806 | | * so as to make the expression indexable using the index's "=" operator. |
2807 | | * Since Postgres 8.1, we must do this because constant simplification does |
2808 | | * the reverse transformation; without this code there'd be no way to use |
2809 | | * such an index at all. |
2810 | | * |
2811 | | * This should be called only when IsBooleanOpfamily() recognizes the |
2812 | | * index's operator family. We check to see if the clause matches the |
2813 | | * index's key, and if so, build a suitable IndexClause. |
2814 | | */ |
2815 | | static IndexClause * |
2816 | | match_boolean_index_clause(PlannerInfo *root, |
2817 | | RestrictInfo *rinfo, |
2818 | | int indexcol, |
2819 | | IndexOptInfo *index) |
2820 | 0 | { |
2821 | 0 | Node *clause = (Node *) rinfo->clause; |
2822 | 0 | Expr *op = NULL; |
2823 | | |
2824 | | /* Direct match? */ |
2825 | 0 | if (match_index_to_operand(clause, indexcol, index)) |
2826 | 0 | { |
2827 | | /* convert to indexkey = TRUE */ |
2828 | 0 | op = make_opclause(BooleanEqualOperator, BOOLOID, false, |
2829 | 0 | (Expr *) clause, |
2830 | 0 | (Expr *) makeBoolConst(true, false), |
2831 | 0 | InvalidOid, InvalidOid); |
2832 | 0 | } |
2833 | | /* NOT clause? */ |
2834 | 0 | else if (is_notclause(clause)) |
2835 | 0 | { |
2836 | 0 | Node *arg = (Node *) get_notclausearg((Expr *) clause); |
2837 | |
|
2838 | 0 | if (match_index_to_operand(arg, indexcol, index)) |
2839 | 0 | { |
2840 | | /* convert to indexkey = FALSE */ |
2841 | 0 | op = make_opclause(BooleanEqualOperator, BOOLOID, false, |
2842 | 0 | (Expr *) arg, |
2843 | 0 | (Expr *) makeBoolConst(false, false), |
2844 | 0 | InvalidOid, InvalidOid); |
2845 | 0 | } |
2846 | 0 | } |
2847 | | |
2848 | | /* |
2849 | | * Since we only consider clauses at top level of WHERE, we can convert |
2850 | | * indexkey IS TRUE and indexkey IS FALSE to index searches as well. The |
2851 | | * different meaning for NULL isn't important. |
2852 | | */ |
2853 | 0 | else if (clause && IsA(clause, BooleanTest)) |
2854 | 0 | { |
2855 | 0 | BooleanTest *btest = (BooleanTest *) clause; |
2856 | 0 | Node *arg = (Node *) btest->arg; |
2857 | |
|
2858 | 0 | if (btest->booltesttype == IS_TRUE && |
2859 | 0 | match_index_to_operand(arg, indexcol, index)) |
2860 | 0 | { |
2861 | | /* convert to indexkey = TRUE */ |
2862 | 0 | op = make_opclause(BooleanEqualOperator, BOOLOID, false, |
2863 | 0 | (Expr *) arg, |
2864 | 0 | (Expr *) makeBoolConst(true, false), |
2865 | 0 | InvalidOid, InvalidOid); |
2866 | 0 | } |
2867 | 0 | else if (btest->booltesttype == IS_FALSE && |
2868 | 0 | match_index_to_operand(arg, indexcol, index)) |
2869 | 0 | { |
2870 | | /* convert to indexkey = FALSE */ |
2871 | 0 | op = make_opclause(BooleanEqualOperator, BOOLOID, false, |
2872 | 0 | (Expr *) arg, |
2873 | 0 | (Expr *) makeBoolConst(false, false), |
2874 | 0 | InvalidOid, InvalidOid); |
2875 | 0 | } |
2876 | 0 | } |
2877 | | |
2878 | | /* |
2879 | | * If we successfully made an operator clause from the given qual, we must |
2880 | | * wrap it in an IndexClause. It's not lossy. |
2881 | | */ |
2882 | 0 | if (op) |
2883 | 0 | { |
2884 | 0 | IndexClause *iclause = makeNode(IndexClause); |
2885 | |
|
2886 | 0 | iclause->rinfo = rinfo; |
2887 | 0 | iclause->indexquals = list_make1(make_simple_restrictinfo(root, op)); |
2888 | 0 | iclause->lossy = false; |
2889 | 0 | iclause->indexcol = indexcol; |
2890 | 0 | iclause->indexcols = NIL; |
2891 | 0 | return iclause; |
2892 | 0 | } |
2893 | | |
2894 | 0 | return NULL; |
2895 | 0 | } |
2896 | | |
2897 | | /* |
2898 | | * match_opclause_to_indexcol() |
2899 | | * Handles the OpExpr case for match_clause_to_indexcol(), |
2900 | | * which see for comments. |
2901 | | */ |
2902 | | static IndexClause * |
2903 | | match_opclause_to_indexcol(PlannerInfo *root, |
2904 | | RestrictInfo *rinfo, |
2905 | | int indexcol, |
2906 | | IndexOptInfo *index) |
2907 | 0 | { |
2908 | 0 | IndexClause *iclause; |
2909 | 0 | OpExpr *clause = (OpExpr *) rinfo->clause; |
2910 | 0 | Node *leftop, |
2911 | 0 | *rightop; |
2912 | 0 | Oid expr_op; |
2913 | 0 | Oid expr_coll; |
2914 | 0 | Index index_relid; |
2915 | 0 | Oid opfamily; |
2916 | 0 | Oid idxcollation; |
2917 | | |
2918 | | /* |
2919 | | * Only binary operators need apply. (In theory, a planner support |
2920 | | * function could do something with a unary operator, but it seems |
2921 | | * unlikely to be worth the cycles to check.) |
2922 | | */ |
2923 | 0 | if (list_length(clause->args) != 2) |
2924 | 0 | return NULL; |
2925 | | |
2926 | 0 | leftop = (Node *) linitial(clause->args); |
2927 | 0 | rightop = (Node *) lsecond(clause->args); |
2928 | 0 | expr_op = clause->opno; |
2929 | 0 | expr_coll = clause->inputcollid; |
2930 | |
|
2931 | 0 | index_relid = index->rel->relid; |
2932 | 0 | opfamily = index->opfamily[indexcol]; |
2933 | 0 | idxcollation = index->indexcollations[indexcol]; |
2934 | | |
2935 | | /* |
2936 | | * Check for clauses of the form: (indexkey operator constant) or |
2937 | | * (constant operator indexkey). See match_clause_to_indexcol's notes |
2938 | | * about const-ness. |
2939 | | * |
2940 | | * Note that we don't ask the support function about clauses that don't |
2941 | | * have one of these forms. Again, in principle it might be possible to |
2942 | | * do something, but it seems unlikely to be worth the cycles to check. |
2943 | | */ |
2944 | 0 | if (match_index_to_operand(leftop, indexcol, index) && |
2945 | 0 | !bms_is_member(index_relid, rinfo->right_relids) && |
2946 | 0 | !contain_volatile_functions(rightop)) |
2947 | 0 | { |
2948 | 0 | if (IndexCollMatchesExprColl(idxcollation, expr_coll) && |
2949 | 0 | op_in_opfamily(expr_op, opfamily)) |
2950 | 0 | { |
2951 | 0 | iclause = makeNode(IndexClause); |
2952 | 0 | iclause->rinfo = rinfo; |
2953 | 0 | iclause->indexquals = list_make1(rinfo); |
2954 | 0 | iclause->lossy = false; |
2955 | 0 | iclause->indexcol = indexcol; |
2956 | 0 | iclause->indexcols = NIL; |
2957 | 0 | return iclause; |
2958 | 0 | } |
2959 | | |
2960 | | /* |
2961 | | * If we didn't find a member of the index's opfamily, try the support |
2962 | | * function for the operator's underlying function. |
2963 | | */ |
2964 | 0 | set_opfuncid(clause); /* make sure we have opfuncid */ |
2965 | 0 | return get_index_clause_from_support(root, |
2966 | 0 | rinfo, |
2967 | 0 | clause->opfuncid, |
2968 | 0 | 0, /* indexarg on left */ |
2969 | 0 | indexcol, |
2970 | 0 | index); |
2971 | 0 | } |
2972 | | |
2973 | 0 | if (match_index_to_operand(rightop, indexcol, index) && |
2974 | 0 | !bms_is_member(index_relid, rinfo->left_relids) && |
2975 | 0 | !contain_volatile_functions(leftop)) |
2976 | 0 | { |
2977 | 0 | if (IndexCollMatchesExprColl(idxcollation, expr_coll)) |
2978 | 0 | { |
2979 | 0 | Oid comm_op = get_commutator(expr_op); |
2980 | |
|
2981 | 0 | if (OidIsValid(comm_op) && |
2982 | 0 | op_in_opfamily(comm_op, opfamily)) |
2983 | 0 | { |
2984 | 0 | RestrictInfo *commrinfo; |
2985 | | |
2986 | | /* Build a commuted OpExpr and RestrictInfo */ |
2987 | 0 | commrinfo = commute_restrictinfo(rinfo, comm_op); |
2988 | | |
2989 | | /* Make an IndexClause showing that as a derived qual */ |
2990 | 0 | iclause = makeNode(IndexClause); |
2991 | 0 | iclause->rinfo = rinfo; |
2992 | 0 | iclause->indexquals = list_make1(commrinfo); |
2993 | 0 | iclause->lossy = false; |
2994 | 0 | iclause->indexcol = indexcol; |
2995 | 0 | iclause->indexcols = NIL; |
2996 | 0 | return iclause; |
2997 | 0 | } |
2998 | 0 | } |
2999 | | |
3000 | | /* |
3001 | | * If we didn't find a member of the index's opfamily, try the support |
3002 | | * function for the operator's underlying function. |
3003 | | */ |
3004 | 0 | set_opfuncid(clause); /* make sure we have opfuncid */ |
3005 | 0 | return get_index_clause_from_support(root, |
3006 | 0 | rinfo, |
3007 | 0 | clause->opfuncid, |
3008 | 0 | 1, /* indexarg on right */ |
3009 | 0 | indexcol, |
3010 | 0 | index); |
3011 | 0 | } |
3012 | | |
3013 | 0 | return NULL; |
3014 | 0 | } |
3015 | | |
3016 | | /* |
3017 | | * match_funcclause_to_indexcol() |
3018 | | * Handles the FuncExpr case for match_clause_to_indexcol(), |
3019 | | * which see for comments. |
3020 | | */ |
3021 | | static IndexClause * |
3022 | | match_funcclause_to_indexcol(PlannerInfo *root, |
3023 | | RestrictInfo *rinfo, |
3024 | | int indexcol, |
3025 | | IndexOptInfo *index) |
3026 | 0 | { |
3027 | 0 | FuncExpr *clause = (FuncExpr *) rinfo->clause; |
3028 | 0 | int indexarg; |
3029 | 0 | ListCell *lc; |
3030 | | |
3031 | | /* |
3032 | | * We have no built-in intelligence about function clauses, but if there's |
3033 | | * a planner support function, it might be able to do something. But, to |
3034 | | * cut down on wasted planning cycles, only call the support function if |
3035 | | * at least one argument matches the target index column. |
3036 | | * |
3037 | | * Note that we don't insist on the other arguments being pseudoconstants; |
3038 | | * the support function has to check that. This is to allow cases where |
3039 | | * only some of the other arguments need to be included in the indexqual. |
3040 | | */ |
3041 | 0 | indexarg = 0; |
3042 | 0 | foreach(lc, clause->args) |
3043 | 0 | { |
3044 | 0 | Node *op = (Node *) lfirst(lc); |
3045 | |
|
3046 | 0 | if (match_index_to_operand(op, indexcol, index)) |
3047 | 0 | { |
3048 | 0 | return get_index_clause_from_support(root, |
3049 | 0 | rinfo, |
3050 | 0 | clause->funcid, |
3051 | 0 | indexarg, |
3052 | 0 | indexcol, |
3053 | 0 | index); |
3054 | 0 | } |
3055 | | |
3056 | 0 | indexarg++; |
3057 | 0 | } |
3058 | | |
3059 | 0 | return NULL; |
3060 | 0 | } |
3061 | | |
3062 | | /* |
3063 | | * get_index_clause_from_support() |
3064 | | * If the function has a planner support function, try to construct |
3065 | | * an IndexClause using indexquals created by the support function. |
3066 | | */ |
3067 | | static IndexClause * |
3068 | | get_index_clause_from_support(PlannerInfo *root, |
3069 | | RestrictInfo *rinfo, |
3070 | | Oid funcid, |
3071 | | int indexarg, |
3072 | | int indexcol, |
3073 | | IndexOptInfo *index) |
3074 | 0 | { |
3075 | 0 | Oid prosupport = get_func_support(funcid); |
3076 | 0 | SupportRequestIndexCondition req; |
3077 | 0 | List *sresult; |
3078 | |
|
3079 | 0 | if (!OidIsValid(prosupport)) |
3080 | 0 | return NULL; |
3081 | | |
3082 | 0 | req.type = T_SupportRequestIndexCondition; |
3083 | 0 | req.root = root; |
3084 | 0 | req.funcid = funcid; |
3085 | 0 | req.node = (Node *) rinfo->clause; |
3086 | 0 | req.indexarg = indexarg; |
3087 | 0 | req.index = index; |
3088 | 0 | req.indexcol = indexcol; |
3089 | 0 | req.opfamily = index->opfamily[indexcol]; |
3090 | 0 | req.indexcollation = index->indexcollations[indexcol]; |
3091 | |
|
3092 | 0 | req.lossy = true; /* default assumption */ |
3093 | |
|
3094 | 0 | sresult = (List *) |
3095 | 0 | DatumGetPointer(OidFunctionCall1(prosupport, |
3096 | 0 | PointerGetDatum(&req))); |
3097 | |
|
3098 | 0 | if (sresult != NIL) |
3099 | 0 | { |
3100 | 0 | IndexClause *iclause = makeNode(IndexClause); |
3101 | 0 | List *indexquals = NIL; |
3102 | 0 | ListCell *lc; |
3103 | | |
3104 | | /* |
3105 | | * The support function API says it should just give back bare |
3106 | | * clauses, so here we must wrap each one in a RestrictInfo. |
3107 | | */ |
3108 | 0 | foreach(lc, sresult) |
3109 | 0 | { |
3110 | 0 | Expr *clause = (Expr *) lfirst(lc); |
3111 | |
|
3112 | 0 | indexquals = lappend(indexquals, |
3113 | 0 | make_simple_restrictinfo(root, clause)); |
3114 | 0 | } |
3115 | |
|
3116 | 0 | iclause->rinfo = rinfo; |
3117 | 0 | iclause->indexquals = indexquals; |
3118 | 0 | iclause->lossy = req.lossy; |
3119 | 0 | iclause->indexcol = indexcol; |
3120 | 0 | iclause->indexcols = NIL; |
3121 | |
|
3122 | 0 | return iclause; |
3123 | 0 | } |
3124 | | |
3125 | 0 | return NULL; |
3126 | 0 | } |
3127 | | |
3128 | | /* |
3129 | | * match_saopclause_to_indexcol() |
3130 | | * Handles the ScalarArrayOpExpr case for match_clause_to_indexcol(), |
3131 | | * which see for comments. |
3132 | | */ |
3133 | | static IndexClause * |
3134 | | match_saopclause_to_indexcol(PlannerInfo *root, |
3135 | | RestrictInfo *rinfo, |
3136 | | int indexcol, |
3137 | | IndexOptInfo *index) |
3138 | 0 | { |
3139 | 0 | ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) rinfo->clause; |
3140 | 0 | Node *leftop, |
3141 | 0 | *rightop; |
3142 | 0 | Relids right_relids; |
3143 | 0 | Oid expr_op; |
3144 | 0 | Oid expr_coll; |
3145 | 0 | Index index_relid; |
3146 | 0 | Oid opfamily; |
3147 | 0 | Oid idxcollation; |
3148 | | |
3149 | | /* We only accept ANY clauses, not ALL */ |
3150 | 0 | if (!saop->useOr) |
3151 | 0 | return NULL; |
3152 | 0 | leftop = (Node *) linitial(saop->args); |
3153 | 0 | rightop = (Node *) lsecond(saop->args); |
3154 | 0 | right_relids = pull_varnos(root, rightop); |
3155 | 0 | expr_op = saop->opno; |
3156 | 0 | expr_coll = saop->inputcollid; |
3157 | |
|
3158 | 0 | index_relid = index->rel->relid; |
3159 | 0 | opfamily = index->opfamily[indexcol]; |
3160 | 0 | idxcollation = index->indexcollations[indexcol]; |
3161 | | |
3162 | | /* |
3163 | | * We must have indexkey on the left and a pseudo-constant array argument. |
3164 | | */ |
3165 | 0 | if (match_index_to_operand(leftop, indexcol, index) && |
3166 | 0 | !bms_is_member(index_relid, right_relids) && |
3167 | 0 | !contain_volatile_functions(rightop)) |
3168 | 0 | { |
3169 | 0 | if (IndexCollMatchesExprColl(idxcollation, expr_coll) && |
3170 | 0 | op_in_opfamily(expr_op, opfamily)) |
3171 | 0 | { |
3172 | 0 | IndexClause *iclause = makeNode(IndexClause); |
3173 | |
|
3174 | 0 | iclause->rinfo = rinfo; |
3175 | 0 | iclause->indexquals = list_make1(rinfo); |
3176 | 0 | iclause->lossy = false; |
3177 | 0 | iclause->indexcol = indexcol; |
3178 | 0 | iclause->indexcols = NIL; |
3179 | 0 | return iclause; |
3180 | 0 | } |
3181 | | |
3182 | | /* |
3183 | | * We do not currently ask support functions about ScalarArrayOpExprs, |
3184 | | * though in principle we could. |
3185 | | */ |
3186 | 0 | } |
3187 | | |
3188 | 0 | return NULL; |
3189 | 0 | } |
3190 | | |
3191 | | /* |
3192 | | * match_rowcompare_to_indexcol() |
3193 | | * Handles the RowCompareExpr case for match_clause_to_indexcol(), |
3194 | | * which see for comments. |
3195 | | * |
3196 | | * In this routine we check whether the first column of the row comparison |
3197 | | * matches the target index column. This is sufficient to guarantee that some |
3198 | | * index condition can be constructed from the RowCompareExpr --- the rest |
3199 | | * is handled by expand_indexqual_rowcompare(). |
3200 | | */ |
3201 | | static IndexClause * |
3202 | | match_rowcompare_to_indexcol(PlannerInfo *root, |
3203 | | RestrictInfo *rinfo, |
3204 | | int indexcol, |
3205 | | IndexOptInfo *index) |
3206 | 0 | { |
3207 | 0 | RowCompareExpr *clause = (RowCompareExpr *) rinfo->clause; |
3208 | 0 | Index index_relid; |
3209 | 0 | Oid opfamily; |
3210 | 0 | Oid idxcollation; |
3211 | 0 | Node *leftop, |
3212 | 0 | *rightop; |
3213 | 0 | bool var_on_left; |
3214 | 0 | Oid expr_op; |
3215 | 0 | Oid expr_coll; |
3216 | | |
3217 | | /* Forget it if we're not dealing with a btree index */ |
3218 | 0 | if (index->relam != BTREE_AM_OID) |
3219 | 0 | return NULL; |
3220 | | |
3221 | 0 | index_relid = index->rel->relid; |
3222 | 0 | opfamily = index->opfamily[indexcol]; |
3223 | 0 | idxcollation = index->indexcollations[indexcol]; |
3224 | | |
3225 | | /* |
3226 | | * We could do the matching on the basis of insisting that the opfamily |
3227 | | * shown in the RowCompareExpr be the same as the index column's opfamily, |
3228 | | * but that could fail in the presence of reverse-sort opfamilies: it'd be |
3229 | | * a matter of chance whether RowCompareExpr had picked the forward or |
3230 | | * reverse-sort family. So look only at the operator, and match if it is |
3231 | | * a member of the index's opfamily (after commutation, if the indexkey is |
3232 | | * on the right). We'll worry later about whether any additional |
3233 | | * operators are matchable to the index. |
3234 | | */ |
3235 | 0 | leftop = (Node *) linitial(clause->largs); |
3236 | 0 | rightop = (Node *) linitial(clause->rargs); |
3237 | 0 | expr_op = linitial_oid(clause->opnos); |
3238 | 0 | expr_coll = linitial_oid(clause->inputcollids); |
3239 | | |
3240 | | /* Collations must match, if relevant */ |
3241 | 0 | if (!IndexCollMatchesExprColl(idxcollation, expr_coll)) |
3242 | 0 | return NULL; |
3243 | | |
3244 | | /* |
3245 | | * These syntactic tests are the same as in match_opclause_to_indexcol() |
3246 | | */ |
3247 | 0 | if (match_index_to_operand(leftop, indexcol, index) && |
3248 | 0 | !bms_is_member(index_relid, pull_varnos(root, rightop)) && |
3249 | 0 | !contain_volatile_functions(rightop)) |
3250 | 0 | { |
3251 | | /* OK, indexkey is on left */ |
3252 | 0 | var_on_left = true; |
3253 | 0 | } |
3254 | 0 | else if (match_index_to_operand(rightop, indexcol, index) && |
3255 | 0 | !bms_is_member(index_relid, pull_varnos(root, leftop)) && |
3256 | 0 | !contain_volatile_functions(leftop)) |
3257 | 0 | { |
3258 | | /* indexkey is on right, so commute the operator */ |
3259 | 0 | expr_op = get_commutator(expr_op); |
3260 | 0 | if (expr_op == InvalidOid) |
3261 | 0 | return NULL; |
3262 | 0 | var_on_left = false; |
3263 | 0 | } |
3264 | 0 | else |
3265 | 0 | return NULL; |
3266 | | |
3267 | | /* We're good if the operator is the right type of opfamily member */ |
3268 | 0 | switch (get_op_opfamily_strategy(expr_op, opfamily)) |
3269 | 0 | { |
3270 | 0 | case BTLessStrategyNumber: |
3271 | 0 | case BTLessEqualStrategyNumber: |
3272 | 0 | case BTGreaterEqualStrategyNumber: |
3273 | 0 | case BTGreaterStrategyNumber: |
3274 | 0 | return expand_indexqual_rowcompare(root, |
3275 | 0 | rinfo, |
3276 | 0 | indexcol, |
3277 | 0 | index, |
3278 | 0 | expr_op, |
3279 | 0 | var_on_left); |
3280 | 0 | } |
3281 | | |
3282 | 0 | return NULL; |
3283 | 0 | } |
3284 | | |
3285 | | /* |
3286 | | * match_orclause_to_indexcol() |
3287 | | * Handles the OR-expr case for match_clause_to_indexcol() in the case |
3288 | | * when it could be transformed to ScalarArrayOpExpr. |
3289 | | * |
3290 | | * In this routine, we attempt to transform a list of OR-clause args into a |
3291 | | * single SAOP expression matching the target index column. On success, |
3292 | | * return an IndexClause, containing the transformed expression or NULL, |
3293 | | * if failed. |
3294 | | */ |
3295 | | static IndexClause * |
3296 | | match_orclause_to_indexcol(PlannerInfo *root, |
3297 | | RestrictInfo *rinfo, |
3298 | | int indexcol, |
3299 | | IndexOptInfo *index) |
3300 | 0 | { |
3301 | 0 | ListCell *lc; |
3302 | 0 | BoolExpr *orclause = (BoolExpr *) rinfo->orclause; |
3303 | 0 | Node *indexExpr = NULL; |
3304 | 0 | List *consts = NIL; |
3305 | 0 | ScalarArrayOpExpr *saopexpr = NULL; |
3306 | 0 | Oid matchOpno = InvalidOid; |
3307 | 0 | IndexClause *iclause; |
3308 | 0 | Oid consttype = InvalidOid; |
3309 | 0 | Oid arraytype = InvalidOid; |
3310 | 0 | Oid inputcollid = InvalidOid; |
3311 | 0 | bool firstTime = true; |
3312 | 0 | bool haveNonConst = false; |
3313 | 0 | Index indexRelid = index->rel->relid; |
3314 | |
|
3315 | 0 | Assert(IsA(orclause, BoolExpr)); |
3316 | 0 | Assert(orclause->boolop == OR_EXPR); |
3317 | | |
3318 | | /* Ignore index if it doesn't support SAOP clauses */ |
3319 | 0 | if (!index->amsearcharray) |
3320 | 0 | return NULL; |
3321 | | |
3322 | | /* |
3323 | | * Try to convert a list of OR-clauses to a single SAOP expression. Each |
3324 | | * OR entry must be in the form: (indexkey operator constant) or (constant |
3325 | | * operator indexkey). Operators of all the entries must match. To be |
3326 | | * effective, give up on the first non-matching entry. Exit is |
3327 | | * implemented as a break from the loop, which is catched afterwards. |
3328 | | */ |
3329 | 0 | foreach(lc, orclause->args) |
3330 | 0 | { |
3331 | 0 | RestrictInfo *subRinfo; |
3332 | 0 | OpExpr *subClause; |
3333 | 0 | Oid opno; |
3334 | 0 | Node *leftop, |
3335 | 0 | *rightop; |
3336 | 0 | Node *constExpr; |
3337 | |
|
3338 | 0 | if (!IsA(lfirst(lc), RestrictInfo)) |
3339 | 0 | break; |
3340 | | |
3341 | 0 | subRinfo = (RestrictInfo *) lfirst(lc); |
3342 | | |
3343 | | /* Only operator clauses can match */ |
3344 | 0 | if (!IsA(subRinfo->clause, OpExpr)) |
3345 | 0 | break; |
3346 | | |
3347 | 0 | subClause = (OpExpr *) subRinfo->clause; |
3348 | 0 | opno = subClause->opno; |
3349 | | |
3350 | | /* Only binary operators can match */ |
3351 | 0 | if (list_length(subClause->args) != 2) |
3352 | 0 | break; |
3353 | | |
3354 | | /* |
3355 | | * The parameters below must match between sub-rinfo and its parent as |
3356 | | * make_restrictinfo() fills them with the same values, and further |
3357 | | * modifications are also the same for the whole subtree. However, |
3358 | | * still make a sanity check. |
3359 | | */ |
3360 | 0 | Assert(subRinfo->is_pushed_down == rinfo->is_pushed_down); |
3361 | 0 | Assert(subRinfo->is_clone == rinfo->is_clone); |
3362 | 0 | Assert(subRinfo->security_level == rinfo->security_level); |
3363 | 0 | Assert(bms_equal(subRinfo->incompatible_relids, rinfo->incompatible_relids)); |
3364 | 0 | Assert(bms_equal(subRinfo->outer_relids, rinfo->outer_relids)); |
3365 | | |
3366 | | /* |
3367 | | * Also, check that required_relids in sub-rinfo is subset of parent's |
3368 | | * required_relids. |
3369 | | */ |
3370 | 0 | Assert(bms_is_subset(subRinfo->required_relids, rinfo->required_relids)); |
3371 | | |
3372 | | /* Only the operator returning a boolean suit the transformation. */ |
3373 | 0 | if (get_op_rettype(opno) != BOOLOID) |
3374 | 0 | break; |
3375 | | |
3376 | | /* |
3377 | | * Check for clauses of the form: (indexkey operator constant) or |
3378 | | * (constant operator indexkey). See match_clause_to_indexcol's notes |
3379 | | * about const-ness. |
3380 | | */ |
3381 | 0 | leftop = (Node *) linitial(subClause->args); |
3382 | 0 | rightop = (Node *) lsecond(subClause->args); |
3383 | 0 | if (match_index_to_operand(leftop, indexcol, index) && |
3384 | 0 | !bms_is_member(indexRelid, subRinfo->right_relids) && |
3385 | 0 | !contain_volatile_functions(rightop)) |
3386 | 0 | { |
3387 | 0 | indexExpr = leftop; |
3388 | 0 | constExpr = rightop; |
3389 | 0 | } |
3390 | 0 | else if (match_index_to_operand(rightop, indexcol, index) && |
3391 | 0 | !bms_is_member(indexRelid, subRinfo->left_relids) && |
3392 | 0 | !contain_volatile_functions(leftop)) |
3393 | 0 | { |
3394 | 0 | opno = get_commutator(opno); |
3395 | 0 | if (!OidIsValid(opno)) |
3396 | 0 | { |
3397 | | /* commutator doesn't exist, we can't reverse the order */ |
3398 | 0 | break; |
3399 | 0 | } |
3400 | 0 | indexExpr = rightop; |
3401 | 0 | constExpr = leftop; |
3402 | 0 | } |
3403 | 0 | else |
3404 | 0 | { |
3405 | 0 | break; |
3406 | 0 | } |
3407 | | |
3408 | | /* |
3409 | | * Ignore any RelabelType node above the operands. This is needed to |
3410 | | * be able to apply indexscanning in binary-compatible-operator cases. |
3411 | | * Note: we can assume there is at most one RelabelType node; |
3412 | | * eval_const_expressions() will have simplified if more than one. |
3413 | | */ |
3414 | 0 | if (IsA(constExpr, RelabelType)) |
3415 | 0 | constExpr = (Node *) ((RelabelType *) constExpr)->arg; |
3416 | 0 | if (IsA(indexExpr, RelabelType)) |
3417 | 0 | indexExpr = (Node *) ((RelabelType *) indexExpr)->arg; |
3418 | | |
3419 | | /* Forbid transformation for composite types, records. */ |
3420 | 0 | if (type_is_rowtype(exprType(constExpr)) || |
3421 | 0 | type_is_rowtype(exprType(indexExpr))) |
3422 | 0 | break; |
3423 | | |
3424 | | /* |
3425 | | * Save information about the operator, type, and collation for the |
3426 | | * first matching qual. Then, check that subsequent quals match the |
3427 | | * first. |
3428 | | */ |
3429 | 0 | if (firstTime) |
3430 | 0 | { |
3431 | 0 | matchOpno = opno; |
3432 | 0 | consttype = exprType(constExpr); |
3433 | 0 | arraytype = get_array_type(consttype); |
3434 | 0 | inputcollid = subClause->inputcollid; |
3435 | | |
3436 | | /* |
3437 | | * Check that the operator is presented in the opfamily and that |
3438 | | * the expression collation matches the index collation. Also, |
3439 | | * there must be an array type to construct an array later. |
3440 | | */ |
3441 | 0 | if (!IndexCollMatchesExprColl(index->indexcollations[indexcol], inputcollid) || |
3442 | 0 | !op_in_opfamily(matchOpno, index->opfamily[indexcol]) || |
3443 | 0 | !OidIsValid(arraytype)) |
3444 | 0 | break; |
3445 | 0 | firstTime = false; |
3446 | 0 | } |
3447 | 0 | else |
3448 | 0 | { |
3449 | 0 | if (opno != matchOpno || |
3450 | 0 | inputcollid != subClause->inputcollid || |
3451 | 0 | consttype != exprType(constExpr)) |
3452 | 0 | break; |
3453 | 0 | } |
3454 | | |
3455 | | /* |
3456 | | * Check if our list of constants in match_clause_to_indexcol's |
3457 | | * understanding of const-ness have something other than Const. |
3458 | | */ |
3459 | 0 | if (!IsA(constExpr, Const)) |
3460 | 0 | haveNonConst = true; |
3461 | 0 | consts = lappend(consts, constExpr); |
3462 | 0 | } |
3463 | | |
3464 | | /* |
3465 | | * Catch the break from the loop above. Normally, a foreach() loop ends |
3466 | | * up with a NULL list cell. A non-NULL list cell indicates a break from |
3467 | | * the foreach() loop. Free the consts list and return NULL then. |
3468 | | */ |
3469 | 0 | if (lc != NULL) |
3470 | 0 | { |
3471 | 0 | list_free(consts); |
3472 | 0 | return NULL; |
3473 | 0 | } |
3474 | | |
3475 | 0 | saopexpr = make_SAOP_expr(matchOpno, indexExpr, consttype, inputcollid, |
3476 | 0 | inputcollid, consts, haveNonConst); |
3477 | | |
3478 | | /* |
3479 | | * Finally, build an IndexClause based on the SAOP node. Use |
3480 | | * make_simple_restrictinfo() to get RestrictInfo with clean selectivity |
3481 | | * estimations, because they may differ from the estimation made for an OR |
3482 | | * clause. Although it is not a lossy expression, keep the original rinfo |
3483 | | * in iclause->rinfo as prescribed. |
3484 | | */ |
3485 | 0 | iclause = makeNode(IndexClause); |
3486 | 0 | iclause->rinfo = rinfo; |
3487 | 0 | iclause->indexquals = list_make1(make_simple_restrictinfo(root, |
3488 | 0 | &saopexpr->xpr)); |
3489 | 0 | iclause->lossy = false; |
3490 | 0 | iclause->indexcol = indexcol; |
3491 | 0 | iclause->indexcols = NIL; |
3492 | 0 | return iclause; |
3493 | 0 | } |
3494 | | |
3495 | | /* |
3496 | | * expand_indexqual_rowcompare --- expand a single indexqual condition |
3497 | | * that is a RowCompareExpr |
3498 | | * |
3499 | | * It's already known that the first column of the row comparison matches |
3500 | | * the specified column of the index. We can use additional columns of the |
3501 | | * row comparison as index qualifications, so long as they match the index |
3502 | | * in the "same direction", ie, the indexkeys are all on the same side of the |
3503 | | * clause and the operators are all the same-type members of the opfamilies. |
3504 | | * |
3505 | | * If all the columns of the RowCompareExpr match in this way, we just use it |
3506 | | * as-is, except for possibly commuting it to put the indexkeys on the left. |
3507 | | * |
3508 | | * Otherwise, we build a shortened RowCompareExpr (if more than one |
3509 | | * column matches) or a simple OpExpr (if the first-column match is all |
3510 | | * there is). In these cases the modified clause is always "<=" or ">=" |
3511 | | * even when the original was "<" or ">" --- this is necessary to match all |
3512 | | * the rows that could match the original. (We are building a lossy version |
3513 | | * of the row comparison when we do this, so we set lossy = true.) |
3514 | | * |
3515 | | * Note: this is really just the last half of match_rowcompare_to_indexcol, |
3516 | | * but we split it out for comprehensibility. |
3517 | | */ |
3518 | | static IndexClause * |
3519 | | expand_indexqual_rowcompare(PlannerInfo *root, |
3520 | | RestrictInfo *rinfo, |
3521 | | int indexcol, |
3522 | | IndexOptInfo *index, |
3523 | | Oid expr_op, |
3524 | | bool var_on_left) |
3525 | 0 | { |
3526 | 0 | IndexClause *iclause = makeNode(IndexClause); |
3527 | 0 | RowCompareExpr *clause = (RowCompareExpr *) rinfo->clause; |
3528 | 0 | int op_strategy; |
3529 | 0 | Oid op_lefttype; |
3530 | 0 | Oid op_righttype; |
3531 | 0 | int matching_cols; |
3532 | 0 | List *expr_ops; |
3533 | 0 | List *opfamilies; |
3534 | 0 | List *lefttypes; |
3535 | 0 | List *righttypes; |
3536 | 0 | List *new_ops; |
3537 | 0 | List *var_args; |
3538 | 0 | List *non_var_args; |
3539 | |
|
3540 | 0 | iclause->rinfo = rinfo; |
3541 | 0 | iclause->indexcol = indexcol; |
3542 | |
|
3543 | 0 | if (var_on_left) |
3544 | 0 | { |
3545 | 0 | var_args = clause->largs; |
3546 | 0 | non_var_args = clause->rargs; |
3547 | 0 | } |
3548 | 0 | else |
3549 | 0 | { |
3550 | 0 | var_args = clause->rargs; |
3551 | 0 | non_var_args = clause->largs; |
3552 | 0 | } |
3553 | |
|
3554 | 0 | get_op_opfamily_properties(expr_op, index->opfamily[indexcol], false, |
3555 | 0 | &op_strategy, |
3556 | 0 | &op_lefttype, |
3557 | 0 | &op_righttype); |
3558 | | |
3559 | | /* Initialize returned list of which index columns are used */ |
3560 | 0 | iclause->indexcols = list_make1_int(indexcol); |
3561 | | |
3562 | | /* Build lists of ops, opfamilies and operator datatypes in case needed */ |
3563 | 0 | expr_ops = list_make1_oid(expr_op); |
3564 | 0 | opfamilies = list_make1_oid(index->opfamily[indexcol]); |
3565 | 0 | lefttypes = list_make1_oid(op_lefttype); |
3566 | 0 | righttypes = list_make1_oid(op_righttype); |
3567 | | |
3568 | | /* |
3569 | | * See how many of the remaining columns match some index column in the |
3570 | | * same way. As in match_clause_to_indexcol(), the "other" side of any |
3571 | | * potential index condition is OK as long as it doesn't use Vars from the |
3572 | | * indexed relation. |
3573 | | */ |
3574 | 0 | matching_cols = 1; |
3575 | |
|
3576 | 0 | while (matching_cols < list_length(var_args)) |
3577 | 0 | { |
3578 | 0 | Node *varop = (Node *) list_nth(var_args, matching_cols); |
3579 | 0 | Node *constop = (Node *) list_nth(non_var_args, matching_cols); |
3580 | 0 | int i; |
3581 | |
|
3582 | 0 | expr_op = list_nth_oid(clause->opnos, matching_cols); |
3583 | 0 | if (!var_on_left) |
3584 | 0 | { |
3585 | | /* indexkey is on right, so commute the operator */ |
3586 | 0 | expr_op = get_commutator(expr_op); |
3587 | 0 | if (expr_op == InvalidOid) |
3588 | 0 | break; /* operator is not usable */ |
3589 | 0 | } |
3590 | 0 | if (bms_is_member(index->rel->relid, pull_varnos(root, constop))) |
3591 | 0 | break; /* no good, Var on wrong side */ |
3592 | 0 | if (contain_volatile_functions(constop)) |
3593 | 0 | break; /* no good, volatile comparison value */ |
3594 | | |
3595 | | /* |
3596 | | * The Var side can match any key column of the index. |
3597 | | */ |
3598 | 0 | for (i = 0; i < index->nkeycolumns; i++) |
3599 | 0 | { |
3600 | 0 | if (match_index_to_operand(varop, i, index) && |
3601 | 0 | get_op_opfamily_strategy(expr_op, |
3602 | 0 | index->opfamily[i]) == op_strategy && |
3603 | 0 | IndexCollMatchesExprColl(index->indexcollations[i], |
3604 | 0 | list_nth_oid(clause->inputcollids, |
3605 | 0 | matching_cols))) |
3606 | 0 | break; |
3607 | 0 | } |
3608 | 0 | if (i >= index->nkeycolumns) |
3609 | 0 | break; /* no match found */ |
3610 | | |
3611 | | /* Add column number to returned list */ |
3612 | 0 | iclause->indexcols = lappend_int(iclause->indexcols, i); |
3613 | | |
3614 | | /* Add operator info to lists */ |
3615 | 0 | get_op_opfamily_properties(expr_op, index->opfamily[i], false, |
3616 | 0 | &op_strategy, |
3617 | 0 | &op_lefttype, |
3618 | 0 | &op_righttype); |
3619 | 0 | expr_ops = lappend_oid(expr_ops, expr_op); |
3620 | 0 | opfamilies = lappend_oid(opfamilies, index->opfamily[i]); |
3621 | 0 | lefttypes = lappend_oid(lefttypes, op_lefttype); |
3622 | 0 | righttypes = lappend_oid(righttypes, op_righttype); |
3623 | | |
3624 | | /* This column matches, keep scanning */ |
3625 | 0 | matching_cols++; |
3626 | 0 | } |
3627 | | |
3628 | | /* Result is non-lossy if all columns are usable as index quals */ |
3629 | 0 | iclause->lossy = (matching_cols != list_length(clause->opnos)); |
3630 | | |
3631 | | /* |
3632 | | * We can use rinfo->clause as-is if we have var on left and it's all |
3633 | | * usable as index quals. |
3634 | | */ |
3635 | 0 | if (var_on_left && !iclause->lossy) |
3636 | 0 | iclause->indexquals = list_make1(rinfo); |
3637 | 0 | else |
3638 | 0 | { |
3639 | | /* |
3640 | | * We have to generate a modified rowcompare (possibly just one |
3641 | | * OpExpr). The painful part of this is changing < to <= or > to >=, |
3642 | | * so deal with that first. |
3643 | | */ |
3644 | 0 | if (!iclause->lossy) |
3645 | 0 | { |
3646 | | /* very easy, just use the commuted operators */ |
3647 | 0 | new_ops = expr_ops; |
3648 | 0 | } |
3649 | 0 | else if (op_strategy == BTLessEqualStrategyNumber || |
3650 | 0 | op_strategy == BTGreaterEqualStrategyNumber) |
3651 | 0 | { |
3652 | | /* easy, just use the same (possibly commuted) operators */ |
3653 | 0 | new_ops = list_truncate(expr_ops, matching_cols); |
3654 | 0 | } |
3655 | 0 | else |
3656 | 0 | { |
3657 | 0 | ListCell *opfamilies_cell; |
3658 | 0 | ListCell *lefttypes_cell; |
3659 | 0 | ListCell *righttypes_cell; |
3660 | |
|
3661 | 0 | if (op_strategy == BTLessStrategyNumber) |
3662 | 0 | op_strategy = BTLessEqualStrategyNumber; |
3663 | 0 | else if (op_strategy == BTGreaterStrategyNumber) |
3664 | 0 | op_strategy = BTGreaterEqualStrategyNumber; |
3665 | 0 | else |
3666 | 0 | elog(ERROR, "unexpected strategy number %d", op_strategy); |
3667 | 0 | new_ops = NIL; |
3668 | 0 | forthree(opfamilies_cell, opfamilies, |
3669 | 0 | lefttypes_cell, lefttypes, |
3670 | 0 | righttypes_cell, righttypes) |
3671 | 0 | { |
3672 | 0 | Oid opfam = lfirst_oid(opfamilies_cell); |
3673 | 0 | Oid lefttype = lfirst_oid(lefttypes_cell); |
3674 | 0 | Oid righttype = lfirst_oid(righttypes_cell); |
3675 | |
|
3676 | 0 | expr_op = get_opfamily_member(opfam, lefttype, righttype, |
3677 | 0 | op_strategy); |
3678 | 0 | if (!OidIsValid(expr_op)) /* should not happen */ |
3679 | 0 | elog(ERROR, "missing operator %d(%u,%u) in opfamily %u", |
3680 | 0 | op_strategy, lefttype, righttype, opfam); |
3681 | 0 | new_ops = lappend_oid(new_ops, expr_op); |
3682 | 0 | } |
3683 | 0 | } |
3684 | | |
3685 | | /* If we have more than one matching col, create a subset rowcompare */ |
3686 | 0 | if (matching_cols > 1) |
3687 | 0 | { |
3688 | 0 | RowCompareExpr *rc = makeNode(RowCompareExpr); |
3689 | |
|
3690 | 0 | rc->cmptype = (CompareType) op_strategy; |
3691 | 0 | rc->opnos = new_ops; |
3692 | 0 | rc->opfamilies = list_copy_head(clause->opfamilies, |
3693 | 0 | matching_cols); |
3694 | 0 | rc->inputcollids = list_copy_head(clause->inputcollids, |
3695 | 0 | matching_cols); |
3696 | 0 | rc->largs = list_copy_head(var_args, matching_cols); |
3697 | 0 | rc->rargs = list_copy_head(non_var_args, matching_cols); |
3698 | 0 | iclause->indexquals = list_make1(make_simple_restrictinfo(root, |
3699 | 0 | (Expr *) rc)); |
3700 | 0 | } |
3701 | 0 | else |
3702 | 0 | { |
3703 | 0 | Expr *op; |
3704 | | |
3705 | | /* We don't report an index column list in this case */ |
3706 | 0 | iclause->indexcols = NIL; |
3707 | |
|
3708 | 0 | op = make_opclause(linitial_oid(new_ops), BOOLOID, false, |
3709 | 0 | copyObject(linitial(var_args)), |
3710 | 0 | copyObject(linitial(non_var_args)), |
3711 | 0 | InvalidOid, |
3712 | 0 | linitial_oid(clause->inputcollids)); |
3713 | 0 | iclause->indexquals = list_make1(make_simple_restrictinfo(root, op)); |
3714 | 0 | } |
3715 | 0 | } |
3716 | | |
3717 | 0 | return iclause; |
3718 | 0 | } |
3719 | | |
3720 | | |
3721 | | /**************************************************************************** |
3722 | | * ---- ROUTINES TO CHECK ORDERING OPERATORS ---- |
3723 | | ****************************************************************************/ |
3724 | | |
3725 | | /* |
3726 | | * match_pathkeys_to_index |
3727 | | * For the given 'index' and 'pathkeys', output a list of suitable ORDER |
3728 | | * BY expressions, each of the form "indexedcol operator pseudoconstant", |
3729 | | * along with an integer list of the index column numbers (zero based) |
3730 | | * that each clause would be used with. |
3731 | | * |
3732 | | * This attempts to find an ORDER BY and index column number for all items in |
3733 | | * the pathkey list, however, if we're unable to match any given pathkey to an |
3734 | | * index column, we return just the ones matched by the function so far. This |
3735 | | * allows callers who are interested in partial matches to get them. Callers |
3736 | | * can determine a partial match vs a full match by checking the outputted |
3737 | | * list lengths. A full match will have one item in the output lists for each |
3738 | | * item in the given 'pathkeys' list. |
3739 | | */ |
3740 | | static void |
3741 | | match_pathkeys_to_index(IndexOptInfo *index, List *pathkeys, |
3742 | | List **orderby_clauses_p, |
3743 | | List **clause_columns_p) |
3744 | 0 | { |
3745 | 0 | ListCell *lc1; |
3746 | |
|
3747 | 0 | *orderby_clauses_p = NIL; /* set default results */ |
3748 | 0 | *clause_columns_p = NIL; |
3749 | | |
3750 | | /* Only indexes with the amcanorderbyop property are interesting here */ |
3751 | 0 | if (!index->amcanorderbyop) |
3752 | 0 | return; |
3753 | | |
3754 | 0 | foreach(lc1, pathkeys) |
3755 | 0 | { |
3756 | 0 | PathKey *pathkey = (PathKey *) lfirst(lc1); |
3757 | 0 | bool found = false; |
3758 | 0 | EquivalenceMemberIterator it; |
3759 | 0 | EquivalenceMember *member; |
3760 | | |
3761 | | |
3762 | | /* Pathkey must request default sort order for the target opfamily */ |
3763 | 0 | if (pathkey->pk_cmptype != COMPARE_LT || pathkey->pk_nulls_first) |
3764 | 0 | return; |
3765 | | |
3766 | | /* If eclass is volatile, no hope of using an indexscan */ |
3767 | 0 | if (pathkey->pk_eclass->ec_has_volatile) |
3768 | 0 | return; |
3769 | | |
3770 | | /* |
3771 | | * Try to match eclass member expression(s) to index. Note that child |
3772 | | * EC members are considered, but only when they belong to the target |
3773 | | * relation. (Unlike regular members, the same expression could be a |
3774 | | * child member of more than one EC. Therefore, the same index could |
3775 | | * be considered to match more than one pathkey list, which is OK |
3776 | | * here. See also get_eclass_for_sort_expr.) |
3777 | | */ |
3778 | 0 | setup_eclass_member_iterator(&it, pathkey->pk_eclass, |
3779 | 0 | index->rel->relids); |
3780 | 0 | while ((member = eclass_member_iterator_next(&it)) != NULL) |
3781 | 0 | { |
3782 | 0 | int indexcol; |
3783 | | |
3784 | | /* No possibility of match if it references other relations */ |
3785 | 0 | if (!bms_equal(member->em_relids, index->rel->relids)) |
3786 | 0 | continue; |
3787 | | |
3788 | | /* |
3789 | | * We allow any column of the index to match each pathkey; they |
3790 | | * don't have to match left-to-right as you might expect. This is |
3791 | | * correct for GiST, and it doesn't matter for SP-GiST because |
3792 | | * that doesn't handle multiple columns anyway, and no other |
3793 | | * existing AMs support amcanorderbyop. We might need different |
3794 | | * logic in future for other implementations. |
3795 | | */ |
3796 | 0 | for (indexcol = 0; indexcol < index->nkeycolumns; indexcol++) |
3797 | 0 | { |
3798 | 0 | Expr *expr; |
3799 | |
|
3800 | 0 | expr = match_clause_to_ordering_op(index, |
3801 | 0 | indexcol, |
3802 | 0 | member->em_expr, |
3803 | 0 | pathkey->pk_opfamily); |
3804 | 0 | if (expr) |
3805 | 0 | { |
3806 | 0 | *orderby_clauses_p = lappend(*orderby_clauses_p, expr); |
3807 | 0 | *clause_columns_p = lappend_int(*clause_columns_p, indexcol); |
3808 | 0 | found = true; |
3809 | 0 | break; |
3810 | 0 | } |
3811 | 0 | } |
3812 | |
|
3813 | 0 | if (found) /* don't want to look at remaining members */ |
3814 | 0 | break; |
3815 | 0 | } |
3816 | | |
3817 | | /* |
3818 | | * Return the matches found so far when this pathkey couldn't be |
3819 | | * matched to the index. |
3820 | | */ |
3821 | 0 | if (!found) |
3822 | 0 | return; |
3823 | 0 | } |
3824 | 0 | } |
3825 | | |
3826 | | /* |
3827 | | * match_clause_to_ordering_op |
3828 | | * Determines whether an ordering operator expression matches an |
3829 | | * index column. |
3830 | | * |
3831 | | * This is similar to, but simpler than, match_clause_to_indexcol. |
3832 | | * We only care about simple OpExpr cases. The input is a bare |
3833 | | * expression that is being ordered by, which must be of the form |
3834 | | * (indexkey op const) or (const op indexkey) where op is an ordering |
3835 | | * operator for the column's opfamily. |
3836 | | * |
3837 | | * 'index' is the index of interest. |
3838 | | * 'indexcol' is a column number of 'index' (counting from 0). |
3839 | | * 'clause' is the ordering expression to be tested. |
3840 | | * 'pk_opfamily' is the btree opfamily describing the required sort order. |
3841 | | * |
3842 | | * Note that we currently do not consider the collation of the ordering |
3843 | | * operator's result. In practical cases the result type will be numeric |
3844 | | * and thus have no collation, and it's not very clear what to match to |
3845 | | * if it did have a collation. The index's collation should match the |
3846 | | * ordering operator's input collation, not its result. |
3847 | | * |
3848 | | * If successful, return 'clause' as-is if the indexkey is on the left, |
3849 | | * otherwise a commuted copy of 'clause'. If no match, return NULL. |
3850 | | */ |
3851 | | static Expr * |
3852 | | match_clause_to_ordering_op(IndexOptInfo *index, |
3853 | | int indexcol, |
3854 | | Expr *clause, |
3855 | | Oid pk_opfamily) |
3856 | 0 | { |
3857 | 0 | Oid opfamily; |
3858 | 0 | Oid idxcollation; |
3859 | 0 | Node *leftop, |
3860 | 0 | *rightop; |
3861 | 0 | Oid expr_op; |
3862 | 0 | Oid expr_coll; |
3863 | 0 | Oid sortfamily; |
3864 | 0 | bool commuted; |
3865 | |
|
3866 | 0 | Assert(indexcol < index->nkeycolumns); |
3867 | |
|
3868 | 0 | opfamily = index->opfamily[indexcol]; |
3869 | 0 | idxcollation = index->indexcollations[indexcol]; |
3870 | | |
3871 | | /* |
3872 | | * Clause must be a binary opclause. |
3873 | | */ |
3874 | 0 | if (!is_opclause(clause)) |
3875 | 0 | return NULL; |
3876 | 0 | leftop = get_leftop(clause); |
3877 | 0 | rightop = get_rightop(clause); |
3878 | 0 | if (!leftop || !rightop) |
3879 | 0 | return NULL; |
3880 | 0 | expr_op = ((OpExpr *) clause)->opno; |
3881 | 0 | expr_coll = ((OpExpr *) clause)->inputcollid; |
3882 | | |
3883 | | /* |
3884 | | * We can forget the whole thing right away if wrong collation. |
3885 | | */ |
3886 | 0 | if (!IndexCollMatchesExprColl(idxcollation, expr_coll)) |
3887 | 0 | return NULL; |
3888 | | |
3889 | | /* |
3890 | | * Check for clauses of the form: (indexkey operator constant) or |
3891 | | * (constant operator indexkey). |
3892 | | */ |
3893 | 0 | if (match_index_to_operand(leftop, indexcol, index) && |
3894 | 0 | !contain_var_clause(rightop) && |
3895 | 0 | !contain_volatile_functions(rightop)) |
3896 | 0 | { |
3897 | 0 | commuted = false; |
3898 | 0 | } |
3899 | 0 | else if (match_index_to_operand(rightop, indexcol, index) && |
3900 | 0 | !contain_var_clause(leftop) && |
3901 | 0 | !contain_volatile_functions(leftop)) |
3902 | 0 | { |
3903 | | /* Might match, but we need a commuted operator */ |
3904 | 0 | expr_op = get_commutator(expr_op); |
3905 | 0 | if (expr_op == InvalidOid) |
3906 | 0 | return NULL; |
3907 | 0 | commuted = true; |
3908 | 0 | } |
3909 | 0 | else |
3910 | 0 | return NULL; |
3911 | | |
3912 | | /* |
3913 | | * Is the (commuted) operator an ordering operator for the opfamily? And |
3914 | | * if so, does it yield the right sorting semantics? |
3915 | | */ |
3916 | 0 | sortfamily = get_op_opfamily_sortfamily(expr_op, opfamily); |
3917 | 0 | if (sortfamily != pk_opfamily) |
3918 | 0 | return NULL; |
3919 | | |
3920 | | /* We have a match. Return clause or a commuted version thereof. */ |
3921 | 0 | if (commuted) |
3922 | 0 | { |
3923 | 0 | OpExpr *newclause = makeNode(OpExpr); |
3924 | | |
3925 | | /* flat-copy all the fields of clause */ |
3926 | 0 | memcpy(newclause, clause, sizeof(OpExpr)); |
3927 | | |
3928 | | /* commute it */ |
3929 | 0 | newclause->opno = expr_op; |
3930 | 0 | newclause->opfuncid = InvalidOid; |
3931 | 0 | newclause->args = list_make2(rightop, leftop); |
3932 | |
|
3933 | 0 | clause = (Expr *) newclause; |
3934 | 0 | } |
3935 | |
|
3936 | 0 | return clause; |
3937 | 0 | } |
3938 | | |
3939 | | |
3940 | | /**************************************************************************** |
3941 | | * ---- ROUTINES TO DO PARTIAL INDEX PREDICATE TESTS ---- |
3942 | | ****************************************************************************/ |
3943 | | |
3944 | | /* |
3945 | | * check_index_predicates |
3946 | | * Set the predicate-derived IndexOptInfo fields for each index |
3947 | | * of the specified relation. |
3948 | | * |
3949 | | * predOK is set true if the index is partial and its predicate is satisfied |
3950 | | * for this query, ie the query's WHERE clauses imply the predicate. |
3951 | | * |
3952 | | * indrestrictinfo is set to the relation's baserestrictinfo list less any |
3953 | | * conditions that are implied by the index's predicate. (Obviously, for a |
3954 | | * non-partial index, this is the same as baserestrictinfo.) Such conditions |
3955 | | * can be dropped from the plan when using the index, in certain cases. |
3956 | | * |
3957 | | * At one time it was possible for this to get re-run after adding more |
3958 | | * restrictions to the rel, thus possibly letting us prove more indexes OK. |
3959 | | * That doesn't happen any more (at least not in the core code's usage), |
3960 | | * but this code still supports it in case extensions want to mess with the |
3961 | | * baserestrictinfo list. We assume that adding more restrictions can't make |
3962 | | * an index not predOK. We must recompute indrestrictinfo each time, though, |
3963 | | * to make sure any newly-added restrictions get into it if needed. |
3964 | | */ |
3965 | | void |
3966 | | check_index_predicates(PlannerInfo *root, RelOptInfo *rel) |
3967 | 0 | { |
3968 | 0 | List *clauselist; |
3969 | 0 | bool have_partial; |
3970 | 0 | bool is_target_rel; |
3971 | 0 | Relids otherrels; |
3972 | 0 | ListCell *lc; |
3973 | | |
3974 | | /* Indexes are available only on base or "other" member relations. */ |
3975 | 0 | Assert(IS_SIMPLE_REL(rel)); |
3976 | | |
3977 | | /* |
3978 | | * Initialize the indrestrictinfo lists to be identical to |
3979 | | * baserestrictinfo, and check whether there are any partial indexes. If |
3980 | | * not, this is all we need to do. |
3981 | | */ |
3982 | 0 | have_partial = false; |
3983 | 0 | foreach(lc, rel->indexlist) |
3984 | 0 | { |
3985 | 0 | IndexOptInfo *index = (IndexOptInfo *) lfirst(lc); |
3986 | |
|
3987 | 0 | index->indrestrictinfo = rel->baserestrictinfo; |
3988 | 0 | if (index->indpred) |
3989 | 0 | have_partial = true; |
3990 | 0 | } |
3991 | 0 | if (!have_partial) |
3992 | 0 | return; |
3993 | | |
3994 | | /* |
3995 | | * Construct a list of clauses that we can assume true for the purpose of |
3996 | | * proving the index(es) usable. Restriction clauses for the rel are |
3997 | | * always usable, and so are any join clauses that are "movable to" this |
3998 | | * rel. Also, we can consider any EC-derivable join clauses (which must |
3999 | | * be "movable to" this rel, by definition). |
4000 | | */ |
4001 | 0 | clauselist = list_copy(rel->baserestrictinfo); |
4002 | | |
4003 | | /* Scan the rel's join clauses */ |
4004 | 0 | foreach(lc, rel->joininfo) |
4005 | 0 | { |
4006 | 0 | RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc); |
4007 | | |
4008 | | /* Check if clause can be moved to this rel */ |
4009 | 0 | if (!join_clause_is_movable_to(rinfo, rel)) |
4010 | 0 | continue; |
4011 | | |
4012 | 0 | clauselist = lappend(clauselist, rinfo); |
4013 | 0 | } |
4014 | | |
4015 | | /* |
4016 | | * Add on any equivalence-derivable join clauses. Computing the correct |
4017 | | * relid sets for generate_join_implied_equalities is slightly tricky |
4018 | | * because the rel could be a child rel rather than a true baserel, and in |
4019 | | * that case we must subtract its parents' relid(s) from all_query_rels. |
4020 | | * Additionally, we mustn't consider clauses that are only computable |
4021 | | * after outer joins that can null the rel. |
4022 | | */ |
4023 | 0 | if (rel->reloptkind == RELOPT_OTHER_MEMBER_REL) |
4024 | 0 | otherrels = bms_difference(root->all_query_rels, |
4025 | 0 | find_childrel_parents(root, rel)); |
4026 | 0 | else |
4027 | 0 | otherrels = bms_difference(root->all_query_rels, rel->relids); |
4028 | 0 | otherrels = bms_del_members(otherrels, rel->nulling_relids); |
4029 | |
|
4030 | 0 | if (!bms_is_empty(otherrels)) |
4031 | 0 | clauselist = |
4032 | 0 | list_concat(clauselist, |
4033 | 0 | generate_join_implied_equalities(root, |
4034 | 0 | bms_union(rel->relids, |
4035 | 0 | otherrels), |
4036 | 0 | otherrels, |
4037 | 0 | rel, |
4038 | 0 | NULL)); |
4039 | | |
4040 | | /* |
4041 | | * Normally we remove quals that are implied by a partial index's |
4042 | | * predicate from indrestrictinfo, indicating that they need not be |
4043 | | * checked explicitly by an indexscan plan using this index. However, if |
4044 | | * the rel is a target relation of UPDATE/DELETE/MERGE/SELECT FOR UPDATE, |
4045 | | * we cannot remove such quals from the plan, because they need to be in |
4046 | | * the plan so that they will be properly rechecked by EvalPlanQual |
4047 | | * testing. Some day we might want to remove such quals from the main |
4048 | | * plan anyway and pass them through to EvalPlanQual via a side channel; |
4049 | | * but for now, we just don't remove implied quals at all for target |
4050 | | * relations. |
4051 | | */ |
4052 | 0 | is_target_rel = (bms_is_member(rel->relid, root->all_result_relids) || |
4053 | 0 | get_plan_rowmark(root->rowMarks, rel->relid) != NULL); |
4054 | | |
4055 | | /* |
4056 | | * Now try to prove each index predicate true, and compute the |
4057 | | * indrestrictinfo lists for partial indexes. Note that we compute the |
4058 | | * indrestrictinfo list even for non-predOK indexes; this might seem |
4059 | | * wasteful, but we may be able to use such indexes in OR clauses, cf |
4060 | | * generate_bitmap_or_paths(). |
4061 | | */ |
4062 | 0 | foreach(lc, rel->indexlist) |
4063 | 0 | { |
4064 | 0 | IndexOptInfo *index = (IndexOptInfo *) lfirst(lc); |
4065 | 0 | ListCell *lcr; |
4066 | |
|
4067 | 0 | if (index->indpred == NIL) |
4068 | 0 | continue; /* ignore non-partial indexes here */ |
4069 | | |
4070 | 0 | if (!index->predOK) /* don't repeat work if already proven OK */ |
4071 | 0 | index->predOK = predicate_implied_by(index->indpred, clauselist, |
4072 | 0 | false); |
4073 | | |
4074 | | /* If rel is an update target, leave indrestrictinfo as set above */ |
4075 | 0 | if (is_target_rel) |
4076 | 0 | continue; |
4077 | | |
4078 | | /* Else compute indrestrictinfo as the non-implied quals */ |
4079 | 0 | index->indrestrictinfo = NIL; |
4080 | 0 | foreach(lcr, rel->baserestrictinfo) |
4081 | 0 | { |
4082 | 0 | RestrictInfo *rinfo = (RestrictInfo *) lfirst(lcr); |
4083 | | |
4084 | | /* predicate_implied_by() assumes first arg is immutable */ |
4085 | 0 | if (contain_mutable_functions((Node *) rinfo->clause) || |
4086 | 0 | !predicate_implied_by(list_make1(rinfo->clause), |
4087 | 0 | index->indpred, false)) |
4088 | 0 | index->indrestrictinfo = lappend(index->indrestrictinfo, rinfo); |
4089 | 0 | } |
4090 | 0 | } |
4091 | 0 | } |
4092 | | |
4093 | | /**************************************************************************** |
4094 | | * ---- ROUTINES TO CHECK EXTERNALLY-VISIBLE CONDITIONS ---- |
4095 | | ****************************************************************************/ |
4096 | | |
4097 | | /* |
4098 | | * ec_member_matches_indexcol |
4099 | | * Test whether an EquivalenceClass member matches an index column. |
4100 | | * |
4101 | | * This is a callback for use by generate_implied_equalities_for_column. |
4102 | | */ |
4103 | | static bool |
4104 | | ec_member_matches_indexcol(PlannerInfo *root, RelOptInfo *rel, |
4105 | | EquivalenceClass *ec, EquivalenceMember *em, |
4106 | | void *arg) |
4107 | 0 | { |
4108 | 0 | IndexOptInfo *index = ((ec_member_matches_arg *) arg)->index; |
4109 | 0 | int indexcol = ((ec_member_matches_arg *) arg)->indexcol; |
4110 | 0 | Oid curFamily; |
4111 | 0 | Oid curCollation; |
4112 | |
|
4113 | 0 | Assert(indexcol < index->nkeycolumns); |
4114 | |
|
4115 | 0 | curFamily = index->opfamily[indexcol]; |
4116 | 0 | curCollation = index->indexcollations[indexcol]; |
4117 | | |
4118 | | /* |
4119 | | * If it's a btree index, we can reject it if its opfamily isn't |
4120 | | * compatible with the EC, since no clause generated from the EC could be |
4121 | | * used with the index. For non-btree indexes, we can't easily tell |
4122 | | * whether clauses generated from the EC could be used with the index, so |
4123 | | * don't check the opfamily. This might mean we return "true" for a |
4124 | | * useless EC, so we have to recheck the results of |
4125 | | * generate_implied_equalities_for_column; see |
4126 | | * match_eclass_clauses_to_index. |
4127 | | */ |
4128 | 0 | if (index->relam == BTREE_AM_OID && |
4129 | 0 | !list_member_oid(ec->ec_opfamilies, curFamily)) |
4130 | 0 | return false; |
4131 | | |
4132 | | /* We insist on collation match for all index types, though */ |
4133 | 0 | if (!IndexCollMatchesExprColl(curCollation, ec->ec_collation)) |
4134 | 0 | return false; |
4135 | | |
4136 | 0 | return match_index_to_operand((Node *) em->em_expr, indexcol, index); |
4137 | 0 | } |
4138 | | |
4139 | | /* |
4140 | | * relation_has_unique_index_for |
4141 | | * Determine whether the relation provably has at most one row satisfying |
4142 | | * a set of equality conditions, because the conditions constrain all |
4143 | | * columns of some unique index. |
4144 | | * |
4145 | | * The conditions can be represented in either or both of two ways: |
4146 | | * 1. A list of RestrictInfo nodes, where the caller has already determined |
4147 | | * that each condition is a mergejoinable equality with an expression in |
4148 | | * this relation on one side, and an expression not involving this relation |
4149 | | * on the other. The transient outer_is_left flag is used to identify which |
4150 | | * side we should look at: left side if outer_is_left is false, right side |
4151 | | * if it is true. |
4152 | | * 2. A list of expressions in this relation, and a corresponding list of |
4153 | | * equality operators. The caller must have already checked that the operators |
4154 | | * represent equality. (Note: the operators could be cross-type; the |
4155 | | * expressions should correspond to their RHS inputs.) |
4156 | | * |
4157 | | * The caller need only supply equality conditions arising from joins; |
4158 | | * this routine automatically adds in any usable baserestrictinfo clauses. |
4159 | | * (Note that the passed-in restrictlist will be destructively modified!) |
4160 | | */ |
4161 | | bool |
4162 | | relation_has_unique_index_for(PlannerInfo *root, RelOptInfo *rel, |
4163 | | List *restrictlist, |
4164 | | List *exprlist, List *oprlist) |
4165 | 0 | { |
4166 | 0 | return relation_has_unique_index_ext(root, rel, restrictlist, |
4167 | 0 | exprlist, oprlist, NULL); |
4168 | 0 | } |
4169 | | |
4170 | | /* |
4171 | | * relation_has_unique_index_ext |
4172 | | * Same as relation_has_unique_index_for(), but supports extra_clauses |
4173 | | * parameter. If extra_clauses isn't NULL, return baserestrictinfo clauses |
4174 | | * which were used to derive uniqueness. |
4175 | | */ |
4176 | | bool |
4177 | | relation_has_unique_index_ext(PlannerInfo *root, RelOptInfo *rel, |
4178 | | List *restrictlist, |
4179 | | List *exprlist, List *oprlist, |
4180 | | List **extra_clauses) |
4181 | 0 | { |
4182 | 0 | ListCell *ic; |
4183 | |
|
4184 | 0 | Assert(list_length(exprlist) == list_length(oprlist)); |
4185 | | |
4186 | | /* Short-circuit if no indexes... */ |
4187 | 0 | if (rel->indexlist == NIL) |
4188 | 0 | return false; |
4189 | | |
4190 | | /* |
4191 | | * Examine the rel's restriction clauses for usable var = const clauses |
4192 | | * that we can add to the restrictlist. |
4193 | | */ |
4194 | 0 | foreach(ic, rel->baserestrictinfo) |
4195 | 0 | { |
4196 | 0 | RestrictInfo *restrictinfo = (RestrictInfo *) lfirst(ic); |
4197 | | |
4198 | | /* |
4199 | | * Note: can_join won't be set for a restriction clause, but |
4200 | | * mergeopfamilies will be if it has a mergejoinable operator and |
4201 | | * doesn't contain volatile functions. |
4202 | | */ |
4203 | 0 | if (restrictinfo->mergeopfamilies == NIL) |
4204 | 0 | continue; /* not mergejoinable */ |
4205 | | |
4206 | | /* |
4207 | | * The clause certainly doesn't refer to anything but the given rel. |
4208 | | * If either side is pseudoconstant then we can use it. |
4209 | | */ |
4210 | 0 | if (bms_is_empty(restrictinfo->left_relids)) |
4211 | 0 | { |
4212 | | /* righthand side is inner */ |
4213 | 0 | restrictinfo->outer_is_left = true; |
4214 | 0 | } |
4215 | 0 | else if (bms_is_empty(restrictinfo->right_relids)) |
4216 | 0 | { |
4217 | | /* lefthand side is inner */ |
4218 | 0 | restrictinfo->outer_is_left = false; |
4219 | 0 | } |
4220 | 0 | else |
4221 | 0 | continue; |
4222 | | |
4223 | | /* OK, add to list */ |
4224 | 0 | restrictlist = lappend(restrictlist, restrictinfo); |
4225 | 0 | } |
4226 | | |
4227 | | /* Short-circuit the easy case */ |
4228 | 0 | if (restrictlist == NIL && exprlist == NIL) |
4229 | 0 | return false; |
4230 | | |
4231 | | /* Examine each index of the relation ... */ |
4232 | 0 | foreach(ic, rel->indexlist) |
4233 | 0 | { |
4234 | 0 | IndexOptInfo *ind = (IndexOptInfo *) lfirst(ic); |
4235 | 0 | int c; |
4236 | 0 | List *exprs = NIL; |
4237 | | |
4238 | | /* |
4239 | | * If the index is not unique, or not immediately enforced, or if it's |
4240 | | * a partial index, it's useless here. We're unable to make use of |
4241 | | * predOK partial unique indexes due to the fact that |
4242 | | * check_index_predicates() also makes use of join predicates to |
4243 | | * determine if the partial index is usable. Here we need proofs that |
4244 | | * hold true before any joins are evaluated. |
4245 | | */ |
4246 | 0 | if (!ind->unique || !ind->immediate || ind->indpred != NIL) |
4247 | 0 | continue; |
4248 | | |
4249 | | /* |
4250 | | * Try to find each index column in the lists of conditions. This is |
4251 | | * O(N^2) or worse, but we expect all the lists to be short. |
4252 | | */ |
4253 | 0 | for (c = 0; c < ind->nkeycolumns; c++) |
4254 | 0 | { |
4255 | 0 | bool matched = false; |
4256 | 0 | ListCell *lc; |
4257 | 0 | ListCell *lc2; |
4258 | |
|
4259 | 0 | foreach(lc, restrictlist) |
4260 | 0 | { |
4261 | 0 | RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc); |
4262 | 0 | Node *rexpr; |
4263 | | |
4264 | | /* |
4265 | | * The condition's equality operator must be a member of the |
4266 | | * index opfamily, else it is not asserting the right kind of |
4267 | | * equality behavior for this index. We check this first |
4268 | | * since it's probably cheaper than match_index_to_operand(). |
4269 | | */ |
4270 | 0 | if (!list_member_oid(rinfo->mergeopfamilies, ind->opfamily[c])) |
4271 | 0 | continue; |
4272 | | |
4273 | | /* |
4274 | | * XXX at some point we may need to check collations here too. |
4275 | | * For the moment we assume all collations reduce to the same |
4276 | | * notion of equality. |
4277 | | */ |
4278 | | |
4279 | | /* OK, see if the condition operand matches the index key */ |
4280 | 0 | if (rinfo->outer_is_left) |
4281 | 0 | rexpr = get_rightop(rinfo->clause); |
4282 | 0 | else |
4283 | 0 | rexpr = get_leftop(rinfo->clause); |
4284 | |
|
4285 | 0 | if (match_index_to_operand(rexpr, c, ind)) |
4286 | 0 | { |
4287 | 0 | matched = true; /* column is unique */ |
4288 | |
|
4289 | 0 | if (bms_membership(rinfo->clause_relids) == BMS_SINGLETON) |
4290 | 0 | { |
4291 | 0 | MemoryContext oldMemCtx = |
4292 | 0 | MemoryContextSwitchTo(root->planner_cxt); |
4293 | | |
4294 | | /* |
4295 | | * Add filter clause into a list allowing caller to |
4296 | | * know if uniqueness have made not only by join |
4297 | | * clauses. |
4298 | | */ |
4299 | 0 | Assert(bms_is_empty(rinfo->left_relids) || |
4300 | 0 | bms_is_empty(rinfo->right_relids)); |
4301 | 0 | if (extra_clauses) |
4302 | 0 | exprs = lappend(exprs, rinfo); |
4303 | 0 | MemoryContextSwitchTo(oldMemCtx); |
4304 | 0 | } |
4305 | |
|
4306 | 0 | break; |
4307 | 0 | } |
4308 | 0 | } |
4309 | |
|
4310 | 0 | if (matched) |
4311 | 0 | continue; |
4312 | | |
4313 | 0 | forboth(lc, exprlist, lc2, oprlist) |
4314 | 0 | { |
4315 | 0 | Node *expr = (Node *) lfirst(lc); |
4316 | 0 | Oid opr = lfirst_oid(lc2); |
4317 | | |
4318 | | /* See if the expression matches the index key */ |
4319 | 0 | if (!match_index_to_operand(expr, c, ind)) |
4320 | 0 | continue; |
4321 | | |
4322 | | /* |
4323 | | * The equality operator must be a member of the index |
4324 | | * opfamily, else it is not asserting the right kind of |
4325 | | * equality behavior for this index. We assume the caller |
4326 | | * determined it is an equality operator, so we don't need to |
4327 | | * check any more tightly than this. |
4328 | | */ |
4329 | 0 | if (!op_in_opfamily(opr, ind->opfamily[c])) |
4330 | 0 | continue; |
4331 | | |
4332 | | /* |
4333 | | * XXX at some point we may need to check collations here too. |
4334 | | * For the moment we assume all collations reduce to the same |
4335 | | * notion of equality. |
4336 | | */ |
4337 | | |
4338 | 0 | matched = true; /* column is unique */ |
4339 | 0 | break; |
4340 | 0 | } |
4341 | |
|
4342 | 0 | if (!matched) |
4343 | 0 | break; /* no match; this index doesn't help us */ |
4344 | 0 | } |
4345 | | |
4346 | | /* Matched all key columns of this index? */ |
4347 | 0 | if (c == ind->nkeycolumns) |
4348 | 0 | { |
4349 | 0 | if (extra_clauses) |
4350 | 0 | *extra_clauses = exprs; |
4351 | 0 | return true; |
4352 | 0 | } |
4353 | 0 | } |
4354 | | |
4355 | 0 | return false; |
4356 | 0 | } |
4357 | | |
4358 | | /* |
4359 | | * indexcol_is_bool_constant_for_query |
4360 | | * |
4361 | | * If an index column is constrained to have a constant value by the query's |
4362 | | * WHERE conditions, then it's irrelevant for sort-order considerations. |
4363 | | * Usually that means we have a restriction clause WHERE indexcol = constant, |
4364 | | * which gets turned into an EquivalenceClass containing a constant, which |
4365 | | * is recognized as redundant by build_index_pathkeys(). But if the index |
4366 | | * column is a boolean variable (or expression), then we are not going to |
4367 | | * see WHERE indexcol = constant, because expression preprocessing will have |
4368 | | * simplified that to "WHERE indexcol" or "WHERE NOT indexcol". So we are not |
4369 | | * going to have a matching EquivalenceClass (unless the query also contains |
4370 | | * "ORDER BY indexcol"). To allow such cases to work the same as they would |
4371 | | * for non-boolean values, this function is provided to detect whether the |
4372 | | * specified index column matches a boolean restriction clause. |
4373 | | */ |
4374 | | bool |
4375 | | indexcol_is_bool_constant_for_query(PlannerInfo *root, |
4376 | | IndexOptInfo *index, |
4377 | | int indexcol) |
4378 | 0 | { |
4379 | 0 | ListCell *lc; |
4380 | | |
4381 | | /* If the index isn't boolean, we can't possibly get a match */ |
4382 | 0 | if (!IsBooleanOpfamily(index->opfamily[indexcol])) |
4383 | 0 | return false; |
4384 | | |
4385 | | /* Check each restriction clause for the index's rel */ |
4386 | 0 | foreach(lc, index->rel->baserestrictinfo) |
4387 | 0 | { |
4388 | 0 | RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc); |
4389 | | |
4390 | | /* |
4391 | | * As in match_clause_to_indexcol, never match pseudoconstants to |
4392 | | * indexes. (It might be semantically okay to do so here, but the |
4393 | | * odds of getting a match are negligible, so don't waste the cycles.) |
4394 | | */ |
4395 | 0 | if (rinfo->pseudoconstant) |
4396 | 0 | continue; |
4397 | | |
4398 | | /* See if we can match the clause's expression to the index column */ |
4399 | 0 | if (match_boolean_index_clause(root, rinfo, indexcol, index)) |
4400 | 0 | return true; |
4401 | 0 | } |
4402 | | |
4403 | 0 | return false; |
4404 | 0 | } |
4405 | | |
4406 | | |
4407 | | /**************************************************************************** |
4408 | | * ---- ROUTINES TO CHECK OPERANDS ---- |
4409 | | ****************************************************************************/ |
4410 | | |
4411 | | /* |
4412 | | * match_index_to_operand() |
4413 | | * Generalized test for a match between an index's key |
4414 | | * and the operand on one side of a restriction or join clause. |
4415 | | * |
4416 | | * operand: the nodetree to be compared to the index |
4417 | | * indexcol: the column number of the index (counting from 0) |
4418 | | * index: the index of interest |
4419 | | * |
4420 | | * Note that we aren't interested in collations here; the caller must check |
4421 | | * for a collation match, if it's dealing with an operator where that matters. |
4422 | | * |
4423 | | * This is exported for use in selfuncs.c. |
4424 | | */ |
4425 | | bool |
4426 | | match_index_to_operand(Node *operand, |
4427 | | int indexcol, |
4428 | | IndexOptInfo *index) |
4429 | 0 | { |
4430 | 0 | int indkey; |
4431 | | |
4432 | | /* |
4433 | | * Ignore any RelabelType node above the operand. This is needed to be |
4434 | | * able to apply indexscanning in binary-compatible-operator cases. Note: |
4435 | | * we can assume there is at most one RelabelType node; |
4436 | | * eval_const_expressions() will have simplified if more than one. |
4437 | | */ |
4438 | 0 | if (operand && IsA(operand, RelabelType)) |
4439 | 0 | operand = (Node *) ((RelabelType *) operand)->arg; |
4440 | |
|
4441 | 0 | indkey = index->indexkeys[indexcol]; |
4442 | 0 | if (indkey != 0) |
4443 | 0 | { |
4444 | | /* |
4445 | | * Simple index column; operand must be a matching Var. |
4446 | | */ |
4447 | 0 | if (operand && IsA(operand, Var) && |
4448 | 0 | index->rel->relid == ((Var *) operand)->varno && |
4449 | 0 | indkey == ((Var *) operand)->varattno && |
4450 | 0 | ((Var *) operand)->varnullingrels == NULL) |
4451 | 0 | return true; |
4452 | 0 | } |
4453 | 0 | else |
4454 | 0 | { |
4455 | | /* |
4456 | | * Index expression; find the correct expression. (This search could |
4457 | | * be avoided, at the cost of complicating all the callers of this |
4458 | | * routine; doesn't seem worth it.) |
4459 | | */ |
4460 | 0 | ListCell *indexpr_item; |
4461 | 0 | int i; |
4462 | 0 | Node *indexkey; |
4463 | |
|
4464 | 0 | indexpr_item = list_head(index->indexprs); |
4465 | 0 | for (i = 0; i < indexcol; i++) |
4466 | 0 | { |
4467 | 0 | if (index->indexkeys[i] == 0) |
4468 | 0 | { |
4469 | 0 | if (indexpr_item == NULL) |
4470 | 0 | elog(ERROR, "wrong number of index expressions"); |
4471 | 0 | indexpr_item = lnext(index->indexprs, indexpr_item); |
4472 | 0 | } |
4473 | 0 | } |
4474 | 0 | if (indexpr_item == NULL) |
4475 | 0 | elog(ERROR, "wrong number of index expressions"); |
4476 | 0 | indexkey = (Node *) lfirst(indexpr_item); |
4477 | | |
4478 | | /* |
4479 | | * Does it match the operand? Again, strip any relabeling. |
4480 | | */ |
4481 | 0 | if (indexkey && IsA(indexkey, RelabelType)) |
4482 | 0 | indexkey = (Node *) ((RelabelType *) indexkey)->arg; |
4483 | |
|
4484 | 0 | if (equal(indexkey, operand)) |
4485 | 0 | return true; |
4486 | 0 | } |
4487 | | |
4488 | 0 | return false; |
4489 | 0 | } |
4490 | | |
4491 | | /* |
4492 | | * is_pseudo_constant_for_index() |
4493 | | * Test whether the given expression can be used as an indexscan |
4494 | | * comparison value. |
4495 | | * |
4496 | | * An indexscan comparison value must not contain any volatile functions, |
4497 | | * and it can't contain any Vars of the index's own table. Vars of |
4498 | | * other tables are okay, though; in that case we'd be producing an |
4499 | | * indexqual usable in a parameterized indexscan. This is, therefore, |
4500 | | * a weaker condition than is_pseudo_constant_clause(). |
4501 | | * |
4502 | | * This function is exported for use by planner support functions, |
4503 | | * which will have available the IndexOptInfo, but not any RestrictInfo |
4504 | | * infrastructure. It is making the same test made by functions above |
4505 | | * such as match_opclause_to_indexcol(), but those rely where possible |
4506 | | * on RestrictInfo information about variable membership. |
4507 | | * |
4508 | | * expr: the nodetree to be checked |
4509 | | * index: the index of interest |
4510 | | */ |
4511 | | bool |
4512 | | is_pseudo_constant_for_index(PlannerInfo *root, Node *expr, IndexOptInfo *index) |
4513 | 0 | { |
4514 | | /* pull_varnos is cheaper than volatility check, so do that first */ |
4515 | 0 | if (bms_is_member(index->rel->relid, pull_varnos(root, expr))) |
4516 | 0 | return false; /* no good, contains Var of table */ |
4517 | 0 | if (contain_volatile_functions(expr)) |
4518 | 0 | return false; /* no good, volatile comparison value */ |
4519 | 0 | return true; |
4520 | 0 | } |