1 /*------------------------------------------------------------------------- 2 * 3 * pathnodes.h 4 * Definitions for planner's internal data structures, especially Paths. 5 * 6 * 7 * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group 8 * Portions Copyright (c) 1994, Regents of the University of California 9 * 10 * src/include/nodes/pathnodes.h 11 * 12 *------------------------------------------------------------------------- 13 */ 14 #ifndef PATHNODES_H 15 #define PATHNODES_H 16 17 #include "access/sdir.h" 18 #include "fmgr.h" 19 #include "lib/stringinfo.h" 20 #include "nodes/params.h" 21 #include "nodes/parsenodes.h" 22 #include "storage/block.h" 23 24 25 /* 26 * Relids 27 * Set of relation identifiers (indexes into the rangetable). 28 */ 29 typedef Bitmapset *Relids; 30 31 /* 32 * When looking for a "cheapest path", this enum specifies whether we want 33 * cheapest startup cost or cheapest total cost. 34 */ 35 typedef enum CostSelector 36 { 37 STARTUP_COST, TOTAL_COST 38 } CostSelector; 39 40 /* 41 * The cost estimate produced by cost_qual_eval() includes both a one-time 42 * (startup) cost, and a per-tuple cost. 43 */ 44 typedef struct QualCost 45 { 46 Cost startup; /* one-time cost */ 47 Cost per_tuple; /* per-evaluation cost */ 48 } QualCost; 49 50 /* 51 * Costing aggregate function execution requires these statistics about 52 * the aggregates to be executed by a given Agg node. Note that the costs 53 * include the execution costs of the aggregates' argument expressions as 54 * well as the aggregate functions themselves. Also, the fields must be 55 * defined so that initializing the struct to zeroes with memset is correct. 56 */ 57 typedef struct AggClauseCosts 58 { 59 int numAggs; /* total number of aggregate functions */ 60 int numOrderedAggs; /* number w/ DISTINCT/ORDER BY/WITHIN GROUP */ 61 bool hasNonPartial; /* does any agg not support partial mode? */ 62 bool hasNonSerial; /* is any partial agg non-serializable? */ 63 QualCost transCost; /* total per-input-row execution costs */ 64 QualCost finalCost; /* total per-aggregated-row costs */ 65 Size transitionSpace; /* space for pass-by-ref transition data */ 66 } AggClauseCosts; 67 68 /* 69 * This enum identifies the different types of "upper" (post-scan/join) 70 * relations that we might deal with during planning. 71 */ 72 typedef enum UpperRelationKind 73 { 74 UPPERREL_SETOP, /* result of UNION/INTERSECT/EXCEPT, if any */ 75 UPPERREL_PARTIAL_GROUP_AGG, /* result of partial grouping/aggregation, if 76 * any */ 77 UPPERREL_GROUP_AGG, /* result of grouping/aggregation, if any */ 78 UPPERREL_WINDOW, /* result of window functions, if any */ 79 UPPERREL_DISTINCT, /* result of "SELECT DISTINCT", if any */ 80 UPPERREL_ORDERED, /* result of ORDER BY, if any */ 81 UPPERREL_FINAL /* result of any remaining top-level actions */ 82 /* NB: UPPERREL_FINAL must be last enum entry; it's used to size arrays */ 83 } UpperRelationKind; 84 85 /* 86 * This enum identifies which type of relation is being planned through the 87 * inheritance planner. INHKIND_NONE indicates the inheritance planner 88 * was not used. 89 */ 90 typedef enum InheritanceKind 91 { 92 INHKIND_NONE, 93 INHKIND_INHERITED, 94 INHKIND_PARTITIONED 95 } InheritanceKind; 96 97 /*---------- 98 * PlannerGlobal 99 * Global information for planning/optimization 100 * 101 * PlannerGlobal holds state for an entire planner invocation; this state 102 * is shared across all levels of sub-Queries that exist in the command being 103 * planned. 104 *---------- 105 */ 106 typedef struct PlannerGlobal 107 { 108 NodeTag type; 109 110 ParamListInfo boundParams; /* Param values provided to planner() */ 111 112 List *subplans; /* Plans for SubPlan nodes */ 113 114 List *subroots; /* PlannerInfos for SubPlan nodes */ 115 116 Bitmapset *rewindPlanIDs; /* indices of subplans that require REWIND */ 117 118 List *finalrtable; /* "flat" rangetable for executor */ 119 120 List *finalrowmarks; /* "flat" list of PlanRowMarks */ 121 122 List *resultRelations; /* "flat" list of integer RT indexes */ 123 124 List *rootResultRelations; /* "flat" list of integer RT indexes */ 125 126 List *relationOids; /* OIDs of relations the plan depends on */ 127 128 List *invalItems; /* other dependencies, as PlanInvalItems */ 129 130 List *paramExecTypes; /* type OIDs for PARAM_EXEC Params */ 131 132 Index lastPHId; /* highest PlaceHolderVar ID assigned */ 133 134 Index lastRowMarkId; /* highest PlanRowMark ID assigned */ 135 136 int lastPlanNodeId; /* highest plan node ID assigned */ 137 138 bool transientPlan; /* redo plan when TransactionXmin changes? */ 139 140 bool dependsOnRole; /* is plan specific to current role? */ 141 142 bool parallelModeOK; /* parallel mode potentially OK? */ 143 144 bool parallelModeNeeded; /* parallel mode actually required? */ 145 146 char maxParallelHazard; /* worst PROPARALLEL hazard level */ 147 148 PartitionDirectory partition_directory; /* partition descriptors */ 149 } PlannerGlobal; 150 151 /* macro for fetching the Plan associated with a SubPlan node */ 152 #define planner_subplan_get_plan(root, subplan) \ 153 ((Plan *) list_nth((root)->glob->subplans, (subplan)->plan_id - 1)) 154 155 156 /*---------- 157 * PlannerInfo 158 * Per-query information for planning/optimization 159 * 160 * This struct is conventionally called "root" in all the planner routines. 161 * It holds links to all of the planner's working state, in addition to the 162 * original Query. Note that at present the planner extensively modifies 163 * the passed-in Query data structure; someday that should stop. 164 * 165 * For reasons explained in optimizer/optimizer.h, we define the typedef 166 * either here or in that header, whichever is read first. 167 *---------- 168 */ 169 #ifndef HAVE_PLANNERINFO_TYPEDEF 170 typedef struct PlannerInfo PlannerInfo; 171 #define HAVE_PLANNERINFO_TYPEDEF 1 172 #endif 173 174 struct PlannerInfo 175 { 176 NodeTag type; 177 178 Query *parse; /* the Query being planned */ 179 180 PlannerGlobal *glob; /* global info for current planner run */ 181 182 Index query_level; /* 1 at the outermost Query */ 183 184 PlannerInfo *parent_root; /* NULL at outermost Query */ 185 186 /* 187 * plan_params contains the expressions that this query level needs to 188 * make available to a lower query level that is currently being planned. 189 * outer_params contains the paramIds of PARAM_EXEC Params that outer 190 * query levels will make available to this query level. 191 */ 192 List *plan_params; /* list of PlannerParamItems, see below */ 193 Bitmapset *outer_params; 194 195 /* 196 * simple_rel_array holds pointers to "base rels" and "other rels" (see 197 * comments for RelOptInfo for more info). It is indexed by rangetable 198 * index (so entry 0 is always wasted). Entries can be NULL when an RTE 199 * does not correspond to a base relation, such as a join RTE or an 200 * unreferenced view RTE; or if the RelOptInfo hasn't been made yet. 201 */ 202 struct RelOptInfo **simple_rel_array; /* All 1-rel RelOptInfos */ 203 int simple_rel_array_size; /* allocated size of array */ 204 205 /* 206 * simple_rte_array is the same length as simple_rel_array and holds 207 * pointers to the associated rangetable entries. This lets us avoid 208 * rt_fetch(), which can be a bit slow once large inheritance sets have 209 * been expanded. 210 */ 211 RangeTblEntry **simple_rte_array; /* rangetable as an array */ 212 213 /* 214 * append_rel_array is the same length as the above arrays, and holds 215 * pointers to the corresponding AppendRelInfo entry indexed by 216 * child_relid, or NULL if none. The array itself is not allocated if 217 * append_rel_list is empty. 218 */ 219 struct AppendRelInfo **append_rel_array; 220 221 /* 222 * all_baserels is a Relids set of all base relids (but not "other" 223 * relids) in the query; that is, the Relids identifier of the final join 224 * we need to form. This is computed in make_one_rel, just before we 225 * start making Paths. 226 */ 227 Relids all_baserels; 228 229 /* 230 * nullable_baserels is a Relids set of base relids that are nullable by 231 * some outer join in the jointree; these are rels that are potentially 232 * nullable below the WHERE clause, SELECT targetlist, etc. This is 233 * computed in deconstruct_jointree. 234 */ 235 Relids nullable_baserels; 236 237 /* 238 * join_rel_list is a list of all join-relation RelOptInfos we have 239 * considered in this planning run. For small problems we just scan the 240 * list to do lookups, but when there are many join relations we build a 241 * hash table for faster lookups. The hash table is present and valid 242 * when join_rel_hash is not NULL. Note that we still maintain the list 243 * even when using the hash table for lookups; this simplifies life for 244 * GEQO. 245 */ 246 List *join_rel_list; /* list of join-relation RelOptInfos */ 247 struct HTAB *join_rel_hash; /* optional hashtable for join relations */ 248 249 /* 250 * When doing a dynamic-programming-style join search, join_rel_level[k] 251 * is a list of all join-relation RelOptInfos of level k, and 252 * join_cur_level is the current level. New join-relation RelOptInfos are 253 * automatically added to the join_rel_level[join_cur_level] list. 254 * join_rel_level is NULL if not in use. 255 */ 256 List **join_rel_level; /* lists of join-relation RelOptInfos */ 257 int join_cur_level; /* index of list being extended */ 258 259 List *init_plans; /* init SubPlans for query */ 260 261 List *cte_plan_ids; /* per-CTE-item list of subplan IDs */ 262 263 List *multiexpr_params; /* List of Lists of Params for MULTIEXPR 264 * subquery outputs */ 265 266 List *eq_classes; /* list of active EquivalenceClasses */ 267 268 List *canon_pathkeys; /* list of "canonical" PathKeys */ 269 270 List *left_join_clauses; /* list of RestrictInfos for mergejoinable 271 * outer join clauses w/nonnullable var on 272 * left */ 273 274 List *right_join_clauses; /* list of RestrictInfos for mergejoinable 275 * outer join clauses w/nonnullable var on 276 * right */ 277 278 List *full_join_clauses; /* list of RestrictInfos for mergejoinable 279 * full join clauses */ 280 281 List *join_info_list; /* list of SpecialJoinInfos */ 282 283 /* 284 * Note: for AppendRelInfos describing partitions of a partitioned table, 285 * we guarantee that partitions that come earlier in the partitioned 286 * table's PartitionDesc will appear earlier in append_rel_list. 287 */ 288 List *append_rel_list; /* list of AppendRelInfos */ 289 290 List *rowMarks; /* list of PlanRowMarks */ 291 292 List *placeholder_list; /* list of PlaceHolderInfos */ 293 294 List *fkey_list; /* list of ForeignKeyOptInfos */ 295 296 List *query_pathkeys; /* desired pathkeys for query_planner() */ 297 298 List *group_pathkeys; /* groupClause pathkeys, if any */ 299 List *window_pathkeys; /* pathkeys of bottom window, if any */ 300 List *distinct_pathkeys; /* distinctClause pathkeys, if any */ 301 List *sort_pathkeys; /* sortClause pathkeys, if any */ 302 303 List *part_schemes; /* Canonicalised partition schemes used in the 304 * query. */ 305 306 List *initial_rels; /* RelOptInfos we are now trying to join */ 307 308 /* Use fetch_upper_rel() to get any particular upper rel */ 309 List *upper_rels[UPPERREL_FINAL + 1]; /* upper-rel RelOptInfos */ 310 311 /* Result tlists chosen by grouping_planner for upper-stage processing */ 312 struct PathTarget *upper_targets[UPPERREL_FINAL + 1]; 313 314 /* 315 * The fully-processed targetlist is kept here. It differs from 316 * parse->targetList in that (for INSERT and UPDATE) it's been reordered 317 * to match the target table, and defaults have been filled in. Also, 318 * additional resjunk targets may be present. preprocess_targetlist() 319 * does most of this work, but note that more resjunk targets can get 320 * added during appendrel expansion. (Hence, upper_targets mustn't get 321 * set up till after that.) 322 */ 323 List *processed_tlist; 324 325 /* Fields filled during create_plan() for use in setrefs.c */ 326 AttrNumber *grouping_map; /* for GroupingFunc fixup */ 327 List *minmax_aggs; /* List of MinMaxAggInfos */ 328 329 MemoryContext planner_cxt; /* context holding PlannerInfo */ 330 331 double total_table_pages; /* # of pages in all non-dummy tables of 332 * query */ 333 334 double tuple_fraction; /* tuple_fraction passed to query_planner */ 335 double limit_tuples; /* limit_tuples passed to query_planner */ 336 337 Index qual_security_level; /* minimum security_level for quals */ 338 /* Note: qual_security_level is zero if there are no securityQuals */ 339 340 InheritanceKind inhTargetKind; /* indicates if the target relation is an 341 * inheritance child or partition or a 342 * partitioned table */ 343 bool hasJoinRTEs; /* true if any RTEs are RTE_JOIN kind */ 344 bool hasLateralRTEs; /* true if any RTEs are marked LATERAL */ 345 bool hasHavingQual; /* true if havingQual was non-null */ 346 bool hasPseudoConstantQuals; /* true if any RestrictInfo has 347 * pseudoconstant = true */ 348 bool hasRecursion; /* true if planning a recursive WITH item */ 349 350 /* These fields are used only when hasRecursion is true: */ 351 int wt_param_id; /* PARAM_EXEC ID for the work table */ 352 struct Path *non_recursive_path; /* a path for non-recursive term */ 353 354 /* These fields are workspace for createplan.c */ 355 Relids curOuterRels; /* outer rels above current node */ 356 List *curOuterParams; /* not-yet-assigned NestLoopParams */ 357 358 /* optional private data for join_search_hook, e.g., GEQO */ 359 void *join_search_private; 360 361 /* Does this query modify any partition key columns? */ 362 bool partColsUpdated; 363 }; 364 365 366 /* 367 * In places where it's known that simple_rte_array[] must have been prepared 368 * already, we just index into it to fetch RTEs. In code that might be 369 * executed before or after entering query_planner(), use this macro. 370 */ 371 #define planner_rt_fetch(rti, root) \ 372 ((root)->simple_rte_array ? (root)->simple_rte_array[rti] : \ 373 rt_fetch(rti, (root)->parse->rtable)) 374 375 /* 376 * If multiple relations are partitioned the same way, all such partitions 377 * will have a pointer to the same PartitionScheme. A list of PartitionScheme 378 * objects is attached to the PlannerInfo. By design, the partition scheme 379 * incorporates only the general properties of the partition method (LIST vs. 380 * RANGE, number of partitioning columns and the type information for each) 381 * and not the specific bounds. 382 * 383 * We store the opclass-declared input data types instead of the partition key 384 * datatypes since the former rather than the latter are used to compare 385 * partition bounds. Since partition key data types and the opclass declared 386 * input data types are expected to be binary compatible (per ResolveOpClass), 387 * both of those should have same byval and length properties. 388 */ 389 typedef struct PartitionSchemeData 390 { 391 char strategy; /* partition strategy */ 392 int16 partnatts; /* number of partition attributes */ 393 Oid *partopfamily; /* OIDs of operator families */ 394 Oid *partopcintype; /* OIDs of opclass declared input data types */ 395 Oid *partcollation; /* OIDs of partitioning collations */ 396 397 /* Cached information about partition key data types. */ 398 int16 *parttyplen; 399 bool *parttypbyval; 400 401 /* Cached information about partition comparison functions. */ 402 FmgrInfo *partsupfunc; 403 } PartitionSchemeData; 404 405 typedef struct PartitionSchemeData *PartitionScheme; 406 407 /*---------- 408 * RelOptInfo 409 * Per-relation information for planning/optimization 410 * 411 * For planning purposes, a "base rel" is either a plain relation (a table) 412 * or the output of a sub-SELECT or function that appears in the range table. 413 * In either case it is uniquely identified by an RT index. A "joinrel" 414 * is the joining of two or more base rels. A joinrel is identified by 415 * the set of RT indexes for its component baserels. We create RelOptInfo 416 * nodes for each baserel and joinrel, and store them in the PlannerInfo's 417 * simple_rel_array and join_rel_list respectively. 418 * 419 * Note that there is only one joinrel for any given set of component 420 * baserels, no matter what order we assemble them in; so an unordered 421 * set is the right datatype to identify it with. 422 * 423 * We also have "other rels", which are like base rels in that they refer to 424 * single RT indexes; but they are not part of the join tree, and are given 425 * a different RelOptKind to identify them. 426 * Currently the only kind of otherrels are those made for member relations 427 * of an "append relation", that is an inheritance set or UNION ALL subquery. 428 * An append relation has a parent RTE that is a base rel, which represents 429 * the entire append relation. The member RTEs are otherrels. The parent 430 * is present in the query join tree but the members are not. The member 431 * RTEs and otherrels are used to plan the scans of the individual tables or 432 * subqueries of the append set; then the parent baserel is given Append 433 * and/or MergeAppend paths comprising the best paths for the individual 434 * member rels. (See comments for AppendRelInfo for more information.) 435 * 436 * At one time we also made otherrels to represent join RTEs, for use in 437 * handling join alias Vars. Currently this is not needed because all join 438 * alias Vars are expanded to non-aliased form during preprocess_expression. 439 * 440 * We also have relations representing joins between child relations of 441 * different partitioned tables. These relations are not added to 442 * join_rel_level lists as they are not joined directly by the dynamic 443 * programming algorithm. 444 * 445 * There is also a RelOptKind for "upper" relations, which are RelOptInfos 446 * that describe post-scan/join processing steps, such as aggregation. 447 * Many of the fields in these RelOptInfos are meaningless, but their Path 448 * fields always hold Paths showing ways to do that processing step. 449 * 450 * Lastly, there is a RelOptKind for "dead" relations, which are base rels 451 * that we have proven we don't need to join after all. 452 * 453 * Parts of this data structure are specific to various scan and join 454 * mechanisms. It didn't seem worth creating new node types for them. 455 * 456 * relids - Set of base-relation identifiers; it is a base relation 457 * if there is just one, a join relation if more than one 458 * rows - estimated number of tuples in the relation after restriction 459 * clauses have been applied (ie, output rows of a plan for it) 460 * consider_startup - true if there is any value in keeping plain paths for 461 * this rel on the basis of having cheap startup cost 462 * consider_param_startup - the same for parameterized paths 463 * reltarget - Default Path output tlist for this rel; normally contains 464 * Var and PlaceHolderVar nodes for the values we need to 465 * output from this relation. 466 * List is in no particular order, but all rels of an 467 * appendrel set must use corresponding orders. 468 * NOTE: in an appendrel child relation, may contain 469 * arbitrary expressions pulled up from a subquery! 470 * pathlist - List of Path nodes, one for each potentially useful 471 * method of generating the relation 472 * ppilist - ParamPathInfo nodes for parameterized Paths, if any 473 * cheapest_startup_path - the pathlist member with lowest startup cost 474 * (regardless of ordering) among the unparameterized paths; 475 * or NULL if there is no unparameterized path 476 * cheapest_total_path - the pathlist member with lowest total cost 477 * (regardless of ordering) among the unparameterized paths; 478 * or if there is no unparameterized path, the path with lowest 479 * total cost among the paths with minimum parameterization 480 * cheapest_unique_path - for caching cheapest path to produce unique 481 * (no duplicates) output from relation; NULL if not yet requested 482 * cheapest_parameterized_paths - best paths for their parameterizations; 483 * always includes cheapest_total_path, even if that's unparameterized 484 * direct_lateral_relids - rels this rel has direct LATERAL references to 485 * lateral_relids - required outer rels for LATERAL, as a Relids set 486 * (includes both direct and indirect lateral references) 487 * 488 * If the relation is a base relation it will have these fields set: 489 * 490 * relid - RTE index (this is redundant with the relids field, but 491 * is provided for convenience of access) 492 * rtekind - copy of RTE's rtekind field 493 * min_attr, max_attr - range of valid AttrNumbers for rel 494 * attr_needed - array of bitmapsets indicating the highest joinrel 495 * in which each attribute is needed; if bit 0 is set then 496 * the attribute is needed as part of final targetlist 497 * attr_widths - cache space for per-attribute width estimates; 498 * zero means not computed yet 499 * lateral_vars - lateral cross-references of rel, if any (list of 500 * Vars and PlaceHolderVars) 501 * lateral_referencers - relids of rels that reference this one laterally 502 * (includes both direct and indirect lateral references) 503 * indexlist - list of IndexOptInfo nodes for relation's indexes 504 * (always NIL if it's not a table) 505 * pages - number of disk pages in relation (zero if not a table) 506 * tuples - number of tuples in relation (not considering restrictions) 507 * allvisfrac - fraction of disk pages that are marked all-visible 508 * subroot - PlannerInfo for subquery (NULL if it's not a subquery) 509 * subplan_params - list of PlannerParamItems to be passed to subquery 510 * 511 * Note: for a subquery, tuples and subroot are not set immediately 512 * upon creation of the RelOptInfo object; they are filled in when 513 * set_subquery_pathlist processes the object. 514 * 515 * For otherrels that are appendrel members, these fields are filled 516 * in just as for a baserel, except we don't bother with lateral_vars. 517 * 518 * If the relation is either a foreign table or a join of foreign tables that 519 * all belong to the same foreign server and are assigned to the same user to 520 * check access permissions as (cf checkAsUser), these fields will be set: 521 * 522 * serverid - OID of foreign server, if foreign table (else InvalidOid) 523 * userid - OID of user to check access as (InvalidOid means current user) 524 * useridiscurrent - we've assumed that userid equals current user 525 * fdwroutine - function hooks for FDW, if foreign table (else NULL) 526 * fdw_private - private state for FDW, if foreign table (else NULL) 527 * 528 * Two fields are used to cache knowledge acquired during the join search 529 * about whether this rel is provably unique when being joined to given other 530 * relation(s), ie, it can have at most one row matching any given row from 531 * that join relation. Currently we only attempt such proofs, and thus only 532 * populate these fields, for base rels; but someday they might be used for 533 * join rels too: 534 * 535 * unique_for_rels - list of Relid sets, each one being a set of other 536 * rels for which this one has been proven unique 537 * non_unique_for_rels - list of Relid sets, each one being a set of 538 * other rels for which we have tried and failed to prove 539 * this one unique 540 * 541 * The presence of the following fields depends on the restrictions 542 * and joins that the relation participates in: 543 * 544 * baserestrictinfo - List of RestrictInfo nodes, containing info about 545 * each non-join qualification clause in which this relation 546 * participates (only used for base rels) 547 * baserestrictcost - Estimated cost of evaluating the baserestrictinfo 548 * clauses at a single tuple (only used for base rels) 549 * baserestrict_min_security - Smallest security_level found among 550 * clauses in baserestrictinfo 551 * joininfo - List of RestrictInfo nodes, containing info about each 552 * join clause in which this relation participates (but 553 * note this excludes clauses that might be derivable from 554 * EquivalenceClasses) 555 * has_eclass_joins - flag that EquivalenceClass joins are possible 556 * 557 * Note: Keeping a restrictinfo list in the RelOptInfo is useful only for 558 * base rels, because for a join rel the set of clauses that are treated as 559 * restrict clauses varies depending on which sub-relations we choose to join. 560 * (For example, in a 3-base-rel join, a clause relating rels 1 and 2 must be 561 * treated as a restrictclause if we join {1} and {2 3} to make {1 2 3}; but 562 * if we join {1 2} and {3} then that clause will be a restrictclause in {1 2} 563 * and should not be processed again at the level of {1 2 3}.) Therefore, 564 * the restrictinfo list in the join case appears in individual JoinPaths 565 * (field joinrestrictinfo), not in the parent relation. But it's OK for 566 * the RelOptInfo to store the joininfo list, because that is the same 567 * for a given rel no matter how we form it. 568 * 569 * We store baserestrictcost in the RelOptInfo (for base relations) because 570 * we know we will need it at least once (to price the sequential scan) 571 * and may need it multiple times to price index scans. 572 * 573 * If the relation is partitioned, these fields will be set: 574 * 575 * part_scheme - Partitioning scheme of the relation 576 * nparts - Number of partitions 577 * boundinfo - Partition bounds 578 * partition_qual - Partition constraint if not the root 579 * part_rels - RelOptInfos for each partition 580 * partexprs, nullable_partexprs - Partition key expressions 581 * partitioned_child_rels - RT indexes of unpruned partitions of 582 * this relation that are partitioned tables 583 * themselves, in hierarchical order 584 * 585 * Note: A base relation always has only one set of partition keys, but a join 586 * relation may have as many sets of partition keys as the number of relations 587 * being joined. partexprs and nullable_partexprs are arrays containing 588 * part_scheme->partnatts elements each. Each of these elements is a list of 589 * partition key expressions. For a base relation each list in partexprs 590 * contains only one expression and nullable_partexprs is not populated. For a 591 * join relation, partexprs and nullable_partexprs contain partition key 592 * expressions from non-nullable and nullable relations resp. Lists at any 593 * given position in those arrays together contain as many elements as the 594 * number of joining relations. 595 *---------- 596 */ 597 typedef enum RelOptKind 598 { 599 RELOPT_BASEREL, 600 RELOPT_JOINREL, 601 RELOPT_OTHER_MEMBER_REL, 602 RELOPT_OTHER_JOINREL, 603 RELOPT_UPPER_REL, 604 RELOPT_OTHER_UPPER_REL, 605 RELOPT_DEADREL 606 } RelOptKind; 607 608 /* 609 * Is the given relation a simple relation i.e a base or "other" member 610 * relation? 611 */ 612 #define IS_SIMPLE_REL(rel) \ 613 ((rel)->reloptkind == RELOPT_BASEREL || \ 614 (rel)->reloptkind == RELOPT_OTHER_MEMBER_REL) 615 616 /* Is the given relation a join relation? */ 617 #define IS_JOIN_REL(rel) \ 618 ((rel)->reloptkind == RELOPT_JOINREL || \ 619 (rel)->reloptkind == RELOPT_OTHER_JOINREL) 620 621 /* Is the given relation an upper relation? */ 622 #define IS_UPPER_REL(rel) \ 623 ((rel)->reloptkind == RELOPT_UPPER_REL || \ 624 (rel)->reloptkind == RELOPT_OTHER_UPPER_REL) 625 626 /* Is the given relation an "other" relation? */ 627 #define IS_OTHER_REL(rel) \ 628 ((rel)->reloptkind == RELOPT_OTHER_MEMBER_REL || \ 629 (rel)->reloptkind == RELOPT_OTHER_JOINREL || \ 630 (rel)->reloptkind == RELOPT_OTHER_UPPER_REL) 631 632 typedef struct RelOptInfo 633 { 634 NodeTag type; 635 636 RelOptKind reloptkind; 637 638 /* all relations included in this RelOptInfo */ 639 Relids relids; /* set of base relids (rangetable indexes) */ 640 641 /* size estimates generated by planner */ 642 double rows; /* estimated number of result tuples */ 643 644 /* per-relation planner control flags */ 645 bool consider_startup; /* keep cheap-startup-cost paths? */ 646 bool consider_param_startup; /* ditto, for parameterized paths? */ 647 bool consider_parallel; /* consider parallel paths? */ 648 649 /* default result targetlist for Paths scanning this relation */ 650 struct PathTarget *reltarget; /* list of Vars/Exprs, cost, width */ 651 652 /* materialization information */ 653 List *pathlist; /* Path structures */ 654 List *ppilist; /* ParamPathInfos used in pathlist */ 655 List *partial_pathlist; /* partial Paths */ 656 struct Path *cheapest_startup_path; 657 struct Path *cheapest_total_path; 658 struct Path *cheapest_unique_path; 659 List *cheapest_parameterized_paths; 660 661 /* parameterization information needed for both base rels and join rels */ 662 /* (see also lateral_vars and lateral_referencers) */ 663 Relids direct_lateral_relids; /* rels directly laterally referenced */ 664 Relids lateral_relids; /* minimum parameterization of rel */ 665 666 /* information about a base rel (not set for join rels!) */ 667 Index relid; 668 Oid reltablespace; /* containing tablespace */ 669 RTEKind rtekind; /* RELATION, SUBQUERY, FUNCTION, etc */ 670 AttrNumber min_attr; /* smallest attrno of rel (often <0) */ 671 AttrNumber max_attr; /* largest attrno of rel */ 672 Relids *attr_needed; /* array indexed [min_attr .. max_attr] */ 673 int32 *attr_widths; /* array indexed [min_attr .. max_attr] */ 674 List *lateral_vars; /* LATERAL Vars and PHVs referenced by rel */ 675 Relids lateral_referencers; /* rels that reference me laterally */ 676 List *indexlist; /* list of IndexOptInfo */ 677 List *statlist; /* list of StatisticExtInfo */ 678 BlockNumber pages; /* size estimates derived from pg_class */ 679 double tuples; 680 double allvisfrac; 681 PlannerInfo *subroot; /* if subquery */ 682 List *subplan_params; /* if subquery */ 683 int rel_parallel_workers; /* wanted number of parallel workers */ 684 685 /* Information about foreign tables and foreign joins */ 686 Oid serverid; /* identifies server for the table or join */ 687 Oid userid; /* identifies user to check access as */ 688 bool useridiscurrent; /* join is only valid for current user */ 689 /* use "struct FdwRoutine" to avoid including fdwapi.h here */ 690 struct FdwRoutine *fdwroutine; 691 void *fdw_private; 692 693 /* cache space for remembering if we have proven this relation unique */ 694 List *unique_for_rels; /* known unique for these other relid 695 * set(s) */ 696 List *non_unique_for_rels; /* known not unique for these set(s) */ 697 698 /* used by various scans and joins: */ 699 List *baserestrictinfo; /* RestrictInfo structures (if base rel) */ 700 QualCost baserestrictcost; /* cost of evaluating the above */ 701 Index baserestrict_min_security; /* min security_level found in 702 * baserestrictinfo */ 703 List *joininfo; /* RestrictInfo structures for join clauses 704 * involving this rel */ 705 bool has_eclass_joins; /* T means joininfo is incomplete */ 706 707 /* used by partitionwise joins: */ 708 bool consider_partitionwise_join; /* consider partitionwise join 709 * paths? (if partitioned rel) */ 710 Relids top_parent_relids; /* Relids of topmost parents (if "other" 711 * rel) */ 712 713 /* used for partitioned relations */ 714 PartitionScheme part_scheme; /* Partitioning scheme. */ 715 int nparts; /* number of partitions */ 716 struct PartitionBoundInfoData *boundinfo; /* Partition bounds */ 717 List *partition_qual; /* partition constraint */ 718 struct RelOptInfo **part_rels; /* Array of RelOptInfos of partitions, 719 * stored in the same order of bounds */ 720 List **partexprs; /* Non-nullable partition key expressions. */ 721 List **nullable_partexprs; /* Nullable partition key expressions. */ 722 List *partitioned_child_rels; /* List of RT indexes. */ 723 } RelOptInfo; 724 725 /* 726 * Is given relation partitioned? 727 * 728 * It's not enough to test whether rel->part_scheme is set, because it might 729 * be that the basic partitioning properties of the input relations matched 730 * but the partition bounds did not. Also, if we are able to prove a rel 731 * dummy (empty), we should henceforth treat it as unpartitioned. 732 */ 733 #define IS_PARTITIONED_REL(rel) \ 734 ((rel)->part_scheme && (rel)->boundinfo && (rel)->nparts > 0 && \ 735 (rel)->part_rels && !IS_DUMMY_REL(rel)) 736 737 /* 738 * Convenience macro to make sure that a partitioned relation has all the 739 * required members set. 740 */ 741 #define REL_HAS_ALL_PART_PROPS(rel) \ 742 ((rel)->part_scheme && (rel)->boundinfo && (rel)->nparts > 0 && \ 743 (rel)->part_rels && (rel)->partexprs && (rel)->nullable_partexprs) 744 745 /* 746 * IndexOptInfo 747 * Per-index information for planning/optimization 748 * 749 * indexkeys[], indexcollations[] each have ncolumns entries. 750 * opfamily[], and opcintype[] each have nkeycolumns entries. They do 751 * not contain any information about included attributes. 752 * 753 * sortopfamily[], reverse_sort[], and nulls_first[] have 754 * nkeycolumns entries, if the index is ordered; but if it is unordered, 755 * those pointers are NULL. 756 * 757 * Zeroes in the indexkeys[] array indicate index columns that are 758 * expressions; there is one element in indexprs for each such column. 759 * 760 * For an ordered index, reverse_sort[] and nulls_first[] describe the 761 * sort ordering of a forward indexscan; we can also consider a backward 762 * indexscan, which will generate the reverse ordering. 763 * 764 * The indexprs and indpred expressions have been run through 765 * prepqual.c and eval_const_expressions() for ease of matching to 766 * WHERE clauses. indpred is in implicit-AND form. 767 * 768 * indextlist is a TargetEntry list representing the index columns. 769 * It provides an equivalent base-relation Var for each simple column, 770 * and links to the matching indexprs element for each expression column. 771 * 772 * While most of these fields are filled when the IndexOptInfo is created 773 * (by plancat.c), indrestrictinfo and predOK are set later, in 774 * check_index_predicates(). 775 */ 776 #ifndef HAVE_INDEXOPTINFO_TYPEDEF 777 typedef struct IndexOptInfo IndexOptInfo; 778 #define HAVE_INDEXOPTINFO_TYPEDEF 1 779 #endif 780 781 struct IndexOptInfo 782 { 783 NodeTag type; 784 785 Oid indexoid; /* OID of the index relation */ 786 Oid reltablespace; /* tablespace of index (not table) */ 787 RelOptInfo *rel; /* back-link to index's table */ 788 789 /* index-size statistics (from pg_class and elsewhere) */ 790 BlockNumber pages; /* number of disk pages in index */ 791 double tuples; /* number of index tuples in index */ 792 int tree_height; /* index tree height, or -1 if unknown */ 793 794 /* index descriptor information */ 795 int ncolumns; /* number of columns in index */ 796 int nkeycolumns; /* number of key columns in index */ 797 int *indexkeys; /* column numbers of index's attributes both 798 * key and included columns, or 0 */ 799 Oid *indexcollations; /* OIDs of collations of index columns */ 800 Oid *opfamily; /* OIDs of operator families for columns */ 801 Oid *opcintype; /* OIDs of opclass declared input data types */ 802 Oid *sortopfamily; /* OIDs of btree opfamilies, if orderable */ 803 bool *reverse_sort; /* is sort order descending? */ 804 bool *nulls_first; /* do NULLs come first in the sort order? */ 805 bool *canreturn; /* which index cols can be returned in an 806 * index-only scan? */ 807 Oid relam; /* OID of the access method (in pg_am) */ 808 809 List *indexprs; /* expressions for non-simple index columns */ 810 List *indpred; /* predicate if a partial index, else NIL */ 811 812 List *indextlist; /* targetlist representing index columns */ 813 814 List *indrestrictinfo; /* parent relation's baserestrictinfo 815 * list, less any conditions implied by 816 * the index's predicate (unless it's a 817 * target rel, see comments in 818 * check_index_predicates()) */ 819 820 bool predOK; /* true if index predicate matches query */ 821 bool unique; /* true if a unique index */ 822 bool immediate; /* is uniqueness enforced immediately? */ 823 bool hypothetical; /* true if index doesn't really exist */ 824 825 /* Remaining fields are copied from the index AM's API struct: */ 826 bool amcanorderbyop; /* does AM support order by operator result? */ 827 bool amoptionalkey; /* can query omit key for the first column? */ 828 bool amsearcharray; /* can AM handle ScalarArrayOpExpr quals? */ 829 bool amsearchnulls; /* can AM search for NULL/NOT NULL entries? */ 830 bool amhasgettuple; /* does AM have amgettuple interface? */ 831 bool amhasgetbitmap; /* does AM have amgetbitmap interface? */ 832 bool amcanparallel; /* does AM support parallel scan? */ 833 bool amcanmarkpos; /* does AM support mark/restore? */ 834 /* Rather than include amapi.h here, we declare amcostestimate like this */ 835 void (*amcostestimate) (); /* AM's cost estimator */ 836 }; 837 838 /* 839 * ForeignKeyOptInfo 840 * Per-foreign-key information for planning/optimization 841 * 842 * The per-FK-column arrays can be fixed-size because we allow at most 843 * INDEX_MAX_KEYS columns in a foreign key constraint. Each array has 844 * nkeys valid entries. 845 */ 846 typedef struct ForeignKeyOptInfo 847 { 848 NodeTag type; 849 850 /* Basic data about the foreign key (fetched from catalogs): */ 851 Index con_relid; /* RT index of the referencing table */ 852 Index ref_relid; /* RT index of the referenced table */ 853 int nkeys; /* number of columns in the foreign key */ 854 AttrNumber conkey[INDEX_MAX_KEYS]; /* cols in referencing table */ 855 AttrNumber confkey[INDEX_MAX_KEYS]; /* cols in referenced table */ 856 Oid conpfeqop[INDEX_MAX_KEYS]; /* PK = FK operator OIDs */ 857 858 /* Derived info about whether FK's equality conditions match the query: */ 859 int nmatched_ec; /* # of FK cols matched by ECs */ 860 int nmatched_rcols; /* # of FK cols matched by non-EC rinfos */ 861 int nmatched_ri; /* total # of non-EC rinfos matched to FK */ 862 /* Pointer to eclass matching each column's condition, if there is one */ 863 struct EquivalenceClass *eclass[INDEX_MAX_KEYS]; 864 /* List of non-EC RestrictInfos matching each column's condition */ 865 List *rinfos[INDEX_MAX_KEYS]; 866 } ForeignKeyOptInfo; 867 868 /* 869 * StatisticExtInfo 870 * Information about extended statistics for planning/optimization 871 * 872 * Each pg_statistic_ext row is represented by one or more nodes of this 873 * type, or even zero if ANALYZE has not computed them. 874 */ 875 typedef struct StatisticExtInfo 876 { 877 NodeTag type; 878 879 Oid statOid; /* OID of the statistics row */ 880 RelOptInfo *rel; /* back-link to statistic's table */ 881 char kind; /* statistics kind of this entry */ 882 Bitmapset *keys; /* attnums of the columns covered */ 883 } StatisticExtInfo; 884 885 /* 886 * EquivalenceClasses 887 * 888 * Whenever we can determine that a mergejoinable equality clause A = B is 889 * not delayed by any outer join, we create an EquivalenceClass containing 890 * the expressions A and B to record this knowledge. If we later find another 891 * equivalence B = C, we add C to the existing EquivalenceClass; this may 892 * require merging two existing EquivalenceClasses. At the end of the qual 893 * distribution process, we have sets of values that are known all transitively 894 * equal to each other, where "equal" is according to the rules of the btree 895 * operator family(s) shown in ec_opfamilies, as well as the collation shown 896 * by ec_collation. (We restrict an EC to contain only equalities whose 897 * operators belong to the same set of opfamilies. This could probably be 898 * relaxed, but for now it's not worth the trouble, since nearly all equality 899 * operators belong to only one btree opclass anyway. Similarly, we suppose 900 * that all or none of the input datatypes are collatable, so that a single 901 * collation value is sufficient.) 902 * 903 * We also use EquivalenceClasses as the base structure for PathKeys, letting 904 * us represent knowledge about different sort orderings being equivalent. 905 * Since every PathKey must reference an EquivalenceClass, we will end up 906 * with single-member EquivalenceClasses whenever a sort key expression has 907 * not been equivalenced to anything else. It is also possible that such an 908 * EquivalenceClass will contain a volatile expression ("ORDER BY random()"), 909 * which is a case that can't arise otherwise since clauses containing 910 * volatile functions are never considered mergejoinable. We mark such 911 * EquivalenceClasses specially to prevent them from being merged with 912 * ordinary EquivalenceClasses. Also, for volatile expressions we have 913 * to be careful to match the EquivalenceClass to the correct targetlist 914 * entry: consider SELECT random() AS a, random() AS b ... ORDER BY b,a. 915 * So we record the SortGroupRef of the originating sort clause. 916 * 917 * We allow equality clauses appearing below the nullable side of an outer join 918 * to form EquivalenceClasses, but these have a slightly different meaning: 919 * the included values might be all NULL rather than all the same non-null 920 * values. See src/backend/optimizer/README for more on that point. 921 * 922 * NB: if ec_merged isn't NULL, this class has been merged into another, and 923 * should be ignored in favor of using the pointed-to class. 924 */ 925 typedef struct EquivalenceClass 926 { 927 NodeTag type; 928 929 List *ec_opfamilies; /* btree operator family OIDs */ 930 Oid ec_collation; /* collation, if datatypes are collatable */ 931 List *ec_members; /* list of EquivalenceMembers */ 932 List *ec_sources; /* list of generating RestrictInfos */ 933 List *ec_derives; /* list of derived RestrictInfos */ 934 Relids ec_relids; /* all relids appearing in ec_members, except 935 * for child members (see below) */ 936 bool ec_has_const; /* any pseudoconstants in ec_members? */ 937 bool ec_has_volatile; /* the (sole) member is a volatile expr */ 938 bool ec_below_outer_join; /* equivalence applies below an OJ */ 939 bool ec_broken; /* failed to generate needed clauses? */ 940 Index ec_sortref; /* originating sortclause label, or 0 */ 941 Index ec_min_security; /* minimum security_level in ec_sources */ 942 Index ec_max_security; /* maximum security_level in ec_sources */ 943 struct EquivalenceClass *ec_merged; /* set if merged into another EC */ 944 } EquivalenceClass; 945 946 /* 947 * If an EC contains a const and isn't below-outer-join, any PathKey depending 948 * on it must be redundant, since there's only one possible value of the key. 949 */ 950 #define EC_MUST_BE_REDUNDANT(eclass) \ 951 ((eclass)->ec_has_const && !(eclass)->ec_below_outer_join) 952 953 /* 954 * EquivalenceMember - one member expression of an EquivalenceClass 955 * 956 * em_is_child signifies that this element was built by transposing a member 957 * for an appendrel parent relation to represent the corresponding expression 958 * for an appendrel child. These members are used for determining the 959 * pathkeys of scans on the child relation and for explicitly sorting the 960 * child when necessary to build a MergeAppend path for the whole appendrel 961 * tree. An em_is_child member has no impact on the properties of the EC as a 962 * whole; in particular the EC's ec_relids field does NOT include the child 963 * relation. An em_is_child member should never be marked em_is_const nor 964 * cause ec_has_const or ec_has_volatile to be set, either. Thus, em_is_child 965 * members are not really full-fledged members of the EC, but just reflections 966 * or doppelgangers of real members. Most operations on EquivalenceClasses 967 * should ignore em_is_child members, and those that don't should test 968 * em_relids to make sure they only consider relevant members. 969 * 970 * em_datatype is usually the same as exprType(em_expr), but can be 971 * different when dealing with a binary-compatible opfamily; in particular 972 * anyarray_ops would never work without this. Use em_datatype when 973 * looking up a specific btree operator to work with this expression. 974 */ 975 typedef struct EquivalenceMember 976 { 977 NodeTag type; 978 979 Expr *em_expr; /* the expression represented */ 980 Relids em_relids; /* all relids appearing in em_expr */ 981 Relids em_nullable_relids; /* nullable by lower outer joins */ 982 bool em_is_const; /* expression is pseudoconstant? */ 983 bool em_is_child; /* derived version for a child relation? */ 984 Oid em_datatype; /* the "nominal type" used by the opfamily */ 985 } EquivalenceMember; 986 987 /* 988 * PathKeys 989 * 990 * The sort ordering of a path is represented by a list of PathKey nodes. 991 * An empty list implies no known ordering. Otherwise the first item 992 * represents the primary sort key, the second the first secondary sort key, 993 * etc. The value being sorted is represented by linking to an 994 * EquivalenceClass containing that value and including pk_opfamily among its 995 * ec_opfamilies. The EquivalenceClass tells which collation to use, too. 996 * This is a convenient method because it makes it trivial to detect 997 * equivalent and closely-related orderings. (See optimizer/README for more 998 * information.) 999 * 1000 * Note: pk_strategy is either BTLessStrategyNumber (for ASC) or 1001 * BTGreaterStrategyNumber (for DESC). We assume that all ordering-capable 1002 * index types will use btree-compatible strategy numbers. 1003 */ 1004 typedef struct PathKey 1005 { 1006 NodeTag type; 1007 1008 EquivalenceClass *pk_eclass; /* the value that is ordered */ 1009 Oid pk_opfamily; /* btree opfamily defining the ordering */ 1010 int pk_strategy; /* sort direction (ASC or DESC) */ 1011 bool pk_nulls_first; /* do NULLs come before normal values? */ 1012 } PathKey; 1013 1014 1015 /* 1016 * PathTarget 1017 * 1018 * This struct contains what we need to know during planning about the 1019 * targetlist (output columns) that a Path will compute. Each RelOptInfo 1020 * includes a default PathTarget, which its individual Paths may simply 1021 * reference. However, in some cases a Path may compute outputs different 1022 * from other Paths, and in that case we make a custom PathTarget for it. 1023 * For example, an indexscan might return index expressions that would 1024 * otherwise need to be explicitly calculated. (Note also that "upper" 1025 * relations generally don't have useful default PathTargets.) 1026 * 1027 * exprs contains bare expressions; they do not have TargetEntry nodes on top, 1028 * though those will appear in finished Plans. 1029 * 1030 * sortgrouprefs[] is an array of the same length as exprs, containing the 1031 * corresponding sort/group refnos, or zeroes for expressions not referenced 1032 * by sort/group clauses. If sortgrouprefs is NULL (which it generally is in 1033 * RelOptInfo.reltarget targets; only upper-level Paths contain this info), 1034 * we have not identified sort/group columns in this tlist. This allows us to 1035 * deal with sort/group refnos when needed with less expense than including 1036 * TargetEntry nodes in the exprs list. 1037 */ 1038 typedef struct PathTarget 1039 { 1040 NodeTag type; 1041 List *exprs; /* list of expressions to be computed */ 1042 Index *sortgrouprefs; /* corresponding sort/group refnos, or 0 */ 1043 QualCost cost; /* cost of evaluating the expressions */ 1044 int width; /* estimated avg width of result tuples */ 1045 } PathTarget; 1046 1047 /* Convenience macro to get a sort/group refno from a PathTarget */ 1048 #define get_pathtarget_sortgroupref(target, colno) \ 1049 ((target)->sortgrouprefs ? (target)->sortgrouprefs[colno] : (Index) 0) 1050 1051 1052 /* 1053 * ParamPathInfo 1054 * 1055 * All parameterized paths for a given relation with given required outer rels 1056 * link to a single ParamPathInfo, which stores common information such as 1057 * the estimated rowcount for this parameterization. We do this partly to 1058 * avoid recalculations, but mostly to ensure that the estimated rowcount 1059 * is in fact the same for every such path. 1060 * 1061 * Note: ppi_clauses is only used in ParamPathInfos for base relation paths; 1062 * in join cases it's NIL because the set of relevant clauses varies depending 1063 * on how the join is formed. The relevant clauses will appear in each 1064 * parameterized join path's joinrestrictinfo list, instead. 1065 */ 1066 typedef struct ParamPathInfo 1067 { 1068 NodeTag type; 1069 1070 Relids ppi_req_outer; /* rels supplying parameters used by path */ 1071 double ppi_rows; /* estimated number of result tuples */ 1072 List *ppi_clauses; /* join clauses available from outer rels */ 1073 } ParamPathInfo; 1074 1075 1076 /* 1077 * Type "Path" is used as-is for sequential-scan paths, as well as some other 1078 * simple plan types that we don't need any extra information in the path for. 1079 * For other path types it is the first component of a larger struct. 1080 * 1081 * "pathtype" is the NodeTag of the Plan node we could build from this Path. 1082 * It is partially redundant with the Path's NodeTag, but allows us to use 1083 * the same Path type for multiple Plan types when there is no need to 1084 * distinguish the Plan type during path processing. 1085 * 1086 * "parent" identifies the relation this Path scans, and "pathtarget" 1087 * describes the precise set of output columns the Path would compute. 1088 * In simple cases all Paths for a given rel share the same targetlist, 1089 * which we represent by having path->pathtarget equal to parent->reltarget. 1090 * 1091 * "param_info", if not NULL, links to a ParamPathInfo that identifies outer 1092 * relation(s) that provide parameter values to each scan of this path. 1093 * That means this path can only be joined to those rels by means of nestloop 1094 * joins with this path on the inside. Also note that a parameterized path 1095 * is responsible for testing all "movable" joinclauses involving this rel 1096 * and the specified outer rel(s). 1097 * 1098 * "rows" is the same as parent->rows in simple paths, but in parameterized 1099 * paths and UniquePaths it can be less than parent->rows, reflecting the 1100 * fact that we've filtered by extra join conditions or removed duplicates. 1101 * 1102 * "pathkeys" is a List of PathKey nodes (see above), describing the sort 1103 * ordering of the path's output rows. 1104 */ 1105 typedef struct Path 1106 { 1107 NodeTag type; 1108 1109 NodeTag pathtype; /* tag identifying scan/join method */ 1110 1111 RelOptInfo *parent; /* the relation this path can build */ 1112 PathTarget *pathtarget; /* list of Vars/Exprs, cost, width */ 1113 1114 ParamPathInfo *param_info; /* parameterization info, or NULL if none */ 1115 1116 bool parallel_aware; /* engage parallel-aware logic? */ 1117 bool parallel_safe; /* OK to use as part of parallel plan? */ 1118 int parallel_workers; /* desired # of workers; 0 = not parallel */ 1119 1120 /* estimated size/costs for path (see costsize.c for more info) */ 1121 double rows; /* estimated number of result tuples */ 1122 Cost startup_cost; /* cost expended before fetching any tuples */ 1123 Cost total_cost; /* total cost (assuming all tuples fetched) */ 1124 1125 List *pathkeys; /* sort ordering of path's output */ 1126 /* pathkeys is a List of PathKey nodes; see above */ 1127 } Path; 1128 1129 /* Macro for extracting a path's parameterization relids; beware double eval */ 1130 #define PATH_REQ_OUTER(path) \ 1131 ((path)->param_info ? (path)->param_info->ppi_req_outer : (Relids) NULL) 1132 1133 /*---------- 1134 * IndexPath represents an index scan over a single index. 1135 * 1136 * This struct is used for both regular indexscans and index-only scans; 1137 * path.pathtype is T_IndexScan or T_IndexOnlyScan to show which is meant. 1138 * 1139 * 'indexinfo' is the index to be scanned. 1140 * 1141 * 'indexclauses' is a list of IndexClause nodes, each representing one 1142 * index-checkable restriction, with implicit AND semantics across the list. 1143 * An empty list implies a full index scan. 1144 * 1145 * 'indexorderbys', if not NIL, is a list of ORDER BY expressions that have 1146 * been found to be usable as ordering operators for an amcanorderbyop index. 1147 * The list must match the path's pathkeys, ie, one expression per pathkey 1148 * in the same order. These are not RestrictInfos, just bare expressions, 1149 * since they generally won't yield booleans. It's guaranteed that each 1150 * expression has the index key on the left side of the operator. 1151 * 1152 * 'indexorderbycols' is an integer list of index column numbers (zero-based) 1153 * of the same length as 'indexorderbys', showing which index column each 1154 * ORDER BY expression is meant to be used with. (There is no restriction 1155 * on which index column each ORDER BY can be used with.) 1156 * 1157 * 'indexscandir' is one of: 1158 * ForwardScanDirection: forward scan of an ordered index 1159 * BackwardScanDirection: backward scan of an ordered index 1160 * NoMovementScanDirection: scan of an unordered index, or don't care 1161 * (The executor doesn't care whether it gets ForwardScanDirection or 1162 * NoMovementScanDirection for an indexscan, but the planner wants to 1163 * distinguish ordered from unordered indexes for building pathkeys.) 1164 * 1165 * 'indextotalcost' and 'indexselectivity' are saved in the IndexPath so that 1166 * we need not recompute them when considering using the same index in a 1167 * bitmap index/heap scan (see BitmapHeapPath). The costs of the IndexPath 1168 * itself represent the costs of an IndexScan or IndexOnlyScan plan type. 1169 *---------- 1170 */ 1171 typedef struct IndexPath 1172 { 1173 Path path; 1174 IndexOptInfo *indexinfo; 1175 List *indexclauses; 1176 List *indexorderbys; 1177 List *indexorderbycols; 1178 ScanDirection indexscandir; 1179 Cost indextotalcost; 1180 Selectivity indexselectivity; 1181 } IndexPath; 1182 1183 /* 1184 * Each IndexClause references a RestrictInfo node from the query's WHERE 1185 * or JOIN conditions, and shows how that restriction can be applied to 1186 * the particular index. We support both indexclauses that are directly 1187 * usable by the index machinery, which are typically of the form 1188 * "indexcol OP pseudoconstant", and those from which an indexable qual 1189 * can be derived. The simplest such transformation is that a clause 1190 * of the form "pseudoconstant OP indexcol" can be commuted to produce an 1191 * indexable qual (the index machinery expects the indexcol to be on the 1192 * left always). Another example is that we might be able to extract an 1193 * indexable range condition from a LIKE condition, as in "x LIKE 'foo%bar'" 1194 * giving rise to "x >= 'foo' AND x < 'fop'". Derivation of such lossy 1195 * conditions is done by a planner support function attached to the 1196 * indexclause's top-level function or operator. 1197 * 1198 * indexquals is a list of RestrictInfos for the directly-usable index 1199 * conditions associated with this IndexClause. In the simplest case 1200 * it's a one-element list whose member is iclause->rinfo. Otherwise, 1201 * it contains one or more directly-usable indexqual conditions extracted 1202 * from the given clause. The 'lossy' flag indicates whether the 1203 * indexquals are semantically equivalent to the original clause, or 1204 * represent a weaker condition. 1205 * 1206 * Normally, indexcol is the index of the single index column the clause 1207 * works on, and indexcols is NIL. But if the clause is a RowCompareExpr, 1208 * indexcol is the index of the leading column, and indexcols is a list of 1209 * all the affected columns. (Note that indexcols matches up with the 1210 * columns of the actual indexable RowCompareExpr in indexquals, which 1211 * might be different from the original in rinfo.) 1212 * 1213 * An IndexPath's IndexClause list is required to be ordered by index 1214 * column, i.e. the indexcol values must form a nondecreasing sequence. 1215 * (The order of multiple clauses for the same index column is unspecified.) 1216 */ 1217 typedef struct IndexClause 1218 { 1219 NodeTag type; 1220 struct RestrictInfo *rinfo; /* original restriction or join clause */ 1221 List *indexquals; /* indexqual(s) derived from it */ 1222 bool lossy; /* are indexquals a lossy version of clause? */ 1223 AttrNumber indexcol; /* index column the clause uses (zero-based) */ 1224 List *indexcols; /* multiple index columns, if RowCompare */ 1225 } IndexClause; 1226 1227 /* 1228 * BitmapHeapPath represents one or more indexscans that generate TID bitmaps 1229 * instead of directly accessing the heap, followed by AND/OR combinations 1230 * to produce a single bitmap, followed by a heap scan that uses the bitmap. 1231 * Note that the output is always considered unordered, since it will come 1232 * out in physical heap order no matter what the underlying indexes did. 1233 * 1234 * The individual indexscans are represented by IndexPath nodes, and any 1235 * logic on top of them is represented by a tree of BitmapAndPath and 1236 * BitmapOrPath nodes. Notice that we can use the same IndexPath node both 1237 * to represent a regular (or index-only) index scan plan, and as the child 1238 * of a BitmapHeapPath that represents scanning the same index using a 1239 * BitmapIndexScan. The startup_cost and total_cost figures of an IndexPath 1240 * always represent the costs to use it as a regular (or index-only) 1241 * IndexScan. The costs of a BitmapIndexScan can be computed using the 1242 * IndexPath's indextotalcost and indexselectivity. 1243 */ 1244 typedef struct BitmapHeapPath 1245 { 1246 Path path; 1247 Path *bitmapqual; /* IndexPath, BitmapAndPath, BitmapOrPath */ 1248 } BitmapHeapPath; 1249 1250 /* 1251 * BitmapAndPath represents a BitmapAnd plan node; it can only appear as 1252 * part of the substructure of a BitmapHeapPath. The Path structure is 1253 * a bit more heavyweight than we really need for this, but for simplicity 1254 * we make it a derivative of Path anyway. 1255 */ 1256 typedef struct BitmapAndPath 1257 { 1258 Path path; 1259 List *bitmapquals; /* IndexPaths and BitmapOrPaths */ 1260 Selectivity bitmapselectivity; 1261 } BitmapAndPath; 1262 1263 /* 1264 * BitmapOrPath represents a BitmapOr plan node; it can only appear as 1265 * part of the substructure of a BitmapHeapPath. The Path structure is 1266 * a bit more heavyweight than we really need for this, but for simplicity 1267 * we make it a derivative of Path anyway. 1268 */ 1269 typedef struct BitmapOrPath 1270 { 1271 Path path; 1272 List *bitmapquals; /* IndexPaths and BitmapAndPaths */ 1273 Selectivity bitmapselectivity; 1274 } BitmapOrPath; 1275 1276 /* 1277 * TidPath represents a scan by TID 1278 * 1279 * tidquals is an implicitly OR'ed list of qual expressions of the form 1280 * "CTID = pseudoconstant", or "CTID = ANY(pseudoconstant_array)", 1281 * or a CurrentOfExpr for the relation. 1282 */ 1283 typedef struct TidPath 1284 { 1285 Path path; 1286 List *tidquals; /* qual(s) involving CTID = something */ 1287 } TidPath; 1288 1289 /* 1290 * SubqueryScanPath represents a scan of an unflattened subquery-in-FROM 1291 * 1292 * Note that the subpath comes from a different planning domain; for example 1293 * RTE indexes within it mean something different from those known to the 1294 * SubqueryScanPath. path.parent->subroot is the planning context needed to 1295 * interpret the subpath. 1296 */ 1297 typedef struct SubqueryScanPath 1298 { 1299 Path path; 1300 Path *subpath; /* path representing subquery execution */ 1301 } SubqueryScanPath; 1302 1303 /* 1304 * ForeignPath represents a potential scan of a foreign table, foreign join 1305 * or foreign upper-relation. 1306 * 1307 * fdw_private stores FDW private data about the scan. While fdw_private is 1308 * not actually touched by the core code during normal operations, it's 1309 * generally a good idea to use a representation that can be dumped by 1310 * nodeToString(), so that you can examine the structure during debugging 1311 * with tools like pprint(). 1312 */ 1313 typedef struct ForeignPath 1314 { 1315 Path path; 1316 Path *fdw_outerpath; 1317 List *fdw_private; 1318 } ForeignPath; 1319 1320 /* 1321 * CustomPath represents a table scan done by some out-of-core extension. 1322 * 1323 * We provide a set of hooks here - which the provider must take care to set 1324 * up correctly - to allow extensions to supply their own methods of scanning 1325 * a relation. For example, a provider might provide GPU acceleration, a 1326 * cache-based scan, or some other kind of logic we haven't dreamed up yet. 1327 * 1328 * CustomPaths can be injected into the planning process for a relation by 1329 * set_rel_pathlist_hook functions. 1330 * 1331 * Core code must avoid assuming that the CustomPath is only as large as 1332 * the structure declared here; providers are allowed to make it the first 1333 * element in a larger structure. (Since the planner never copies Paths, 1334 * this doesn't add any complication.) However, for consistency with the 1335 * FDW case, we provide a "custom_private" field in CustomPath; providers 1336 * may prefer to use that rather than define another struct type. 1337 */ 1338 1339 struct CustomPathMethods; 1340 1341 typedef struct CustomPath 1342 { 1343 Path path; 1344 uint32 flags; /* mask of CUSTOMPATH_* flags, see 1345 * nodes/extensible.h */ 1346 List *custom_paths; /* list of child Path nodes, if any */ 1347 List *custom_private; 1348 const struct CustomPathMethods *methods; 1349 } CustomPath; 1350 1351 /* 1352 * AppendPath represents an Append plan, ie, successive execution of 1353 * several member plans. 1354 * 1355 * For partial Append, 'subpaths' contains non-partial subpaths followed by 1356 * partial subpaths. 1357 * 1358 * Note: it is possible for "subpaths" to contain only one, or even no, 1359 * elements. These cases are optimized during create_append_plan. 1360 * In particular, an AppendPath with no subpaths is a "dummy" path that 1361 * is created to represent the case that a relation is provably empty. 1362 * (This is a convenient representation because it means that when we build 1363 * an appendrel and find that all its children have been excluded, no extra 1364 * action is needed to recognize the relation as dummy.) 1365 */ 1366 typedef struct AppendPath 1367 { 1368 Path path; 1369 /* RT indexes of non-leaf tables in a partition tree */ 1370 List *partitioned_rels; 1371 List *subpaths; /* list of component Paths */ 1372 /* Index of first partial path in subpaths; list_length(subpaths) if none */ 1373 int first_partial_path; 1374 double limit_tuples; /* hard limit on output tuples, or -1 */ 1375 } AppendPath; 1376 1377 #define IS_DUMMY_APPEND(p) \ 1378 (IsA((p), AppendPath) && ((AppendPath *) (p))->subpaths == NIL) 1379 1380 /* 1381 * A relation that's been proven empty will have one path that is dummy 1382 * (but might have projection paths on top). For historical reasons, 1383 * this is provided as a macro that wraps is_dummy_rel(). 1384 */ 1385 #define IS_DUMMY_REL(r) is_dummy_rel(r) 1386 extern bool is_dummy_rel(RelOptInfo *rel); 1387 1388 /* 1389 * MergeAppendPath represents a MergeAppend plan, ie, the merging of sorted 1390 * results from several member plans to produce similarly-sorted output. 1391 */ 1392 typedef struct MergeAppendPath 1393 { 1394 Path path; 1395 /* RT indexes of non-leaf tables in a partition tree */ 1396 List *partitioned_rels; 1397 List *subpaths; /* list of component Paths */ 1398 double limit_tuples; /* hard limit on output tuples, or -1 */ 1399 } MergeAppendPath; 1400 1401 /* 1402 * GroupResultPath represents use of a Result plan node to compute the 1403 * output of a degenerate GROUP BY case, wherein we know we should produce 1404 * exactly one row, which might then be filtered by a HAVING qual. 1405 * 1406 * Note that quals is a list of bare clauses, not RestrictInfos. 1407 */ 1408 typedef struct GroupResultPath 1409 { 1410 Path path; 1411 List *quals; 1412 } GroupResultPath; 1413 1414 /* 1415 * MaterialPath represents use of a Material plan node, i.e., caching of 1416 * the output of its subpath. This is used when the subpath is expensive 1417 * and needs to be scanned repeatedly, or when we need mark/restore ability 1418 * and the subpath doesn't have it. 1419 */ 1420 typedef struct MaterialPath 1421 { 1422 Path path; 1423 Path *subpath; 1424 } MaterialPath; 1425 1426 /* 1427 * UniquePath represents elimination of distinct rows from the output of 1428 * its subpath. 1429 * 1430 * This can represent significantly different plans: either hash-based or 1431 * sort-based implementation, or a no-op if the input path can be proven 1432 * distinct already. The decision is sufficiently localized that it's not 1433 * worth having separate Path node types. (Note: in the no-op case, we could 1434 * eliminate the UniquePath node entirely and just return the subpath; but 1435 * it's convenient to have a UniquePath in the path tree to signal upper-level 1436 * routines that the input is known distinct.) 1437 */ 1438 typedef enum 1439 { 1440 UNIQUE_PATH_NOOP, /* input is known unique already */ 1441 UNIQUE_PATH_HASH, /* use hashing */ 1442 UNIQUE_PATH_SORT /* use sorting */ 1443 } UniquePathMethod; 1444 1445 typedef struct UniquePath 1446 { 1447 Path path; 1448 Path *subpath; 1449 UniquePathMethod umethod; 1450 List *in_operators; /* equality operators of the IN clause */ 1451 List *uniq_exprs; /* expressions to be made unique */ 1452 } UniquePath; 1453 1454 /* 1455 * GatherPath runs several copies of a plan in parallel and collects the 1456 * results. The parallel leader may also execute the plan, unless the 1457 * single_copy flag is set. 1458 */ 1459 typedef struct GatherPath 1460 { 1461 Path path; 1462 Path *subpath; /* path for each worker */ 1463 bool single_copy; /* don't execute path more than once */ 1464 int num_workers; /* number of workers sought to help */ 1465 } GatherPath; 1466 1467 /* 1468 * GatherMergePath runs several copies of a plan in parallel and collects 1469 * the results, preserving their common sort order. 1470 */ 1471 typedef struct GatherMergePath 1472 { 1473 Path path; 1474 Path *subpath; /* path for each worker */ 1475 int num_workers; /* number of workers sought to help */ 1476 } GatherMergePath; 1477 1478 1479 /* 1480 * All join-type paths share these fields. 1481 */ 1482 1483 typedef struct JoinPath 1484 { 1485 Path path; 1486 1487 JoinType jointype; 1488 1489 bool inner_unique; /* each outer tuple provably matches no more 1490 * than one inner tuple */ 1491 1492 Path *outerjoinpath; /* path for the outer side of the join */ 1493 Path *innerjoinpath; /* path for the inner side of the join */ 1494 1495 List *joinrestrictinfo; /* RestrictInfos to apply to join */ 1496 1497 /* 1498 * See the notes for RelOptInfo and ParamPathInfo to understand why 1499 * joinrestrictinfo is needed in JoinPath, and can't be merged into the 1500 * parent RelOptInfo. 1501 */ 1502 } JoinPath; 1503 1504 /* 1505 * A nested-loop path needs no special fields. 1506 */ 1507 1508 typedef JoinPath NestPath; 1509 1510 /* 1511 * A mergejoin path has these fields. 1512 * 1513 * Unlike other path types, a MergePath node doesn't represent just a single 1514 * run-time plan node: it can represent up to four. Aside from the MergeJoin 1515 * node itself, there can be a Sort node for the outer input, a Sort node 1516 * for the inner input, and/or a Material node for the inner input. We could 1517 * represent these nodes by separate path nodes, but considering how many 1518 * different merge paths are investigated during a complex join problem, 1519 * it seems better to avoid unnecessary palloc overhead. 1520 * 1521 * path_mergeclauses lists the clauses (in the form of RestrictInfos) 1522 * that will be used in the merge. 1523 * 1524 * Note that the mergeclauses are a subset of the parent relation's 1525 * restriction-clause list. Any join clauses that are not mergejoinable 1526 * appear only in the parent's restrict list, and must be checked by a 1527 * qpqual at execution time. 1528 * 1529 * outersortkeys (resp. innersortkeys) is NIL if the outer path 1530 * (resp. inner path) is already ordered appropriately for the 1531 * mergejoin. If it is not NIL then it is a PathKeys list describing 1532 * the ordering that must be created by an explicit Sort node. 1533 * 1534 * skip_mark_restore is true if the executor need not do mark/restore calls. 1535 * Mark/restore overhead is usually required, but can be skipped if we know 1536 * that the executor need find only one match per outer tuple, and that the 1537 * mergeclauses are sufficient to identify a match. In such cases the 1538 * executor can immediately advance the outer relation after processing a 1539 * match, and therefore it need never back up the inner relation. 1540 * 1541 * materialize_inner is true if a Material node should be placed atop the 1542 * inner input. This may appear with or without an inner Sort step. 1543 */ 1544 1545 typedef struct MergePath 1546 { 1547 JoinPath jpath; 1548 List *path_mergeclauses; /* join clauses to be used for merge */ 1549 List *outersortkeys; /* keys for explicit sort, if any */ 1550 List *innersortkeys; /* keys for explicit sort, if any */ 1551 bool skip_mark_restore; /* can executor skip mark/restore? */ 1552 bool materialize_inner; /* add Materialize to inner? */ 1553 } MergePath; 1554 1555 /* 1556 * A hashjoin path has these fields. 1557 * 1558 * The remarks above for mergeclauses apply for hashclauses as well. 1559 * 1560 * Hashjoin does not care what order its inputs appear in, so we have 1561 * no need for sortkeys. 1562 */ 1563 1564 typedef struct HashPath 1565 { 1566 JoinPath jpath; 1567 List *path_hashclauses; /* join clauses used for hashing */ 1568 int num_batches; /* number of batches expected */ 1569 double inner_rows_total; /* total inner rows expected */ 1570 } HashPath; 1571 1572 /* 1573 * ProjectionPath represents a projection (that is, targetlist computation) 1574 * 1575 * Nominally, this path node represents using a Result plan node to do a 1576 * projection step. However, if the input plan node supports projection, 1577 * we can just modify its output targetlist to do the required calculations 1578 * directly, and not need a Result. In some places in the planner we can just 1579 * jam the desired PathTarget into the input path node (and adjust its cost 1580 * accordingly), so we don't need a ProjectionPath. But in other places 1581 * it's necessary to not modify the input path node, so we need a separate 1582 * ProjectionPath node, which is marked dummy to indicate that we intend to 1583 * assign the work to the input plan node. The estimated cost for the 1584 * ProjectionPath node will account for whether a Result will be used or not. 1585 */ 1586 typedef struct ProjectionPath 1587 { 1588 Path path; 1589 Path *subpath; /* path representing input source */ 1590 bool dummypp; /* true if no separate Result is needed */ 1591 } ProjectionPath; 1592 1593 /* 1594 * ProjectSetPath represents evaluation of a targetlist that includes 1595 * set-returning function(s), which will need to be implemented by a 1596 * ProjectSet plan node. 1597 */ 1598 typedef struct ProjectSetPath 1599 { 1600 Path path; 1601 Path *subpath; /* path representing input source */ 1602 } ProjectSetPath; 1603 1604 /* 1605 * SortPath represents an explicit sort step 1606 * 1607 * The sort keys are, by definition, the same as path.pathkeys. 1608 * 1609 * Note: the Sort plan node cannot project, so path.pathtarget must be the 1610 * same as the input's pathtarget. 1611 */ 1612 typedef struct SortPath 1613 { 1614 Path path; 1615 Path *subpath; /* path representing input source */ 1616 } SortPath; 1617 1618 /* 1619 * GroupPath represents grouping (of presorted input) 1620 * 1621 * groupClause represents the columns to be grouped on; the input path 1622 * must be at least that well sorted. 1623 * 1624 * We can also apply a qual to the grouped rows (equivalent of HAVING) 1625 */ 1626 typedef struct GroupPath 1627 { 1628 Path path; 1629 Path *subpath; /* path representing input source */ 1630 List *groupClause; /* a list of SortGroupClause's */ 1631 List *qual; /* quals (HAVING quals), if any */ 1632 } GroupPath; 1633 1634 /* 1635 * UpperUniquePath represents adjacent-duplicate removal (in presorted input) 1636 * 1637 * The columns to be compared are the first numkeys columns of the path's 1638 * pathkeys. The input is presumed already sorted that way. 1639 */ 1640 typedef struct UpperUniquePath 1641 { 1642 Path path; 1643 Path *subpath; /* path representing input source */ 1644 int numkeys; /* number of pathkey columns to compare */ 1645 } UpperUniquePath; 1646 1647 /* 1648 * AggPath represents generic computation of aggregate functions 1649 * 1650 * This may involve plain grouping (but not grouping sets), using either 1651 * sorted or hashed grouping; for the AGG_SORTED case, the input must be 1652 * appropriately presorted. 1653 */ 1654 typedef struct AggPath 1655 { 1656 Path path; 1657 Path *subpath; /* path representing input source */ 1658 AggStrategy aggstrategy; /* basic strategy, see nodes.h */ 1659 AggSplit aggsplit; /* agg-splitting mode, see nodes.h */ 1660 double numGroups; /* estimated number of groups in input */ 1661 List *groupClause; /* a list of SortGroupClause's */ 1662 List *qual; /* quals (HAVING quals), if any */ 1663 } AggPath; 1664 1665 /* 1666 * Various annotations used for grouping sets in the planner. 1667 */ 1668 1669 typedef struct GroupingSetData 1670 { 1671 NodeTag type; 1672 List *set; /* grouping set as list of sortgrouprefs */ 1673 double numGroups; /* est. number of result groups */ 1674 } GroupingSetData; 1675 1676 typedef struct RollupData 1677 { 1678 NodeTag type; 1679 List *groupClause; /* applicable subset of parse->groupClause */ 1680 List *gsets; /* lists of integer indexes into groupClause */ 1681 List *gsets_data; /* list of GroupingSetData */ 1682 double numGroups; /* est. number of result groups */ 1683 bool hashable; /* can be hashed */ 1684 bool is_hashed; /* to be implemented as a hashagg */ 1685 } RollupData; 1686 1687 /* 1688 * GroupingSetsPath represents a GROUPING SETS aggregation 1689 */ 1690 1691 typedef struct GroupingSetsPath 1692 { 1693 Path path; 1694 Path *subpath; /* path representing input source */ 1695 AggStrategy aggstrategy; /* basic strategy */ 1696 List *rollups; /* list of RollupData */ 1697 List *qual; /* quals (HAVING quals), if any */ 1698 } GroupingSetsPath; 1699 1700 /* 1701 * MinMaxAggPath represents computation of MIN/MAX aggregates from indexes 1702 */ 1703 typedef struct MinMaxAggPath 1704 { 1705 Path path; 1706 List *mmaggregates; /* list of MinMaxAggInfo */ 1707 List *quals; /* HAVING quals, if any */ 1708 } MinMaxAggPath; 1709 1710 /* 1711 * WindowAggPath represents generic computation of window functions 1712 */ 1713 typedef struct WindowAggPath 1714 { 1715 Path path; 1716 Path *subpath; /* path representing input source */ 1717 WindowClause *winclause; /* WindowClause we'll be using */ 1718 } WindowAggPath; 1719 1720 /* 1721 * SetOpPath represents a set-operation, that is INTERSECT or EXCEPT 1722 */ 1723 typedef struct SetOpPath 1724 { 1725 Path path; 1726 Path *subpath; /* path representing input source */ 1727 SetOpCmd cmd; /* what to do, see nodes.h */ 1728 SetOpStrategy strategy; /* how to do it, see nodes.h */ 1729 List *distinctList; /* SortGroupClauses identifying target cols */ 1730 AttrNumber flagColIdx; /* where is the flag column, if any */ 1731 int firstFlag; /* flag value for first input relation */ 1732 double numGroups; /* estimated number of groups in input */ 1733 } SetOpPath; 1734 1735 /* 1736 * RecursiveUnionPath represents a recursive UNION node 1737 */ 1738 typedef struct RecursiveUnionPath 1739 { 1740 Path path; 1741 Path *leftpath; /* paths representing input sources */ 1742 Path *rightpath; 1743 List *distinctList; /* SortGroupClauses identifying target cols */ 1744 int wtParam; /* ID of Param representing work table */ 1745 double numGroups; /* estimated number of groups in input */ 1746 } RecursiveUnionPath; 1747 1748 /* 1749 * LockRowsPath represents acquiring row locks for SELECT FOR UPDATE/SHARE 1750 */ 1751 typedef struct LockRowsPath 1752 { 1753 Path path; 1754 Path *subpath; /* path representing input source */ 1755 List *rowMarks; /* a list of PlanRowMark's */ 1756 int epqParam; /* ID of Param for EvalPlanQual re-eval */ 1757 } LockRowsPath; 1758 1759 /* 1760 * ModifyTablePath represents performing INSERT/UPDATE/DELETE modifications 1761 * 1762 * We represent most things that will be in the ModifyTable plan node 1763 * literally, except we have child Path(s) not Plan(s). But analysis of the 1764 * OnConflictExpr is deferred to createplan.c, as is collection of FDW data. 1765 */ 1766 typedef struct ModifyTablePath 1767 { 1768 Path path; 1769 CmdType operation; /* INSERT, UPDATE, or DELETE */ 1770 bool canSetTag; /* do we set the command tag/es_processed? */ 1771 Index nominalRelation; /* Parent RT index for use of EXPLAIN */ 1772 Index rootRelation; /* Root RT index, if target is partitioned */ 1773 bool partColsUpdated; /* some part key in hierarchy updated */ 1774 List *resultRelations; /* integer list of RT indexes */ 1775 List *subpaths; /* Path(s) producing source data */ 1776 List *subroots; /* per-target-table PlannerInfos */ 1777 List *withCheckOptionLists; /* per-target-table WCO lists */ 1778 List *returningLists; /* per-target-table RETURNING tlists */ 1779 List *rowMarks; /* PlanRowMarks (non-locking only) */ 1780 OnConflictExpr *onconflict; /* ON CONFLICT clause, or NULL */ 1781 int epqParam; /* ID of Param for EvalPlanQual re-eval */ 1782 } ModifyTablePath; 1783 1784 /* 1785 * LimitPath represents applying LIMIT/OFFSET restrictions 1786 */ 1787 typedef struct LimitPath 1788 { 1789 Path path; 1790 Path *subpath; /* path representing input source */ 1791 Node *limitOffset; /* OFFSET parameter, or NULL if none */ 1792 Node *limitCount; /* COUNT parameter, or NULL if none */ 1793 } LimitPath; 1794 1795 1796 /* 1797 * Restriction clause info. 1798 * 1799 * We create one of these for each AND sub-clause of a restriction condition 1800 * (WHERE or JOIN/ON clause). Since the restriction clauses are logically 1801 * ANDed, we can use any one of them or any subset of them to filter out 1802 * tuples, without having to evaluate the rest. The RestrictInfo node itself 1803 * stores data used by the optimizer while choosing the best query plan. 1804 * 1805 * If a restriction clause references a single base relation, it will appear 1806 * in the baserestrictinfo list of the RelOptInfo for that base rel. 1807 * 1808 * If a restriction clause references more than one base rel, it will 1809 * appear in the joininfo list of every RelOptInfo that describes a strict 1810 * subset of the base rels mentioned in the clause. The joininfo lists are 1811 * used to drive join tree building by selecting plausible join candidates. 1812 * The clause cannot actually be applied until we have built a join rel 1813 * containing all the base rels it references, however. 1814 * 1815 * When we construct a join rel that includes all the base rels referenced 1816 * in a multi-relation restriction clause, we place that clause into the 1817 * joinrestrictinfo lists of paths for the join rel, if neither left nor 1818 * right sub-path includes all base rels referenced in the clause. The clause 1819 * will be applied at that join level, and will not propagate any further up 1820 * the join tree. (Note: the "predicate migration" code was once intended to 1821 * push restriction clauses up and down the plan tree based on evaluation 1822 * costs, but it's dead code and is unlikely to be resurrected in the 1823 * foreseeable future.) 1824 * 1825 * Note that in the presence of more than two rels, a multi-rel restriction 1826 * might reach different heights in the join tree depending on the join 1827 * sequence we use. So, these clauses cannot be associated directly with 1828 * the join RelOptInfo, but must be kept track of on a per-join-path basis. 1829 * 1830 * RestrictInfos that represent equivalence conditions (i.e., mergejoinable 1831 * equalities that are not outerjoin-delayed) are handled a bit differently. 1832 * Initially we attach them to the EquivalenceClasses that are derived from 1833 * them. When we construct a scan or join path, we look through all the 1834 * EquivalenceClasses and generate derived RestrictInfos representing the 1835 * minimal set of conditions that need to be checked for this particular scan 1836 * or join to enforce that all members of each EquivalenceClass are in fact 1837 * equal in all rows emitted by the scan or join. 1838 * 1839 * When dealing with outer joins we have to be very careful about pushing qual 1840 * clauses up and down the tree. An outer join's own JOIN/ON conditions must 1841 * be evaluated exactly at that join node, unless they are "degenerate" 1842 * conditions that reference only Vars from the nullable side of the join. 1843 * Quals appearing in WHERE or in a JOIN above the outer join cannot be pushed 1844 * down below the outer join, if they reference any nullable Vars. 1845 * RestrictInfo nodes contain a flag to indicate whether a qual has been 1846 * pushed down to a lower level than its original syntactic placement in the 1847 * join tree would suggest. If an outer join prevents us from pushing a qual 1848 * down to its "natural" semantic level (the level associated with just the 1849 * base rels used in the qual) then we mark the qual with a "required_relids" 1850 * value including more than just the base rels it actually uses. By 1851 * pretending that the qual references all the rels required to form the outer 1852 * join, we prevent it from being evaluated below the outer join's joinrel. 1853 * When we do form the outer join's joinrel, we still need to distinguish 1854 * those quals that are actually in that join's JOIN/ON condition from those 1855 * that appeared elsewhere in the tree and were pushed down to the join rel 1856 * because they used no other rels. That's what the is_pushed_down flag is 1857 * for; it tells us that a qual is not an OUTER JOIN qual for the set of base 1858 * rels listed in required_relids. A clause that originally came from WHERE 1859 * or an INNER JOIN condition will *always* have its is_pushed_down flag set. 1860 * It's possible for an OUTER JOIN clause to be marked is_pushed_down too, 1861 * if we decide that it can be pushed down into the nullable side of the join. 1862 * In that case it acts as a plain filter qual for wherever it gets evaluated. 1863 * (In short, is_pushed_down is only false for non-degenerate outer join 1864 * conditions. Possibly we should rename it to reflect that meaning? But 1865 * see also the comments for RINFO_IS_PUSHED_DOWN, below.) 1866 * 1867 * RestrictInfo nodes also contain an outerjoin_delayed flag, which is true 1868 * if the clause's applicability must be delayed due to any outer joins 1869 * appearing below it (ie, it has to be postponed to some join level higher 1870 * than the set of relations it actually references). 1871 * 1872 * There is also an outer_relids field, which is NULL except for outer join 1873 * clauses; for those, it is the set of relids on the outer side of the 1874 * clause's outer join. (These are rels that the clause cannot be applied to 1875 * in parameterized scans, since pushing it into the join's outer side would 1876 * lead to wrong answers.) 1877 * 1878 * There is also a nullable_relids field, which is the set of rels the clause 1879 * references that can be forced null by some outer join below the clause. 1880 * 1881 * outerjoin_delayed = true is subtly different from nullable_relids != NULL: 1882 * a clause might reference some nullable rels and yet not be 1883 * outerjoin_delayed because it also references all the other rels of the 1884 * outer join(s). A clause that is not outerjoin_delayed can be enforced 1885 * anywhere it is computable. 1886 * 1887 * To handle security-barrier conditions efficiently, we mark RestrictInfo 1888 * nodes with a security_level field, in which higher values identify clauses 1889 * coming from less-trusted sources. The exact semantics are that a clause 1890 * cannot be evaluated before another clause with a lower security_level value 1891 * unless the first clause is leakproof. As with outer-join clauses, this 1892 * creates a reason for clauses to sometimes need to be evaluated higher in 1893 * the join tree than their contents would suggest; and even at a single plan 1894 * node, this rule constrains the order of application of clauses. 1895 * 1896 * In general, the referenced clause might be arbitrarily complex. The 1897 * kinds of clauses we can handle as indexscan quals, mergejoin clauses, 1898 * or hashjoin clauses are limited (e.g., no volatile functions). The code 1899 * for each kind of path is responsible for identifying the restrict clauses 1900 * it can use and ignoring the rest. Clauses not implemented by an indexscan, 1901 * mergejoin, or hashjoin will be placed in the plan qual or joinqual field 1902 * of the finished Plan node, where they will be enforced by general-purpose 1903 * qual-expression-evaluation code. (But we are still entitled to count 1904 * their selectivity when estimating the result tuple count, if we 1905 * can guess what it is...) 1906 * 1907 * When the referenced clause is an OR clause, we generate a modified copy 1908 * in which additional RestrictInfo nodes are inserted below the top-level 1909 * OR/AND structure. This is a convenience for OR indexscan processing: 1910 * indexquals taken from either the top level or an OR subclause will have 1911 * associated RestrictInfo nodes. 1912 * 1913 * The can_join flag is set true if the clause looks potentially useful as 1914 * a merge or hash join clause, that is if it is a binary opclause with 1915 * nonoverlapping sets of relids referenced in the left and right sides. 1916 * (Whether the operator is actually merge or hash joinable isn't checked, 1917 * however.) 1918 * 1919 * The pseudoconstant flag is set true if the clause contains no Vars of 1920 * the current query level and no volatile functions. Such a clause can be 1921 * pulled out and used as a one-time qual in a gating Result node. We keep 1922 * pseudoconstant clauses in the same lists as other RestrictInfos so that 1923 * the regular clause-pushing machinery can assign them to the correct join 1924 * level, but they need to be treated specially for cost and selectivity 1925 * estimates. Note that a pseudoconstant clause can never be an indexqual 1926 * or merge or hash join clause, so it's of no interest to large parts of 1927 * the planner. 1928 * 1929 * When join clauses are generated from EquivalenceClasses, there may be 1930 * several equally valid ways to enforce join equivalence, of which we need 1931 * apply only one. We mark clauses of this kind by setting parent_ec to 1932 * point to the generating EquivalenceClass. Multiple clauses with the same 1933 * parent_ec in the same join are redundant. 1934 */ 1935 1936 typedef struct RestrictInfo 1937 { 1938 NodeTag type; 1939 1940 Expr *clause; /* the represented clause of WHERE or JOIN */ 1941 1942 bool is_pushed_down; /* true if clause was pushed down in level */ 1943 1944 bool outerjoin_delayed; /* true if delayed by lower outer join */ 1945 1946 bool can_join; /* see comment above */ 1947 1948 bool pseudoconstant; /* see comment above */ 1949 1950 bool leakproof; /* true if known to contain no leaked Vars */ 1951 1952 Index security_level; /* see comment above */ 1953 1954 /* The set of relids (varnos) actually referenced in the clause: */ 1955 Relids clause_relids; 1956 1957 /* The set of relids required to evaluate the clause: */ 1958 Relids required_relids; 1959 1960 /* If an outer-join clause, the outer-side relations, else NULL: */ 1961 Relids outer_relids; 1962 1963 /* The relids used in the clause that are nullable by lower outer joins: */ 1964 Relids nullable_relids; 1965 1966 /* These fields are set for any binary opclause: */ 1967 Relids left_relids; /* relids in left side of clause */ 1968 Relids right_relids; /* relids in right side of clause */ 1969 1970 /* This field is NULL unless clause is an OR clause: */ 1971 Expr *orclause; /* modified clause with RestrictInfos */ 1972 1973 /* This field is NULL unless clause is potentially redundant: */ 1974 EquivalenceClass *parent_ec; /* generating EquivalenceClass */ 1975 1976 /* cache space for cost and selectivity */ 1977 QualCost eval_cost; /* eval cost of clause; -1 if not yet set */ 1978 Selectivity norm_selec; /* selectivity for "normal" (JOIN_INNER) 1979 * semantics; -1 if not yet set; >1 means a 1980 * redundant clause */ 1981 Selectivity outer_selec; /* selectivity for outer join semantics; -1 if 1982 * not yet set */ 1983 1984 /* valid if clause is mergejoinable, else NIL */ 1985 List *mergeopfamilies; /* opfamilies containing clause operator */ 1986 1987 /* cache space for mergeclause processing; NULL if not yet set */ 1988 EquivalenceClass *left_ec; /* EquivalenceClass containing lefthand */ 1989 EquivalenceClass *right_ec; /* EquivalenceClass containing righthand */ 1990 EquivalenceMember *left_em; /* EquivalenceMember for lefthand */ 1991 EquivalenceMember *right_em; /* EquivalenceMember for righthand */ 1992 List *scansel_cache; /* list of MergeScanSelCache structs */ 1993 1994 /* transient workspace for use while considering a specific join path */ 1995 bool outer_is_left; /* T = outer var on left, F = on right */ 1996 1997 /* valid if clause is hashjoinable, else InvalidOid: */ 1998 Oid hashjoinoperator; /* copy of clause operator */ 1999 2000 /* cache space for hashclause processing; -1 if not yet set */ 2001 Selectivity left_bucketsize; /* avg bucketsize of left side */ 2002 Selectivity right_bucketsize; /* avg bucketsize of right side */ 2003 Selectivity left_mcvfreq; /* left side's most common val's freq */ 2004 Selectivity right_mcvfreq; /* right side's most common val's freq */ 2005 } RestrictInfo; 2006 2007 /* 2008 * This macro embodies the correct way to test whether a RestrictInfo is 2009 * "pushed down" to a given outer join, that is, should be treated as a filter 2010 * clause rather than a join clause at that outer join. This is certainly so 2011 * if is_pushed_down is true; but examining that is not sufficient anymore, 2012 * because outer-join clauses will get pushed down to lower outer joins when 2013 * we generate a path for the lower outer join that is parameterized by the 2014 * LHS of the upper one. We can detect such a clause by noting that its 2015 * required_relids exceed the scope of the join. 2016 */ 2017 #define RINFO_IS_PUSHED_DOWN(rinfo, joinrelids) \ 2018 ((rinfo)->is_pushed_down || \ 2019 !bms_is_subset((rinfo)->required_relids, joinrelids)) 2020 2021 /* 2022 * Since mergejoinscansel() is a relatively expensive function, and would 2023 * otherwise be invoked many times while planning a large join tree, 2024 * we go out of our way to cache its results. Each mergejoinable 2025 * RestrictInfo carries a list of the specific sort orderings that have 2026 * been considered for use with it, and the resulting selectivities. 2027 */ 2028 typedef struct MergeScanSelCache 2029 { 2030 /* Ordering details (cache lookup key) */ 2031 Oid opfamily; /* btree opfamily defining the ordering */ 2032 Oid collation; /* collation for the ordering */ 2033 int strategy; /* sort direction (ASC or DESC) */ 2034 bool nulls_first; /* do NULLs come before normal values? */ 2035 /* Results */ 2036 Selectivity leftstartsel; /* first-join fraction for clause left side */ 2037 Selectivity leftendsel; /* last-join fraction for clause left side */ 2038 Selectivity rightstartsel; /* first-join fraction for clause right side */ 2039 Selectivity rightendsel; /* last-join fraction for clause right side */ 2040 } MergeScanSelCache; 2041 2042 /* 2043 * Placeholder node for an expression to be evaluated below the top level 2044 * of a plan tree. This is used during planning to represent the contained 2045 * expression. At the end of the planning process it is replaced by either 2046 * the contained expression or a Var referring to a lower-level evaluation of 2047 * the contained expression. Typically the evaluation occurs below an outer 2048 * join, and Var references above the outer join might thereby yield NULL 2049 * instead of the expression value. 2050 * 2051 * Although the planner treats this as an expression node type, it is not 2052 * recognized by the parser or executor, so we declare it here rather than 2053 * in primnodes.h. 2054 */ 2055 2056 typedef struct PlaceHolderVar 2057 { 2058 Expr xpr; 2059 Expr *phexpr; /* the represented expression */ 2060 Relids phrels; /* base relids syntactically within expr src */ 2061 Index phid; /* ID for PHV (unique within planner run) */ 2062 Index phlevelsup; /* > 0 if PHV belongs to outer query */ 2063 } PlaceHolderVar; 2064 2065 /* 2066 * "Special join" info. 2067 * 2068 * One-sided outer joins constrain the order of joining partially but not 2069 * completely. We flatten such joins into the planner's top-level list of 2070 * relations to join, but record information about each outer join in a 2071 * SpecialJoinInfo struct. These structs are kept in the PlannerInfo node's 2072 * join_info_list. 2073 * 2074 * Similarly, semijoins and antijoins created by flattening IN (subselect) 2075 * and EXISTS(subselect) clauses create partial constraints on join order. 2076 * These are likewise recorded in SpecialJoinInfo structs. 2077 * 2078 * We make SpecialJoinInfos for FULL JOINs even though there is no flexibility 2079 * of planning for them, because this simplifies make_join_rel()'s API. 2080 * 2081 * min_lefthand and min_righthand are the sets of base relids that must be 2082 * available on each side when performing the special join. lhs_strict is 2083 * true if the special join's condition cannot succeed when the LHS variables 2084 * are all NULL (this means that an outer join can commute with upper-level 2085 * outer joins even if it appears in their RHS). We don't bother to set 2086 * lhs_strict for FULL JOINs, however. 2087 * 2088 * It is not valid for either min_lefthand or min_righthand to be empty sets; 2089 * if they were, this would break the logic that enforces join order. 2090 * 2091 * syn_lefthand and syn_righthand are the sets of base relids that are 2092 * syntactically below this special join. (These are needed to help compute 2093 * min_lefthand and min_righthand for higher joins.) 2094 * 2095 * delay_upper_joins is set true if we detect a pushed-down clause that has 2096 * to be evaluated after this join is formed (because it references the RHS). 2097 * Any outer joins that have such a clause and this join in their RHS cannot 2098 * commute with this join, because that would leave noplace to check the 2099 * pushed-down clause. (We don't track this for FULL JOINs, either.) 2100 * 2101 * For a semijoin, we also extract the join operators and their RHS arguments 2102 * and set semi_operators, semi_rhs_exprs, semi_can_btree, and semi_can_hash. 2103 * This is done in support of possibly unique-ifying the RHS, so we don't 2104 * bother unless at least one of semi_can_btree and semi_can_hash can be set 2105 * true. (You might expect that this information would be computed during 2106 * join planning; but it's helpful to have it available during planning of 2107 * parameterized table scans, so we store it in the SpecialJoinInfo structs.) 2108 * 2109 * jointype is never JOIN_RIGHT; a RIGHT JOIN is handled by switching 2110 * the inputs to make it a LEFT JOIN. So the allowed values of jointype 2111 * in a join_info_list member are only LEFT, FULL, SEMI, or ANTI. 2112 * 2113 * For purposes of join selectivity estimation, we create transient 2114 * SpecialJoinInfo structures for regular inner joins; so it is possible 2115 * to have jointype == JOIN_INNER in such a structure, even though this is 2116 * not allowed within join_info_list. We also create transient 2117 * SpecialJoinInfos with jointype == JOIN_INNER for outer joins, since for 2118 * cost estimation purposes it is sometimes useful to know the join size under 2119 * plain innerjoin semantics. Note that lhs_strict, delay_upper_joins, and 2120 * of course the semi_xxx fields are not set meaningfully within such structs. 2121 */ 2122 #ifndef HAVE_SPECIALJOININFO_TYPEDEF 2123 typedef struct SpecialJoinInfo SpecialJoinInfo; 2124 #define HAVE_SPECIALJOININFO_TYPEDEF 1 2125 #endif 2126 2127 struct SpecialJoinInfo 2128 { 2129 NodeTag type; 2130 Relids min_lefthand; /* base relids in minimum LHS for join */ 2131 Relids min_righthand; /* base relids in minimum RHS for join */ 2132 Relids syn_lefthand; /* base relids syntactically within LHS */ 2133 Relids syn_righthand; /* base relids syntactically within RHS */ 2134 JoinType jointype; /* always INNER, LEFT, FULL, SEMI, or ANTI */ 2135 bool lhs_strict; /* joinclause is strict for some LHS rel */ 2136 bool delay_upper_joins; /* can't commute with upper RHS */ 2137 /* Remaining fields are set only for JOIN_SEMI jointype: */ 2138 bool semi_can_btree; /* true if semi_operators are all btree */ 2139 bool semi_can_hash; /* true if semi_operators are all hash */ 2140 List *semi_operators; /* OIDs of equality join operators */ 2141 List *semi_rhs_exprs; /* righthand-side expressions of these ops */ 2142 }; 2143 2144 /* 2145 * Append-relation info. 2146 * 2147 * When we expand an inheritable table or a UNION-ALL subselect into an 2148 * "append relation" (essentially, a list of child RTEs), we build an 2149 * AppendRelInfo for each child RTE. The list of AppendRelInfos indicates 2150 * which child RTEs must be included when expanding the parent, and each node 2151 * carries information needed to translate Vars referencing the parent into 2152 * Vars referencing that child. 2153 * 2154 * These structs are kept in the PlannerInfo node's append_rel_list. 2155 * Note that we just throw all the structs into one list, and scan the 2156 * whole list when desiring to expand any one parent. We could have used 2157 * a more complex data structure (eg, one list per parent), but this would 2158 * be harder to update during operations such as pulling up subqueries, 2159 * and not really any easier to scan. Considering that typical queries 2160 * will not have many different append parents, it doesn't seem worthwhile 2161 * to complicate things. 2162 * 2163 * Note: after completion of the planner prep phase, any given RTE is an 2164 * append parent having entries in append_rel_list if and only if its 2165 * "inh" flag is set. We clear "inh" for plain tables that turn out not 2166 * to have inheritance children, and (in an abuse of the original meaning 2167 * of the flag) we set "inh" for subquery RTEs that turn out to be 2168 * flattenable UNION ALL queries. This lets us avoid useless searches 2169 * of append_rel_list. 2170 * 2171 * Note: the data structure assumes that append-rel members are single 2172 * baserels. This is OK for inheritance, but it prevents us from pulling 2173 * up a UNION ALL member subquery if it contains a join. While that could 2174 * be fixed with a more complex data structure, at present there's not much 2175 * point because no improvement in the plan could result. 2176 */ 2177 2178 typedef struct AppendRelInfo 2179 { 2180 NodeTag type; 2181 2182 /* 2183 * These fields uniquely identify this append relationship. There can be 2184 * (in fact, always should be) multiple AppendRelInfos for the same 2185 * parent_relid, but never more than one per child_relid, since a given 2186 * RTE cannot be a child of more than one append parent. 2187 */ 2188 Index parent_relid; /* RT index of append parent rel */ 2189 Index child_relid; /* RT index of append child rel */ 2190 2191 /* 2192 * For an inheritance appendrel, the parent and child are both regular 2193 * relations, and we store their rowtype OIDs here for use in translating 2194 * whole-row Vars. For a UNION-ALL appendrel, the parent and child are 2195 * both subqueries with no named rowtype, and we store InvalidOid here. 2196 */ 2197 Oid parent_reltype; /* OID of parent's composite type */ 2198 Oid child_reltype; /* OID of child's composite type */ 2199 2200 /* 2201 * The N'th element of this list is a Var or expression representing the 2202 * child column corresponding to the N'th column of the parent. This is 2203 * used to translate Vars referencing the parent rel into references to 2204 * the child. A list element is NULL if it corresponds to a dropped 2205 * column of the parent (this is only possible for inheritance cases, not 2206 * UNION ALL). The list elements are always simple Vars for inheritance 2207 * cases, but can be arbitrary expressions in UNION ALL cases. 2208 * 2209 * Notice we only store entries for user columns (attno > 0). Whole-row 2210 * Vars are special-cased, and system columns (attno < 0) need no special 2211 * translation since their attnos are the same for all tables. 2212 * 2213 * Caution: the Vars have varlevelsup = 0. Be careful to adjust as needed 2214 * when copying into a subquery. 2215 */ 2216 List *translated_vars; /* Expressions in the child's Vars */ 2217 2218 /* 2219 * We store the parent table's OID here for inheritance, or InvalidOid for 2220 * UNION ALL. This is only needed to help in generating error messages if 2221 * an attempt is made to reference a dropped parent column. 2222 */ 2223 Oid parent_reloid; /* OID of parent relation */ 2224 } AppendRelInfo; 2225 2226 /* 2227 * For each distinct placeholder expression generated during planning, we 2228 * store a PlaceHolderInfo node in the PlannerInfo node's placeholder_list. 2229 * This stores info that is needed centrally rather than in each copy of the 2230 * PlaceHolderVar. The phid fields identify which PlaceHolderInfo goes with 2231 * each PlaceHolderVar. Note that phid is unique throughout a planner run, 2232 * not just within a query level --- this is so that we need not reassign ID's 2233 * when pulling a subquery into its parent. 2234 * 2235 * The idea is to evaluate the expression at (only) the ph_eval_at join level, 2236 * then allow it to bubble up like a Var until the ph_needed join level. 2237 * ph_needed has the same definition as attr_needed for a regular Var. 2238 * 2239 * The PlaceHolderVar's expression might contain LATERAL references to vars 2240 * coming from outside its syntactic scope. If so, those rels are *not* 2241 * included in ph_eval_at, but they are recorded in ph_lateral. 2242 * 2243 * Notice that when ph_eval_at is a join rather than a single baserel, the 2244 * PlaceHolderInfo may create constraints on join order: the ph_eval_at join 2245 * has to be formed below any outer joins that should null the PlaceHolderVar. 2246 * 2247 * We create a PlaceHolderInfo only after determining that the PlaceHolderVar 2248 * is actually referenced in the plan tree, so that unreferenced placeholders 2249 * don't result in unnecessary constraints on join order. 2250 */ 2251 2252 typedef struct PlaceHolderInfo 2253 { 2254 NodeTag type; 2255 2256 Index phid; /* ID for PH (unique within planner run) */ 2257 PlaceHolderVar *ph_var; /* copy of PlaceHolderVar tree */ 2258 Relids ph_eval_at; /* lowest level we can evaluate value at */ 2259 Relids ph_lateral; /* relids of contained lateral refs, if any */ 2260 Relids ph_needed; /* highest level the value is needed at */ 2261 int32 ph_width; /* estimated attribute width */ 2262 } PlaceHolderInfo; 2263 2264 /* 2265 * This struct describes one potentially index-optimizable MIN/MAX aggregate 2266 * function. MinMaxAggPath contains a list of these, and if we accept that 2267 * path, the list is stored into root->minmax_aggs for use during setrefs.c. 2268 */ 2269 typedef struct MinMaxAggInfo 2270 { 2271 NodeTag type; 2272 2273 Oid aggfnoid; /* pg_proc Oid of the aggregate */ 2274 Oid aggsortop; /* Oid of its sort operator */ 2275 Expr *target; /* expression we are aggregating on */ 2276 PlannerInfo *subroot; /* modified "root" for planning the subquery */ 2277 Path *path; /* access path for subquery */ 2278 Cost pathcost; /* estimated cost to fetch first row */ 2279 Param *param; /* param for subplan's output */ 2280 } MinMaxAggInfo; 2281 2282 /* 2283 * At runtime, PARAM_EXEC slots are used to pass values around from one plan 2284 * node to another. They can be used to pass values down into subqueries (for 2285 * outer references in subqueries), or up out of subqueries (for the results 2286 * of a subplan), or from a NestLoop plan node into its inner relation (when 2287 * the inner scan is parameterized with values from the outer relation). 2288 * The planner is responsible for assigning nonconflicting PARAM_EXEC IDs to 2289 * the PARAM_EXEC Params it generates. 2290 * 2291 * Outer references are managed via root->plan_params, which is a list of 2292 * PlannerParamItems. While planning a subquery, each parent query level's 2293 * plan_params contains the values required from it by the current subquery. 2294 * During create_plan(), we use plan_params to track values that must be 2295 * passed from outer to inner sides of NestLoop plan nodes. 2296 * 2297 * The item a PlannerParamItem represents can be one of three kinds: 2298 * 2299 * A Var: the slot represents a variable of this level that must be passed 2300 * down because subqueries have outer references to it, or must be passed 2301 * from a NestLoop node to its inner scan. The varlevelsup value in the Var 2302 * will always be zero. 2303 * 2304 * A PlaceHolderVar: this works much like the Var case, except that the 2305 * entry is a PlaceHolderVar node with a contained expression. The PHV 2306 * will have phlevelsup = 0, and the contained expression is adjusted 2307 * to match in level. 2308 * 2309 * An Aggref (with an expression tree representing its argument): the slot 2310 * represents an aggregate expression that is an outer reference for some 2311 * subquery. The Aggref itself has agglevelsup = 0, and its argument tree 2312 * is adjusted to match in level. 2313 * 2314 * Note: we detect duplicate Var and PlaceHolderVar parameters and coalesce 2315 * them into one slot, but we do not bother to do that for Aggrefs. 2316 * The scope of duplicate-elimination only extends across the set of 2317 * parameters passed from one query level into a single subquery, or for 2318 * nestloop parameters across the set of nestloop parameters used in a single 2319 * query level. So there is no possibility of a PARAM_EXEC slot being used 2320 * for conflicting purposes. 2321 * 2322 * In addition, PARAM_EXEC slots are assigned for Params representing outputs 2323 * from subplans (values that are setParam items for those subplans). These 2324 * IDs need not be tracked via PlannerParamItems, since we do not need any 2325 * duplicate-elimination nor later processing of the represented expressions. 2326 * Instead, we just record the assignment of the slot number by appending to 2327 * root->glob->paramExecTypes. 2328 */ 2329 typedef struct PlannerParamItem 2330 { 2331 NodeTag type; 2332 2333 Node *item; /* the Var, PlaceHolderVar, or Aggref */ 2334 int paramId; /* its assigned PARAM_EXEC slot number */ 2335 } PlannerParamItem; 2336 2337 /* 2338 * When making cost estimates for a SEMI/ANTI/inner_unique join, there are 2339 * some correction factors that are needed in both nestloop and hash joins 2340 * to account for the fact that the executor can stop scanning inner rows 2341 * as soon as it finds a match to the current outer row. These numbers 2342 * depend only on the selected outer and inner join relations, not on the 2343 * particular paths used for them, so it's worthwhile to calculate them 2344 * just once per relation pair not once per considered path. This struct 2345 * is filled by compute_semi_anti_join_factors and must be passed along 2346 * to the join cost estimation functions. 2347 * 2348 * outer_match_frac is the fraction of the outer tuples that are 2349 * expected to have at least one match. 2350 * match_count is the average number of matches expected for 2351 * outer tuples that have at least one match. 2352 */ 2353 typedef struct SemiAntiJoinFactors 2354 { 2355 Selectivity outer_match_frac; 2356 Selectivity match_count; 2357 } SemiAntiJoinFactors; 2358 2359 /* 2360 * Struct for extra information passed to subroutines of add_paths_to_joinrel 2361 * 2362 * restrictlist contains all of the RestrictInfo nodes for restriction 2363 * clauses that apply to this join 2364 * mergeclause_list is a list of RestrictInfo nodes for available 2365 * mergejoin clauses in this join 2366 * inner_unique is true if each outer tuple provably matches no more 2367 * than one inner tuple 2368 * sjinfo is extra info about special joins for selectivity estimation 2369 * semifactors is as shown above (only valid for SEMI/ANTI/inner_unique joins) 2370 * param_source_rels are OK targets for parameterization of result paths 2371 */ 2372 typedef struct JoinPathExtraData 2373 { 2374 List *restrictlist; 2375 List *mergeclause_list; 2376 bool inner_unique; 2377 SpecialJoinInfo *sjinfo; 2378 SemiAntiJoinFactors semifactors; 2379 Relids param_source_rels; 2380 } JoinPathExtraData; 2381 2382 /* 2383 * Various flags indicating what kinds of grouping are possible. 2384 * 2385 * GROUPING_CAN_USE_SORT should be set if it's possible to perform 2386 * sort-based implementations of grouping. When grouping sets are in use, 2387 * this will be true if sorting is potentially usable for any of the grouping 2388 * sets, even if it's not usable for all of them. 2389 * 2390 * GROUPING_CAN_USE_HASH should be set if it's possible to perform 2391 * hash-based implementations of grouping. 2392 * 2393 * GROUPING_CAN_PARTIAL_AGG should be set if the aggregation is of a type 2394 * for which we support partial aggregation (not, for example, grouping sets). 2395 * It says nothing about parallel-safety or the availability of suitable paths. 2396 */ 2397 #define GROUPING_CAN_USE_SORT 0x0001 2398 #define GROUPING_CAN_USE_HASH 0x0002 2399 #define GROUPING_CAN_PARTIAL_AGG 0x0004 2400 2401 /* 2402 * What kind of partitionwise aggregation is in use? 2403 * 2404 * PARTITIONWISE_AGGREGATE_NONE: Not used. 2405 * 2406 * PARTITIONWISE_AGGREGATE_FULL: Aggregate each partition separately, and 2407 * append the results. 2408 * 2409 * PARTITIONWISE_AGGREGATE_PARTIAL: Partially aggregate each partition 2410 * separately, append the results, and then finalize aggregation. 2411 */ 2412 typedef enum 2413 { 2414 PARTITIONWISE_AGGREGATE_NONE, 2415 PARTITIONWISE_AGGREGATE_FULL, 2416 PARTITIONWISE_AGGREGATE_PARTIAL 2417 } PartitionwiseAggregateType; 2418 2419 /* 2420 * Struct for extra information passed to subroutines of create_grouping_paths 2421 * 2422 * flags indicating what kinds of grouping are possible. 2423 * partial_costs_set is true if the agg_partial_costs and agg_final_costs 2424 * have been initialized. 2425 * agg_partial_costs gives partial aggregation costs. 2426 * agg_final_costs gives finalization costs. 2427 * target_parallel_safe is true if target is parallel safe. 2428 * havingQual gives list of quals to be applied after aggregation. 2429 * targetList gives list of columns to be projected. 2430 * patype is the type of partitionwise aggregation that is being performed. 2431 */ 2432 typedef struct 2433 { 2434 /* Data which remains constant once set. */ 2435 int flags; 2436 bool partial_costs_set; 2437 AggClauseCosts agg_partial_costs; 2438 AggClauseCosts agg_final_costs; 2439 2440 /* Data which may differ across partitions. */ 2441 bool target_parallel_safe; 2442 Node *havingQual; 2443 List *targetList; 2444 PartitionwiseAggregateType patype; 2445 } GroupPathExtraData; 2446 2447 /* 2448 * Struct for extra information passed to subroutines of grouping_planner 2449 * 2450 * limit_needed is true if we actually need a Limit plan node. 2451 * limit_tuples is an estimated bound on the number of output tuples, 2452 * or -1 if no LIMIT or couldn't estimate. 2453 * count_est and offset_est are the estimated values of the LIMIT and OFFSET 2454 * expressions computed by preprocess_limit() (see comments for 2455 * preprocess_limit() for more information). 2456 */ 2457 typedef struct 2458 { 2459 bool limit_needed; 2460 double limit_tuples; 2461 int64 count_est; 2462 int64 offset_est; 2463 } FinalPathExtraData; 2464 2465 /* 2466 * For speed reasons, cost estimation for join paths is performed in two 2467 * phases: the first phase tries to quickly derive a lower bound for the 2468 * join cost, and then we check if that's sufficient to reject the path. 2469 * If not, we come back for a more refined cost estimate. The first phase 2470 * fills a JoinCostWorkspace struct with its preliminary cost estimates 2471 * and possibly additional intermediate values. The second phase takes 2472 * these values as inputs to avoid repeating work. 2473 * 2474 * (Ideally we'd declare this in cost.h, but it's also needed in pathnode.h, 2475 * so seems best to put it here.) 2476 */ 2477 typedef struct JoinCostWorkspace 2478 { 2479 /* Preliminary cost estimates --- must not be larger than final ones! */ 2480 Cost startup_cost; /* cost expended before fetching any tuples */ 2481 Cost total_cost; /* total cost (assuming all tuples fetched) */ 2482 2483 /* Fields below here should be treated as private to costsize.c */ 2484 Cost run_cost; /* non-startup cost components */ 2485 2486 /* private for cost_nestloop code */ 2487 Cost inner_run_cost; /* also used by cost_mergejoin code */ 2488 Cost inner_rescan_run_cost; 2489 2490 /* private for cost_mergejoin code */ 2491 double outer_rows; 2492 double inner_rows; 2493 double outer_skip_rows; 2494 double inner_skip_rows; 2495 2496 /* private for cost_hashjoin code */ 2497 int numbuckets; 2498 int numbatches; 2499 double inner_rows_total; 2500 } JoinCostWorkspace; 2501 2502 #endif /* PATHNODES_H */ 2503