1 /*-------------------------------------------------------------------------
4 * Definitions for planner's internal data structures.
7 * Portions Copyright (c) 1996-2015, PostgreSQL Global Development Group
8 * Portions Copyright (c) 1994, Regents of the University of California
10 * src/include/nodes/relation.h
12 *-------------------------------------------------------------------------
17 #include "access/sdir.h"
18 #include "lib/stringinfo.h"
19 #include "nodes/params.h"
20 #include "nodes/parsenodes.h"
21 #include "storage/block.h"
26 * Set of relation identifiers (indexes into the rangetable).
28 typedef Bitmapset *Relids;
31 * When looking for a "cheapest path", this enum specifies whether we want
32 * cheapest startup cost or cheapest total cost.
34 typedef enum CostSelector
36 STARTUP_COST, TOTAL_COST
40 * The cost estimate produced by cost_qual_eval() includes both a one-time
41 * (startup) cost, and a per-tuple cost.
43 typedef struct QualCost
45 Cost startup; /* one-time cost */
46 Cost per_tuple; /* per-evaluation cost */
50 * Costing aggregate function execution requires these statistics about
51 * the aggregates to be executed by a given Agg node. Note that the costs
52 * include the execution costs of the aggregates' argument expressions as
53 * well as the aggregate functions themselves.
55 typedef struct AggClauseCosts
57 int numAggs; /* total number of aggregate functions */
58 int numOrderedAggs; /* number w/ DISTINCT/ORDER BY/WITHIN GROUP */
59 QualCost transCost; /* total per-input-row execution costs */
60 Cost finalCost; /* total per-aggregated-row costs */
61 Size transitionSpace; /* space for pass-by-ref transition data */
67 * Global information for planning/optimization
69 * PlannerGlobal holds state for an entire planner invocation; this state
70 * is shared across all levels of sub-Queries that exist in the command being
74 typedef struct PlannerGlobal
78 ParamListInfo boundParams; /* Param values provided to planner() */
80 List *subplans; /* Plans for SubPlan nodes */
82 List *subroots; /* PlannerInfos for SubPlan nodes */
84 Bitmapset *rewindPlanIDs; /* indices of subplans that require REWIND */
86 List *finalrtable; /* "flat" rangetable for executor */
88 List *finalrowmarks; /* "flat" list of PlanRowMarks */
90 List *resultRelations; /* "flat" list of integer RT indexes */
92 List *relationOids; /* OIDs of relations the plan depends on */
94 List *invalItems; /* other dependencies, as PlanInvalItems */
96 int nParamExec; /* number of PARAM_EXEC Params used */
98 Index lastPHId; /* highest PlaceHolderVar ID assigned */
100 Index lastRowMarkId; /* highest PlanRowMark ID assigned */
102 int lastPlanNodeId; /* highest plan node ID assigned */
104 bool transientPlan; /* redo plan when TransactionXmin changes? */
106 bool hasRowSecurity; /* row security applied? */
108 bool parallelModeOK; /* parallel mode potentially OK? */
110 bool parallelModeNeeded; /* parallel mode actually required? */
113 /* macro for fetching the Plan associated with a SubPlan node */
114 #define planner_subplan_get_plan(root, subplan) \
115 ((Plan *) list_nth((root)->glob->subplans, (subplan)->plan_id - 1))
120 * Per-query information for planning/optimization
122 * This struct is conventionally called "root" in all the planner routines.
123 * It holds links to all of the planner's working state, in addition to the
124 * original Query. Note that at present the planner extensively modifies
125 * the passed-in Query data structure; someday that should stop.
128 typedef struct PlannerInfo
132 Query *parse; /* the Query being planned */
134 PlannerGlobal *glob; /* global info for current planner run */
136 Index query_level; /* 1 at the outermost Query */
138 struct PlannerInfo *parent_root; /* NULL at outermost Query */
141 * plan_params contains the expressions that this query level needs to
142 * make available to a lower query level that is currently being planned.
143 * outer_params contains the paramIds of PARAM_EXEC Params that outer
144 * query levels will make available to this query level.
146 List *plan_params; /* list of PlannerParamItems, see below */
147 Bitmapset *outer_params;
150 * simple_rel_array holds pointers to "base rels" and "other rels" (see
151 * comments for RelOptInfo for more info). It is indexed by rangetable
152 * index (so entry 0 is always wasted). Entries can be NULL when an RTE
153 * does not correspond to a base relation, such as a join RTE or an
154 * unreferenced view RTE; or if the RelOptInfo hasn't been made yet.
156 struct RelOptInfo **simple_rel_array; /* All 1-rel RelOptInfos */
157 int simple_rel_array_size; /* allocated size of array */
160 * simple_rte_array is the same length as simple_rel_array and holds
161 * pointers to the associated rangetable entries. This lets us avoid
162 * rt_fetch(), which can be a bit slow once large inheritance sets have
165 RangeTblEntry **simple_rte_array; /* rangetable as an array */
168 * all_baserels is a Relids set of all base relids (but not "other"
169 * relids) in the query; that is, the Relids identifier of the final join
170 * we need to form. This is computed in make_one_rel, just before we
171 * start making Paths.
176 * nullable_baserels is a Relids set of base relids that are nullable by
177 * some outer join in the jointree; these are rels that are potentially
178 * nullable below the WHERE clause, SELECT targetlist, etc. This is
179 * computed in deconstruct_jointree.
181 Relids nullable_baserels;
184 * join_rel_list is a list of all join-relation RelOptInfos we have
185 * considered in this planning run. For small problems we just scan the
186 * list to do lookups, but when there are many join relations we build a
187 * hash table for faster lookups. The hash table is present and valid
188 * when join_rel_hash is not NULL. Note that we still maintain the list
189 * even when using the hash table for lookups; this simplifies life for
192 List *join_rel_list; /* list of join-relation RelOptInfos */
193 struct HTAB *join_rel_hash; /* optional hashtable for join relations */
196 * When doing a dynamic-programming-style join search, join_rel_level[k]
197 * is a list of all join-relation RelOptInfos of level k, and
198 * join_cur_level is the current level. New join-relation RelOptInfos are
199 * automatically added to the join_rel_level[join_cur_level] list.
200 * join_rel_level is NULL if not in use.
202 List **join_rel_level; /* lists of join-relation RelOptInfos */
203 int join_cur_level; /* index of list being extended */
205 List *init_plans; /* init SubPlans for query */
207 List *cte_plan_ids; /* per-CTE-item list of subplan IDs */
209 List *multiexpr_params; /* List of Lists of Params for
210 * MULTIEXPR subquery outputs */
212 List *eq_classes; /* list of active EquivalenceClasses */
214 List *canon_pathkeys; /* list of "canonical" PathKeys */
216 List *left_join_clauses; /* list of RestrictInfos for
217 * mergejoinable outer join clauses
218 * w/nonnullable var on left */
220 List *right_join_clauses; /* list of RestrictInfos for
221 * mergejoinable outer join clauses
222 * w/nonnullable var on right */
224 List *full_join_clauses; /* list of RestrictInfos for
225 * mergejoinable full join clauses */
227 List *join_info_list; /* list of SpecialJoinInfos */
229 List *lateral_info_list; /* list of LateralJoinInfos */
231 List *append_rel_list; /* list of AppendRelInfos */
233 List *rowMarks; /* list of PlanRowMarks */
235 List *placeholder_list; /* list of PlaceHolderInfos */
237 List *query_pathkeys; /* desired pathkeys for query_planner(), and
238 * actual pathkeys after planning */
240 List *group_pathkeys; /* groupClause pathkeys, if any */
241 List *window_pathkeys; /* pathkeys of bottom window, if any */
242 List *distinct_pathkeys; /* distinctClause pathkeys, if any */
243 List *sort_pathkeys; /* sortClause pathkeys, if any */
245 List *minmax_aggs; /* List of MinMaxAggInfos */
247 List *initial_rels; /* RelOptInfos we are now trying to join */
249 MemoryContext planner_cxt; /* context holding PlannerInfo */
251 double total_table_pages; /* # of pages in all tables of query */
253 double tuple_fraction; /* tuple_fraction passed to query_planner */
254 double limit_tuples; /* limit_tuples passed to query_planner */
256 bool hasInheritedTarget; /* true if parse->resultRelation is an
257 * inheritance child rel */
258 bool hasJoinRTEs; /* true if any RTEs are RTE_JOIN kind */
259 bool hasLateralRTEs; /* true if any RTEs are marked LATERAL */
260 bool hasDeletedRTEs; /* true if any RTE was deleted from jointree */
261 bool hasHavingQual; /* true if havingQual was non-null */
262 bool hasPseudoConstantQuals; /* true if any RestrictInfo has
263 * pseudoconstant = true */
264 bool hasRecursion; /* true if planning a recursive WITH item */
266 /* These fields are used only when hasRecursion is true: */
267 int wt_param_id; /* PARAM_EXEC ID for the work table */
268 struct Plan *non_recursive_plan; /* plan for non-recursive term */
270 /* These fields are workspace for createplan.c */
271 Relids curOuterRels; /* outer rels above current node */
272 List *curOuterParams; /* not-yet-assigned NestLoopParams */
274 /* optional private data for join_search_hook, e.g., GEQO */
275 void *join_search_private;
277 /* for GroupingFunc fixup in setrefs */
278 AttrNumber *grouping_map;
283 * In places where it's known that simple_rte_array[] must have been prepared
284 * already, we just index into it to fetch RTEs. In code that might be
285 * executed before or after entering query_planner(), use this macro.
287 #define planner_rt_fetch(rti, root) \
288 ((root)->simple_rte_array ? (root)->simple_rte_array[rti] : \
289 rt_fetch(rti, (root)->parse->rtable))
294 * Per-relation information for planning/optimization
296 * For planning purposes, a "base rel" is either a plain relation (a table)
297 * or the output of a sub-SELECT or function that appears in the range table.
298 * In either case it is uniquely identified by an RT index. A "joinrel"
299 * is the joining of two or more base rels. A joinrel is identified by
300 * the set of RT indexes for its component baserels. We create RelOptInfo
301 * nodes for each baserel and joinrel, and store them in the PlannerInfo's
302 * simple_rel_array and join_rel_list respectively.
304 * Note that there is only one joinrel for any given set of component
305 * baserels, no matter what order we assemble them in; so an unordered
306 * set is the right datatype to identify it with.
308 * We also have "other rels", which are like base rels in that they refer to
309 * single RT indexes; but they are not part of the join tree, and are given
310 * a different RelOptKind to identify them. Lastly, there is a RelOptKind
311 * for "dead" relations, which are base rels that we have proven we don't
312 * need to join after all.
314 * Currently the only kind of otherrels are those made for member relations
315 * of an "append relation", that is an inheritance set or UNION ALL subquery.
316 * An append relation has a parent RTE that is a base rel, which represents
317 * the entire append relation. The member RTEs are otherrels. The parent
318 * is present in the query join tree but the members are not. The member
319 * RTEs and otherrels are used to plan the scans of the individual tables or
320 * subqueries of the append set; then the parent baserel is given Append
321 * and/or MergeAppend paths comprising the best paths for the individual
322 * member rels. (See comments for AppendRelInfo for more information.)
324 * At one time we also made otherrels to represent join RTEs, for use in
325 * handling join alias Vars. Currently this is not needed because all join
326 * alias Vars are expanded to non-aliased form during preprocess_expression.
328 * Parts of this data structure are specific to various scan and join
329 * mechanisms. It didn't seem worth creating new node types for them.
331 * relids - Set of base-relation identifiers; it is a base relation
332 * if there is just one, a join relation if more than one
333 * rows - estimated number of tuples in the relation after restriction
334 * clauses have been applied (ie, output rows of a plan for it)
335 * width - avg. number of bytes per tuple in the relation after the
336 * appropriate projections have been done (ie, output width)
337 * consider_startup - true if there is any value in keeping plain paths for
338 * this rel on the basis of having cheap startup cost
339 * consider_param_startup - the same for parameterized paths
340 * reltargetlist - List of Var and PlaceHolderVar nodes for the values
341 * we need to output from this relation.
342 * List is in no particular order, but all rels of an
343 * appendrel set must use corresponding orders.
344 * NOTE: in an appendrel child relation, may contain
345 * arbitrary expressions pulled up from a subquery!
346 * pathlist - List of Path nodes, one for each potentially useful
347 * method of generating the relation
348 * ppilist - ParamPathInfo nodes for parameterized Paths, if any
349 * cheapest_startup_path - the pathlist member with lowest startup cost
350 * (regardless of ordering) among the unparameterized paths;
351 * or NULL if there is no unparameterized path
352 * cheapest_total_path - the pathlist member with lowest total cost
353 * (regardless of ordering) among the unparameterized paths;
354 * or if there is no unparameterized path, the path with lowest
355 * total cost among the paths with minimum parameterization
356 * cheapest_unique_path - for caching cheapest path to produce unique
357 * (no duplicates) output from relation; NULL if not yet requested
358 * cheapest_parameterized_paths - best paths for their parameterizations;
359 * always includes cheapest_total_path, even if that's unparameterized
361 * If the relation is a base relation it will have these fields set:
363 * relid - RTE index (this is redundant with the relids field, but
364 * is provided for convenience of access)
365 * rtekind - distinguishes plain relation, subquery, or function RTE
366 * min_attr, max_attr - range of valid AttrNumbers for rel
367 * attr_needed - array of bitmapsets indicating the highest joinrel
368 * in which each attribute is needed; if bit 0 is set then
369 * the attribute is needed as part of final targetlist
370 * attr_widths - cache space for per-attribute width estimates;
371 * zero means not computed yet
372 * lateral_vars - lateral cross-references of rel, if any (list of
373 * Vars and PlaceHolderVars)
374 * lateral_relids - required outer rels for LATERAL, as a Relids set
375 * (for child rels this can be more than lateral_vars)
376 * lateral_referencers - relids of rels that reference this one laterally
377 * indexlist - list of IndexOptInfo nodes for relation's indexes
378 * (always NIL if it's not a table)
379 * pages - number of disk pages in relation (zero if not a table)
380 * tuples - number of tuples in relation (not considering restrictions)
381 * allvisfrac - fraction of disk pages that are marked all-visible
382 * subplan - plan for subquery (NULL if it's not a subquery)
383 * subroot - PlannerInfo for subquery (NULL if it's not a subquery)
384 * subplan_params - list of PlannerParamItems to be passed to subquery
386 * Note: for a subquery, tuples, subplan, subroot are not set immediately
387 * upon creation of the RelOptInfo object; they are filled in when
388 * set_subquery_pathlist processes the object.
390 * For otherrels that are appendrel members, these fields are filled
391 * in just as for a baserel.
393 * If the relation is either a foreign table or a join of foreign tables that
394 * all belong to the same foreign server, these fields will be set:
396 * serverid - OID of foreign server, if foreign table (else InvalidOid)
397 * fdwroutine - function hooks for FDW, if foreign table (else NULL)
398 * fdw_private - private state for FDW, if foreign table (else NULL)
400 * The presence of the remaining fields depends on the restrictions
401 * and joins that the relation participates in:
403 * baserestrictinfo - List of RestrictInfo nodes, containing info about
404 * each non-join qualification clause in which this relation
405 * participates (only used for base rels)
406 * baserestrictcost - Estimated cost of evaluating the baserestrictinfo
407 * clauses at a single tuple (only used for base rels)
408 * joininfo - List of RestrictInfo nodes, containing info about each
409 * join clause in which this relation participates (but
410 * note this excludes clauses that might be derivable from
411 * EquivalenceClasses)
412 * has_eclass_joins - flag that EquivalenceClass joins are possible
414 * Note: Keeping a restrictinfo list in the RelOptInfo is useful only for
415 * base rels, because for a join rel the set of clauses that are treated as
416 * restrict clauses varies depending on which sub-relations we choose to join.
417 * (For example, in a 3-base-rel join, a clause relating rels 1 and 2 must be
418 * treated as a restrictclause if we join {1} and {2 3} to make {1 2 3}; but
419 * if we join {1 2} and {3} then that clause will be a restrictclause in {1 2}
420 * and should not be processed again at the level of {1 2 3}.) Therefore,
421 * the restrictinfo list in the join case appears in individual JoinPaths
422 * (field joinrestrictinfo), not in the parent relation. But it's OK for
423 * the RelOptInfo to store the joininfo list, because that is the same
424 * for a given rel no matter how we form it.
426 * We store baserestrictcost in the RelOptInfo (for base relations) because
427 * we know we will need it at least once (to price the sequential scan)
428 * and may need it multiple times to price index scans.
431 typedef enum RelOptKind
435 RELOPT_OTHER_MEMBER_REL,
439 typedef struct RelOptInfo
443 RelOptKind reloptkind;
445 /* all relations included in this RelOptInfo */
446 Relids relids; /* set of base relids (rangetable indexes) */
448 /* size estimates generated by planner */
449 double rows; /* estimated number of result tuples */
450 int width; /* estimated avg width of result tuples */
452 /* per-relation planner control flags */
453 bool consider_startup; /* keep cheap-startup-cost paths? */
454 bool consider_param_startup; /* ditto, for parameterized paths? */
456 /* materialization information */
457 List *reltargetlist; /* Vars to be output by scan of relation */
458 List *pathlist; /* Path structures */
459 List *ppilist; /* ParamPathInfos used in pathlist */
460 struct Path *cheapest_startup_path;
461 struct Path *cheapest_total_path;
462 struct Path *cheapest_unique_path;
463 List *cheapest_parameterized_paths;
465 /* information about a base rel (not set for join rels!) */
467 Oid reltablespace; /* containing tablespace */
468 RTEKind rtekind; /* RELATION, SUBQUERY, or FUNCTION */
469 AttrNumber min_attr; /* smallest attrno of rel (often <0) */
470 AttrNumber max_attr; /* largest attrno of rel */
471 Relids *attr_needed; /* array indexed [min_attr .. max_attr] */
472 int32 *attr_widths; /* array indexed [min_attr .. max_attr] */
473 List *lateral_vars; /* LATERAL Vars and PHVs referenced by rel */
474 Relids lateral_relids; /* minimum parameterization of rel */
475 Relids lateral_referencers; /* rels that reference me laterally */
476 List *indexlist; /* list of IndexOptInfo */
477 BlockNumber pages; /* size estimates derived from pg_class */
480 /* use "struct Plan" to avoid including plannodes.h here */
481 struct Plan *subplan; /* if subquery */
482 PlannerInfo *subroot; /* if subquery */
483 List *subplan_params; /* if subquery */
485 /* Information about foreign tables and foreign joins */
486 Oid serverid; /* identifies server for the table or join */
487 /* use "struct FdwRoutine" to avoid including fdwapi.h here */
488 struct FdwRoutine *fdwroutine;
491 /* used by various scans and joins: */
492 List *baserestrictinfo; /* RestrictInfo structures (if base
494 QualCost baserestrictcost; /* cost of evaluating the above */
495 List *joininfo; /* RestrictInfo structures for join clauses
496 * involving this rel */
497 bool has_eclass_joins; /* T means joininfo is incomplete */
502 * Per-index information for planning/optimization
504 * indexkeys[], indexcollations[], opfamily[], and opcintype[]
505 * each have ncolumns entries.
507 * sortopfamily[], reverse_sort[], and nulls_first[] likewise have
508 * ncolumns entries, if the index is ordered; but if it is unordered,
509 * those pointers are NULL.
511 * Zeroes in the indexkeys[] array indicate index columns that are
512 * expressions; there is one element in indexprs for each such column.
514 * For an ordered index, reverse_sort[] and nulls_first[] describe the
515 * sort ordering of a forward indexscan; we can also consider a backward
516 * indexscan, which will generate the reverse ordering.
518 * The indexprs and indpred expressions have been run through
519 * prepqual.c and eval_const_expressions() for ease of matching to
520 * WHERE clauses. indpred is in implicit-AND form.
522 * indextlist is a TargetEntry list representing the index columns.
523 * It provides an equivalent base-relation Var for each simple column,
524 * and links to the matching indexprs element for each expression column.
526 typedef struct IndexOptInfo
530 Oid indexoid; /* OID of the index relation */
531 Oid reltablespace; /* tablespace of index (not table) */
532 RelOptInfo *rel; /* back-link to index's table */
534 /* index-size statistics (from pg_class and elsewhere) */
535 BlockNumber pages; /* number of disk pages in index */
536 double tuples; /* number of index tuples in index */
537 int tree_height; /* index tree height, or -1 if unknown */
539 /* index descriptor information */
540 int ncolumns; /* number of columns in index */
541 int *indexkeys; /* column numbers of index's keys, or 0 */
542 Oid *indexcollations; /* OIDs of collations of index columns */
543 Oid *opfamily; /* OIDs of operator families for columns */
544 Oid *opcintype; /* OIDs of opclass declared input data types */
545 Oid *sortopfamily; /* OIDs of btree opfamilies, if orderable */
546 bool *reverse_sort; /* is sort order descending? */
547 bool *nulls_first; /* do NULLs come first in the sort order? */
548 bool *canreturn; /* which index cols can be returned in an
549 * index-only scan? */
550 Oid relam; /* OID of the access method (in pg_am) */
552 RegProcedure amcostestimate; /* OID of the access method's cost fcn */
554 List *indexprs; /* expressions for non-simple index columns */
555 List *indpred; /* predicate if a partial index, else NIL */
557 List *indextlist; /* targetlist representing index columns */
559 bool predOK; /* true if predicate matches query */
560 bool unique; /* true if a unique index */
561 bool immediate; /* is uniqueness enforced immediately? */
562 bool hypothetical; /* true if index doesn't really exist */
563 bool amcanorderbyop; /* does AM support order by operator result? */
564 bool amoptionalkey; /* can query omit key for the first column? */
565 bool amsearcharray; /* can AM handle ScalarArrayOpExpr quals? */
566 bool amsearchnulls; /* can AM search for NULL/NOT NULL entries? */
567 bool amhasgettuple; /* does AM have amgettuple interface? */
568 bool amhasgetbitmap; /* does AM have amgetbitmap interface? */
575 * Whenever we can determine that a mergejoinable equality clause A = B is
576 * not delayed by any outer join, we create an EquivalenceClass containing
577 * the expressions A and B to record this knowledge. If we later find another
578 * equivalence B = C, we add C to the existing EquivalenceClass; this may
579 * require merging two existing EquivalenceClasses. At the end of the qual
580 * distribution process, we have sets of values that are known all transitively
581 * equal to each other, where "equal" is according to the rules of the btree
582 * operator family(s) shown in ec_opfamilies, as well as the collation shown
583 * by ec_collation. (We restrict an EC to contain only equalities whose
584 * operators belong to the same set of opfamilies. This could probably be
585 * relaxed, but for now it's not worth the trouble, since nearly all equality
586 * operators belong to only one btree opclass anyway. Similarly, we suppose
587 * that all or none of the input datatypes are collatable, so that a single
588 * collation value is sufficient.)
590 * We also use EquivalenceClasses as the base structure for PathKeys, letting
591 * us represent knowledge about different sort orderings being equivalent.
592 * Since every PathKey must reference an EquivalenceClass, we will end up
593 * with single-member EquivalenceClasses whenever a sort key expression has
594 * not been equivalenced to anything else. It is also possible that such an
595 * EquivalenceClass will contain a volatile expression ("ORDER BY random()"),
596 * which is a case that can't arise otherwise since clauses containing
597 * volatile functions are never considered mergejoinable. We mark such
598 * EquivalenceClasses specially to prevent them from being merged with
599 * ordinary EquivalenceClasses. Also, for volatile expressions we have
600 * to be careful to match the EquivalenceClass to the correct targetlist
601 * entry: consider SELECT random() AS a, random() AS b ... ORDER BY b,a.
602 * So we record the SortGroupRef of the originating sort clause.
604 * We allow equality clauses appearing below the nullable side of an outer join
605 * to form EquivalenceClasses, but these have a slightly different meaning:
606 * the included values might be all NULL rather than all the same non-null
607 * values. See src/backend/optimizer/README for more on that point.
609 * NB: if ec_merged isn't NULL, this class has been merged into another, and
610 * should be ignored in favor of using the pointed-to class.
612 typedef struct EquivalenceClass
616 List *ec_opfamilies; /* btree operator family OIDs */
617 Oid ec_collation; /* collation, if datatypes are collatable */
618 List *ec_members; /* list of EquivalenceMembers */
619 List *ec_sources; /* list of generating RestrictInfos */
620 List *ec_derives; /* list of derived RestrictInfos */
621 Relids ec_relids; /* all relids appearing in ec_members, except
622 * for child members (see below) */
623 bool ec_has_const; /* any pseudoconstants in ec_members? */
624 bool ec_has_volatile; /* the (sole) member is a volatile expr */
625 bool ec_below_outer_join; /* equivalence applies below an OJ */
626 bool ec_broken; /* failed to generate needed clauses? */
627 Index ec_sortref; /* originating sortclause label, or 0 */
628 struct EquivalenceClass *ec_merged; /* set if merged into another EC */
632 * If an EC contains a const and isn't below-outer-join, any PathKey depending
633 * on it must be redundant, since there's only one possible value of the key.
635 #define EC_MUST_BE_REDUNDANT(eclass) \
636 ((eclass)->ec_has_const && !(eclass)->ec_below_outer_join)
639 * EquivalenceMember - one member expression of an EquivalenceClass
641 * em_is_child signifies that this element was built by transposing a member
642 * for an appendrel parent relation to represent the corresponding expression
643 * for an appendrel child. These members are used for determining the
644 * pathkeys of scans on the child relation and for explicitly sorting the
645 * child when necessary to build a MergeAppend path for the whole appendrel
646 * tree. An em_is_child member has no impact on the properties of the EC as a
647 * whole; in particular the EC's ec_relids field does NOT include the child
648 * relation. An em_is_child member should never be marked em_is_const nor
649 * cause ec_has_const or ec_has_volatile to be set, either. Thus, em_is_child
650 * members are not really full-fledged members of the EC, but just reflections
651 * or doppelgangers of real members. Most operations on EquivalenceClasses
652 * should ignore em_is_child members, and those that don't should test
653 * em_relids to make sure they only consider relevant members.
655 * em_datatype is usually the same as exprType(em_expr), but can be
656 * different when dealing with a binary-compatible opfamily; in particular
657 * anyarray_ops would never work without this. Use em_datatype when
658 * looking up a specific btree operator to work with this expression.
660 typedef struct EquivalenceMember
664 Expr *em_expr; /* the expression represented */
665 Relids em_relids; /* all relids appearing in em_expr */
666 Relids em_nullable_relids; /* nullable by lower outer joins */
667 bool em_is_const; /* expression is pseudoconstant? */
668 bool em_is_child; /* derived version for a child relation? */
669 Oid em_datatype; /* the "nominal type" used by the opfamily */
675 * The sort ordering of a path is represented by a list of PathKey nodes.
676 * An empty list implies no known ordering. Otherwise the first item
677 * represents the primary sort key, the second the first secondary sort key,
678 * etc. The value being sorted is represented by linking to an
679 * EquivalenceClass containing that value and including pk_opfamily among its
680 * ec_opfamilies. The EquivalenceClass tells which collation to use, too.
681 * This is a convenient method because it makes it trivial to detect
682 * equivalent and closely-related orderings. (See optimizer/README for more
685 * Note: pk_strategy is either BTLessStrategyNumber (for ASC) or
686 * BTGreaterStrategyNumber (for DESC). We assume that all ordering-capable
687 * index types will use btree-compatible strategy numbers.
689 typedef struct PathKey
693 EquivalenceClass *pk_eclass; /* the value that is ordered */
694 Oid pk_opfamily; /* btree opfamily defining the ordering */
695 int pk_strategy; /* sort direction (ASC or DESC) */
696 bool pk_nulls_first; /* do NULLs come before normal values? */
703 * All parameterized paths for a given relation with given required outer rels
704 * link to a single ParamPathInfo, which stores common information such as
705 * the estimated rowcount for this parameterization. We do this partly to
706 * avoid recalculations, but mostly to ensure that the estimated rowcount
707 * is in fact the same for every such path.
709 * Note: ppi_clauses is only used in ParamPathInfos for base relation paths;
710 * in join cases it's NIL because the set of relevant clauses varies depending
711 * on how the join is formed. The relevant clauses will appear in each
712 * parameterized join path's joinrestrictinfo list, instead.
714 typedef struct ParamPathInfo
718 Relids ppi_req_outer; /* rels supplying parameters used by path */
719 double ppi_rows; /* estimated number of result tuples */
720 List *ppi_clauses; /* join clauses available from outer rels */
725 * Type "Path" is used as-is for sequential-scan paths, as well as some other
726 * simple plan types that we don't need any extra information in the path for.
727 * For other path types it is the first component of a larger struct.
729 * "pathtype" is the NodeTag of the Plan node we could build from this Path.
730 * It is partially redundant with the Path's NodeTag, but allows us to use
731 * the same Path type for multiple Plan types when there is no need to
732 * distinguish the Plan type during path processing.
734 * "param_info", if not NULL, links to a ParamPathInfo that identifies outer
735 * relation(s) that provide parameter values to each scan of this path.
736 * That means this path can only be joined to those rels by means of nestloop
737 * joins with this path on the inside. Also note that a parameterized path
738 * is responsible for testing all "movable" joinclauses involving this rel
739 * and the specified outer rel(s).
741 * "rows" is the same as parent->rows in simple paths, but in parameterized
742 * paths and UniquePaths it can be less than parent->rows, reflecting the
743 * fact that we've filtered by extra join conditions or removed duplicates.
745 * "pathkeys" is a List of PathKey nodes (see above), describing the sort
746 * ordering of the path's output rows.
752 NodeTag pathtype; /* tag identifying scan/join method */
754 RelOptInfo *parent; /* the relation this path can build */
755 ParamPathInfo *param_info; /* parameterization info, or NULL if none */
757 /* estimated size/costs for path (see costsize.c for more info) */
758 double rows; /* estimated number of result tuples */
759 Cost startup_cost; /* cost expended before fetching any tuples */
760 Cost total_cost; /* total cost (assuming all tuples fetched) */
762 List *pathkeys; /* sort ordering of path's output */
763 /* pathkeys is a List of PathKey nodes; see above */
766 /* Macro for extracting a path's parameterization relids; beware double eval */
767 #define PATH_REQ_OUTER(path) \
768 ((path)->param_info ? (path)->param_info->ppi_req_outer : (Relids) NULL)
771 * IndexPath represents an index scan over a single index.
773 * This struct is used for both regular indexscans and index-only scans;
774 * path.pathtype is T_IndexScan or T_IndexOnlyScan to show which is meant.
776 * 'indexinfo' is the index to be scanned.
778 * 'indexclauses' is a list of index qualification clauses, with implicit
779 * AND semantics across the list. Each clause is a RestrictInfo node from
780 * the query's WHERE or JOIN conditions. An empty list implies a full
783 * 'indexquals' has the same structure as 'indexclauses', but it contains
784 * the actual index qual conditions that can be used with the index.
785 * In simple cases this is identical to 'indexclauses', but when special
786 * indexable operators appear in 'indexclauses', they are replaced by the
787 * derived indexscannable conditions in 'indexquals'.
789 * 'indexqualcols' is an integer list of index column numbers (zero-based)
790 * of the same length as 'indexquals', showing which index column each qual
791 * is meant to be used with. 'indexquals' is required to be ordered by
792 * index column, so 'indexqualcols' must form a nondecreasing sequence.
793 * (The order of multiple quals for the same index column is unspecified.)
795 * 'indexorderbys', if not NIL, is a list of ORDER BY expressions that have
796 * been found to be usable as ordering operators for an amcanorderbyop index.
797 * The list must match the path's pathkeys, ie, one expression per pathkey
798 * in the same order. These are not RestrictInfos, just bare expressions,
799 * since they generally won't yield booleans. Also, unlike the case for
800 * quals, it's guaranteed that each expression has the index key on the left
801 * side of the operator.
803 * 'indexorderbycols' is an integer list of index column numbers (zero-based)
804 * of the same length as 'indexorderbys', showing which index column each
805 * ORDER BY expression is meant to be used with. (There is no restriction
806 * on which index column each ORDER BY can be used with.)
808 * 'indexscandir' is one of:
809 * ForwardScanDirection: forward scan of an ordered index
810 * BackwardScanDirection: backward scan of an ordered index
811 * NoMovementScanDirection: scan of an unordered index, or don't care
812 * (The executor doesn't care whether it gets ForwardScanDirection or
813 * NoMovementScanDirection for an indexscan, but the planner wants to
814 * distinguish ordered from unordered indexes for building pathkeys.)
816 * 'indextotalcost' and 'indexselectivity' are saved in the IndexPath so that
817 * we need not recompute them when considering using the same index in a
818 * bitmap index/heap scan (see BitmapHeapPath). The costs of the IndexPath
819 * itself represent the costs of an IndexScan or IndexOnlyScan plan type.
822 typedef struct IndexPath
825 IndexOptInfo *indexinfo;
830 List *indexorderbycols;
831 ScanDirection indexscandir;
833 Selectivity indexselectivity;
837 * BitmapHeapPath represents one or more indexscans that generate TID bitmaps
838 * instead of directly accessing the heap, followed by AND/OR combinations
839 * to produce a single bitmap, followed by a heap scan that uses the bitmap.
840 * Note that the output is always considered unordered, since it will come
841 * out in physical heap order no matter what the underlying indexes did.
843 * The individual indexscans are represented by IndexPath nodes, and any
844 * logic on top of them is represented by a tree of BitmapAndPath and
845 * BitmapOrPath nodes. Notice that we can use the same IndexPath node both
846 * to represent a regular (or index-only) index scan plan, and as the child
847 * of a BitmapHeapPath that represents scanning the same index using a
848 * BitmapIndexScan. The startup_cost and total_cost figures of an IndexPath
849 * always represent the costs to use it as a regular (or index-only)
850 * IndexScan. The costs of a BitmapIndexScan can be computed using the
851 * IndexPath's indextotalcost and indexselectivity.
853 typedef struct BitmapHeapPath
856 Path *bitmapqual; /* IndexPath, BitmapAndPath, BitmapOrPath */
860 * BitmapAndPath represents a BitmapAnd plan node; it can only appear as
861 * part of the substructure of a BitmapHeapPath. The Path structure is
862 * a bit more heavyweight than we really need for this, but for simplicity
863 * we make it a derivative of Path anyway.
865 typedef struct BitmapAndPath
868 List *bitmapquals; /* IndexPaths and BitmapOrPaths */
869 Selectivity bitmapselectivity;
873 * BitmapOrPath represents a BitmapOr plan node; it can only appear as
874 * part of the substructure of a BitmapHeapPath. The Path structure is
875 * a bit more heavyweight than we really need for this, but for simplicity
876 * we make it a derivative of Path anyway.
878 typedef struct BitmapOrPath
881 List *bitmapquals; /* IndexPaths and BitmapAndPaths */
882 Selectivity bitmapselectivity;
886 * TidPath represents a scan by TID
888 * tidquals is an implicitly OR'ed list of qual expressions of the form
889 * "CTID = pseudoconstant" or "CTID = ANY(pseudoconstant_array)".
890 * Note they are bare expressions, not RestrictInfos.
892 typedef struct TidPath
895 List *tidquals; /* qual(s) involving CTID = something */
899 * ForeignPath represents a potential scan of a foreign table
901 * fdw_private stores FDW private data about the scan. While fdw_private is
902 * not actually touched by the core code during normal operations, it's
903 * generally a good idea to use a representation that can be dumped by
904 * nodeToString(), so that you can examine the structure during debugging
905 * with tools like pprint().
907 typedef struct ForeignPath
914 * CustomPath represents a table scan done by some out-of-core extension.
916 * We provide a set of hooks here - which the provider must take care to set
917 * up correctly - to allow extensions to supply their own methods of scanning
918 * a relation. For example, a provider might provide GPU acceleration, a
919 * cache-based scan, or some other kind of logic we haven't dreamed up yet.
921 * CustomPaths can be injected into the planning process for a relation by
922 * set_rel_pathlist_hook functions.
924 * Core code must avoid assuming that the CustomPath is only as large as
925 * the structure declared here; providers are allowed to make it the first
926 * element in a larger structure. (Since the planner never copies Paths,
927 * this doesn't add any complication.) However, for consistency with the
928 * FDW case, we provide a "custom_private" field in CustomPath; providers
929 * may prefer to use that rather than define another struct type.
933 #define CUSTOMPATH_SUPPORT_BACKWARD_SCAN 0x0001
934 #define CUSTOMPATH_SUPPORT_MARK_RESTORE 0x0002
936 typedef struct CustomPathMethods
938 const char *CustomName;
940 /* Convert Path to a Plan */
941 struct Plan *(*PlanCustomPath) (PlannerInfo *root,
943 struct CustomPath *best_path,
947 /* Optional: print additional fields besides "private" */
948 void (*TextOutCustomPath) (StringInfo str,
949 const struct CustomPath *node);
952 typedef struct CustomPath
955 uint32 flags; /* mask of CUSTOMPATH_* flags, see above */
956 List *custom_paths; /* list of child Path nodes, if any */
957 List *custom_private;
958 const CustomPathMethods *methods;
962 * AppendPath represents an Append plan, ie, successive execution of
963 * several member plans.
965 * Note: it is possible for "subpaths" to contain only one, or even no,
966 * elements. These cases are optimized during create_append_plan.
967 * In particular, an AppendPath with no subpaths is a "dummy" path that
968 * is created to represent the case that a relation is provably empty.
970 typedef struct AppendPath
973 List *subpaths; /* list of component Paths */
976 #define IS_DUMMY_PATH(p) \
977 (IsA((p), AppendPath) && ((AppendPath *) (p))->subpaths == NIL)
979 /* A relation that's been proven empty will have one path that is dummy */
980 #define IS_DUMMY_REL(r) \
981 ((r)->cheapest_total_path != NULL && \
982 IS_DUMMY_PATH((r)->cheapest_total_path))
985 * MergeAppendPath represents a MergeAppend plan, ie, the merging of sorted
986 * results from several member plans to produce similarly-sorted output.
988 typedef struct MergeAppendPath
991 List *subpaths; /* list of component Paths */
992 double limit_tuples; /* hard limit on output tuples, or -1 */
996 * ResultPath represents use of a Result plan node to compute a variable-free
997 * targetlist with no underlying tables (a "SELECT expressions" query).
998 * The query could have a WHERE clause, too, represented by "quals".
1000 * Note that quals is a list of bare clauses, not RestrictInfos.
1002 typedef struct ResultPath
1009 * MaterialPath represents use of a Material plan node, i.e., caching of
1010 * the output of its subpath. This is used when the subpath is expensive
1011 * and needs to be scanned repeatedly, or when we need mark/restore ability
1012 * and the subpath doesn't have it.
1014 typedef struct MaterialPath
1021 * UniquePath represents elimination of distinct rows from the output of
1024 * This is unlike the other Path nodes in that it can actually generate
1025 * different plans: either hash-based or sort-based implementation, or a
1026 * no-op if the input path can be proven distinct already. The decision
1027 * is sufficiently localized that it's not worth having separate Path node
1028 * types. (Note: in the no-op case, we could eliminate the UniquePath node
1029 * entirely and just return the subpath; but it's convenient to have a
1030 * UniquePath in the path tree to signal upper-level routines that the input
1031 * is known distinct.)
1035 UNIQUE_PATH_NOOP, /* input is known unique already */
1036 UNIQUE_PATH_HASH, /* use hashing */
1037 UNIQUE_PATH_SORT /* use sorting */
1040 typedef struct UniquePath
1044 UniquePathMethod umethod;
1045 List *in_operators; /* equality operators of the IN clause */
1046 List *uniq_exprs; /* expressions to be made unique */
1050 * GatherPath runs several copies of a plan in parallel and collects the
1051 * results. The parallel leader may also execute the plan, unless the
1052 * single_copy flag is set.
1054 typedef struct GatherPath
1057 Path *subpath; /* path for each worker */
1058 int num_workers; /* number of workers sought to help */
1059 bool single_copy; /* path must not be executed >1x */
1063 * All join-type paths share these fields.
1066 typedef struct JoinPath
1072 Path *outerjoinpath; /* path for the outer side of the join */
1073 Path *innerjoinpath; /* path for the inner side of the join */
1075 List *joinrestrictinfo; /* RestrictInfos to apply to join */
1078 * See the notes for RelOptInfo and ParamPathInfo to understand why
1079 * joinrestrictinfo is needed in JoinPath, and can't be merged into the
1080 * parent RelOptInfo.
1085 * A nested-loop path needs no special fields.
1088 typedef JoinPath NestPath;
1091 * A mergejoin path has these fields.
1093 * Unlike other path types, a MergePath node doesn't represent just a single
1094 * run-time plan node: it can represent up to four. Aside from the MergeJoin
1095 * node itself, there can be a Sort node for the outer input, a Sort node
1096 * for the inner input, and/or a Material node for the inner input. We could
1097 * represent these nodes by separate path nodes, but considering how many
1098 * different merge paths are investigated during a complex join problem,
1099 * it seems better to avoid unnecessary palloc overhead.
1101 * path_mergeclauses lists the clauses (in the form of RestrictInfos)
1102 * that will be used in the merge.
1104 * Note that the mergeclauses are a subset of the parent relation's
1105 * restriction-clause list. Any join clauses that are not mergejoinable
1106 * appear only in the parent's restrict list, and must be checked by a
1107 * qpqual at execution time.
1109 * outersortkeys (resp. innersortkeys) is NIL if the outer path
1110 * (resp. inner path) is already ordered appropriately for the
1111 * mergejoin. If it is not NIL then it is a PathKeys list describing
1112 * the ordering that must be created by an explicit Sort node.
1114 * materialize_inner is TRUE if a Material node should be placed atop the
1115 * inner input. This may appear with or without an inner Sort step.
1118 typedef struct MergePath
1121 List *path_mergeclauses; /* join clauses to be used for merge */
1122 List *outersortkeys; /* keys for explicit sort, if any */
1123 List *innersortkeys; /* keys for explicit sort, if any */
1124 bool materialize_inner; /* add Materialize to inner? */
1128 * A hashjoin path has these fields.
1130 * The remarks above for mergeclauses apply for hashclauses as well.
1132 * Hashjoin does not care what order its inputs appear in, so we have
1133 * no need for sortkeys.
1136 typedef struct HashPath
1139 List *path_hashclauses; /* join clauses used for hashing */
1140 int num_batches; /* number of batches expected */
1144 * Restriction clause info.
1146 * We create one of these for each AND sub-clause of a restriction condition
1147 * (WHERE or JOIN/ON clause). Since the restriction clauses are logically
1148 * ANDed, we can use any one of them or any subset of them to filter out
1149 * tuples, without having to evaluate the rest. The RestrictInfo node itself
1150 * stores data used by the optimizer while choosing the best query plan.
1152 * If a restriction clause references a single base relation, it will appear
1153 * in the baserestrictinfo list of the RelOptInfo for that base rel.
1155 * If a restriction clause references more than one base rel, it will
1156 * appear in the joininfo list of every RelOptInfo that describes a strict
1157 * subset of the base rels mentioned in the clause. The joininfo lists are
1158 * used to drive join tree building by selecting plausible join candidates.
1159 * The clause cannot actually be applied until we have built a join rel
1160 * containing all the base rels it references, however.
1162 * When we construct a join rel that includes all the base rels referenced
1163 * in a multi-relation restriction clause, we place that clause into the
1164 * joinrestrictinfo lists of paths for the join rel, if neither left nor
1165 * right sub-path includes all base rels referenced in the clause. The clause
1166 * will be applied at that join level, and will not propagate any further up
1167 * the join tree. (Note: the "predicate migration" code was once intended to
1168 * push restriction clauses up and down the plan tree based on evaluation
1169 * costs, but it's dead code and is unlikely to be resurrected in the
1170 * foreseeable future.)
1172 * Note that in the presence of more than two rels, a multi-rel restriction
1173 * might reach different heights in the join tree depending on the join
1174 * sequence we use. So, these clauses cannot be associated directly with
1175 * the join RelOptInfo, but must be kept track of on a per-join-path basis.
1177 * RestrictInfos that represent equivalence conditions (i.e., mergejoinable
1178 * equalities that are not outerjoin-delayed) are handled a bit differently.
1179 * Initially we attach them to the EquivalenceClasses that are derived from
1180 * them. When we construct a scan or join path, we look through all the
1181 * EquivalenceClasses and generate derived RestrictInfos representing the
1182 * minimal set of conditions that need to be checked for this particular scan
1183 * or join to enforce that all members of each EquivalenceClass are in fact
1184 * equal in all rows emitted by the scan or join.
1186 * When dealing with outer joins we have to be very careful about pushing qual
1187 * clauses up and down the tree. An outer join's own JOIN/ON conditions must
1188 * be evaluated exactly at that join node, unless they are "degenerate"
1189 * conditions that reference only Vars from the nullable side of the join.
1190 * Quals appearing in WHERE or in a JOIN above the outer join cannot be pushed
1191 * down below the outer join, if they reference any nullable Vars.
1192 * RestrictInfo nodes contain a flag to indicate whether a qual has been
1193 * pushed down to a lower level than its original syntactic placement in the
1194 * join tree would suggest. If an outer join prevents us from pushing a qual
1195 * down to its "natural" semantic level (the level associated with just the
1196 * base rels used in the qual) then we mark the qual with a "required_relids"
1197 * value including more than just the base rels it actually uses. By
1198 * pretending that the qual references all the rels required to form the outer
1199 * join, we prevent it from being evaluated below the outer join's joinrel.
1200 * When we do form the outer join's joinrel, we still need to distinguish
1201 * those quals that are actually in that join's JOIN/ON condition from those
1202 * that appeared elsewhere in the tree and were pushed down to the join rel
1203 * because they used no other rels. That's what the is_pushed_down flag is
1204 * for; it tells us that a qual is not an OUTER JOIN qual for the set of base
1205 * rels listed in required_relids. A clause that originally came from WHERE
1206 * or an INNER JOIN condition will *always* have its is_pushed_down flag set.
1207 * It's possible for an OUTER JOIN clause to be marked is_pushed_down too,
1208 * if we decide that it can be pushed down into the nullable side of the join.
1209 * In that case it acts as a plain filter qual for wherever it gets evaluated.
1210 * (In short, is_pushed_down is only false for non-degenerate outer join
1211 * conditions. Possibly we should rename it to reflect that meaning?)
1213 * RestrictInfo nodes also contain an outerjoin_delayed flag, which is true
1214 * if the clause's applicability must be delayed due to any outer joins
1215 * appearing below it (ie, it has to be postponed to some join level higher
1216 * than the set of relations it actually references).
1218 * There is also an outer_relids field, which is NULL except for outer join
1219 * clauses; for those, it is the set of relids on the outer side of the
1220 * clause's outer join. (These are rels that the clause cannot be applied to
1221 * in parameterized scans, since pushing it into the join's outer side would
1222 * lead to wrong answers.)
1224 * There is also a nullable_relids field, which is the set of rels the clause
1225 * references that can be forced null by some outer join below the clause.
1227 * outerjoin_delayed = true is subtly different from nullable_relids != NULL:
1228 * a clause might reference some nullable rels and yet not be
1229 * outerjoin_delayed because it also references all the other rels of the
1230 * outer join(s). A clause that is not outerjoin_delayed can be enforced
1231 * anywhere it is computable.
1233 * In general, the referenced clause might be arbitrarily complex. The
1234 * kinds of clauses we can handle as indexscan quals, mergejoin clauses,
1235 * or hashjoin clauses are limited (e.g., no volatile functions). The code
1236 * for each kind of path is responsible for identifying the restrict clauses
1237 * it can use and ignoring the rest. Clauses not implemented by an indexscan,
1238 * mergejoin, or hashjoin will be placed in the plan qual or joinqual field
1239 * of the finished Plan node, where they will be enforced by general-purpose
1240 * qual-expression-evaluation code. (But we are still entitled to count
1241 * their selectivity when estimating the result tuple count, if we
1242 * can guess what it is...)
1244 * When the referenced clause is an OR clause, we generate a modified copy
1245 * in which additional RestrictInfo nodes are inserted below the top-level
1246 * OR/AND structure. This is a convenience for OR indexscan processing:
1247 * indexquals taken from either the top level or an OR subclause will have
1248 * associated RestrictInfo nodes.
1250 * The can_join flag is set true if the clause looks potentially useful as
1251 * a merge or hash join clause, that is if it is a binary opclause with
1252 * nonoverlapping sets of relids referenced in the left and right sides.
1253 * (Whether the operator is actually merge or hash joinable isn't checked,
1256 * The pseudoconstant flag is set true if the clause contains no Vars of
1257 * the current query level and no volatile functions. Such a clause can be
1258 * pulled out and used as a one-time qual in a gating Result node. We keep
1259 * pseudoconstant clauses in the same lists as other RestrictInfos so that
1260 * the regular clause-pushing machinery can assign them to the correct join
1261 * level, but they need to be treated specially for cost and selectivity
1262 * estimates. Note that a pseudoconstant clause can never be an indexqual
1263 * or merge or hash join clause, so it's of no interest to large parts of
1266 * When join clauses are generated from EquivalenceClasses, there may be
1267 * several equally valid ways to enforce join equivalence, of which we need
1268 * apply only one. We mark clauses of this kind by setting parent_ec to
1269 * point to the generating EquivalenceClass. Multiple clauses with the same
1270 * parent_ec in the same join are redundant.
1273 typedef struct RestrictInfo
1277 Expr *clause; /* the represented clause of WHERE or JOIN */
1279 bool is_pushed_down; /* TRUE if clause was pushed down in level */
1281 bool outerjoin_delayed; /* TRUE if delayed by lower outer join */
1283 bool can_join; /* see comment above */
1285 bool pseudoconstant; /* see comment above */
1287 /* The set of relids (varnos) actually referenced in the clause: */
1288 Relids clause_relids;
1290 /* The set of relids required to evaluate the clause: */
1291 Relids required_relids;
1293 /* If an outer-join clause, the outer-side relations, else NULL: */
1294 Relids outer_relids;
1296 /* The relids used in the clause that are nullable by lower outer joins: */
1297 Relids nullable_relids;
1299 /* These fields are set for any binary opclause: */
1300 Relids left_relids; /* relids in left side of clause */
1301 Relids right_relids; /* relids in right side of clause */
1303 /* This field is NULL unless clause is an OR clause: */
1304 Expr *orclause; /* modified clause with RestrictInfos */
1306 /* This field is NULL unless clause is potentially redundant: */
1307 EquivalenceClass *parent_ec; /* generating EquivalenceClass */
1309 /* cache space for cost and selectivity */
1310 QualCost eval_cost; /* eval cost of clause; -1 if not yet set */
1311 Selectivity norm_selec; /* selectivity for "normal" (JOIN_INNER)
1312 * semantics; -1 if not yet set; >1 means a
1313 * redundant clause */
1314 Selectivity outer_selec; /* selectivity for outer join semantics; -1 if
1317 /* valid if clause is mergejoinable, else NIL */
1318 List *mergeopfamilies; /* opfamilies containing clause operator */
1320 /* cache space for mergeclause processing; NULL if not yet set */
1321 EquivalenceClass *left_ec; /* EquivalenceClass containing lefthand */
1322 EquivalenceClass *right_ec; /* EquivalenceClass containing righthand */
1323 EquivalenceMember *left_em; /* EquivalenceMember for lefthand */
1324 EquivalenceMember *right_em; /* EquivalenceMember for righthand */
1325 List *scansel_cache; /* list of MergeScanSelCache structs */
1327 /* transient workspace for use while considering a specific join path */
1328 bool outer_is_left; /* T = outer var on left, F = on right */
1330 /* valid if clause is hashjoinable, else InvalidOid: */
1331 Oid hashjoinoperator; /* copy of clause operator */
1333 /* cache space for hashclause processing; -1 if not yet set */
1334 Selectivity left_bucketsize; /* avg bucketsize of left side */
1335 Selectivity right_bucketsize; /* avg bucketsize of right side */
1339 * Since mergejoinscansel() is a relatively expensive function, and would
1340 * otherwise be invoked many times while planning a large join tree,
1341 * we go out of our way to cache its results. Each mergejoinable
1342 * RestrictInfo carries a list of the specific sort orderings that have
1343 * been considered for use with it, and the resulting selectivities.
1345 typedef struct MergeScanSelCache
1347 /* Ordering details (cache lookup key) */
1348 Oid opfamily; /* btree opfamily defining the ordering */
1349 Oid collation; /* collation for the ordering */
1350 int strategy; /* sort direction (ASC or DESC) */
1351 bool nulls_first; /* do NULLs come before normal values? */
1353 Selectivity leftstartsel; /* first-join fraction for clause left side */
1354 Selectivity leftendsel; /* last-join fraction for clause left side */
1355 Selectivity rightstartsel; /* first-join fraction for clause right side */
1356 Selectivity rightendsel; /* last-join fraction for clause right side */
1357 } MergeScanSelCache;
1360 * Placeholder node for an expression to be evaluated below the top level
1361 * of a plan tree. This is used during planning to represent the contained
1362 * expression. At the end of the planning process it is replaced by either
1363 * the contained expression or a Var referring to a lower-level evaluation of
1364 * the contained expression. Typically the evaluation occurs below an outer
1365 * join, and Var references above the outer join might thereby yield NULL
1366 * instead of the expression value.
1368 * Although the planner treats this as an expression node type, it is not
1369 * recognized by the parser or executor, so we declare it here rather than
1373 typedef struct PlaceHolderVar
1376 Expr *phexpr; /* the represented expression */
1377 Relids phrels; /* base relids syntactically within expr src */
1378 Index phid; /* ID for PHV (unique within planner run) */
1379 Index phlevelsup; /* > 0 if PHV belongs to outer query */
1383 * "Special join" info.
1385 * One-sided outer joins constrain the order of joining partially but not
1386 * completely. We flatten such joins into the planner's top-level list of
1387 * relations to join, but record information about each outer join in a
1388 * SpecialJoinInfo struct. These structs are kept in the PlannerInfo node's
1391 * Similarly, semijoins and antijoins created by flattening IN (subselect)
1392 * and EXISTS(subselect) clauses create partial constraints on join order.
1393 * These are likewise recorded in SpecialJoinInfo structs.
1395 * We make SpecialJoinInfos for FULL JOINs even though there is no flexibility
1396 * of planning for them, because this simplifies make_join_rel()'s API.
1398 * min_lefthand and min_righthand are the sets of base relids that must be
1399 * available on each side when performing the special join. lhs_strict is
1400 * true if the special join's condition cannot succeed when the LHS variables
1401 * are all NULL (this means that an outer join can commute with upper-level
1402 * outer joins even if it appears in their RHS). We don't bother to set
1403 * lhs_strict for FULL JOINs, however.
1405 * It is not valid for either min_lefthand or min_righthand to be empty sets;
1406 * if they were, this would break the logic that enforces join order.
1408 * syn_lefthand and syn_righthand are the sets of base relids that are
1409 * syntactically below this special join. (These are needed to help compute
1410 * min_lefthand and min_righthand for higher joins.)
1412 * delay_upper_joins is set TRUE if we detect a pushed-down clause that has
1413 * to be evaluated after this join is formed (because it references the RHS).
1414 * Any outer joins that have such a clause and this join in their RHS cannot
1415 * commute with this join, because that would leave noplace to check the
1416 * pushed-down clause. (We don't track this for FULL JOINs, either.)
1418 * For a semijoin, we also extract the join operators and their RHS arguments
1419 * and set semi_operators, semi_rhs_exprs, semi_can_btree, and semi_can_hash.
1420 * This is done in support of possibly unique-ifying the RHS, so we don't
1421 * bother unless at least one of semi_can_btree and semi_can_hash can be set
1422 * true. (You might expect that this information would be computed during
1423 * join planning; but it's helpful to have it available during planning of
1424 * parameterized table scans, so we store it in the SpecialJoinInfo structs.)
1426 * jointype is never JOIN_RIGHT; a RIGHT JOIN is handled by switching
1427 * the inputs to make it a LEFT JOIN. So the allowed values of jointype
1428 * in a join_info_list member are only LEFT, FULL, SEMI, or ANTI.
1430 * For purposes of join selectivity estimation, we create transient
1431 * SpecialJoinInfo structures for regular inner joins; so it is possible
1432 * to have jointype == JOIN_INNER in such a structure, even though this is
1433 * not allowed within join_info_list. We also create transient
1434 * SpecialJoinInfos with jointype == JOIN_INNER for outer joins, since for
1435 * cost estimation purposes it is sometimes useful to know the join size under
1436 * plain innerjoin semantics. Note that lhs_strict, delay_upper_joins, and
1437 * of course the semi_xxx fields are not set meaningfully within such structs.
1440 typedef struct SpecialJoinInfo
1443 Relids min_lefthand; /* base relids in minimum LHS for join */
1444 Relids min_righthand; /* base relids in minimum RHS for join */
1445 Relids syn_lefthand; /* base relids syntactically within LHS */
1446 Relids syn_righthand; /* base relids syntactically within RHS */
1447 JoinType jointype; /* always INNER, LEFT, FULL, SEMI, or ANTI */
1448 bool lhs_strict; /* joinclause is strict for some LHS rel */
1449 bool delay_upper_joins; /* can't commute with upper RHS */
1450 /* Remaining fields are set only for JOIN_SEMI jointype: */
1451 bool semi_can_btree; /* true if semi_operators are all btree */
1452 bool semi_can_hash; /* true if semi_operators are all hash */
1453 List *semi_operators; /* OIDs of equality join operators */
1454 List *semi_rhs_exprs; /* righthand-side expressions of these ops */
1458 * "Lateral join" info.
1460 * Lateral references constrain the join order in a way that's somewhat like
1461 * outer joins, though different in detail. We construct a LateralJoinInfo
1462 * for each lateral cross-reference, placing them in the PlannerInfo node's
1463 * lateral_info_list.
1465 * For unflattened LATERAL RTEs, we generate LateralJoinInfo(s) in which
1466 * lateral_rhs is the relid of the LATERAL baserel, and lateral_lhs is a set
1467 * of relids of baserels it references, all of which must be present on the
1468 * LHS to compute a parameter needed by the RHS. Typically, lateral_lhs is
1469 * a singleton, but it can include multiple rels if the RHS references a
1470 * PlaceHolderVar with a multi-rel ph_eval_at level. We disallow joining to
1471 * only part of the LHS in such cases, since that would result in a join tree
1472 * with no convenient place to compute the PHV.
1474 * When an appendrel contains lateral references (eg "LATERAL (SELECT x.col1
1475 * UNION ALL SELECT y.col2)"), the LateralJoinInfos reference the parent
1476 * baserel not the member otherrels, since it is the parent relid that is
1477 * considered for joining purposes.
1479 * If any LATERAL RTEs were flattened into the parent query, it is possible
1480 * that the query now contains PlaceHolderVars containing lateral references,
1481 * representing expressions that need to be evaluated at particular spots in
1482 * the jointree but contain lateral references to Vars from elsewhere. These
1483 * give rise to LateralJoinInfos in which lateral_rhs is the evaluation point
1484 * of a PlaceHolderVar and lateral_lhs is the set of lateral rels it needs.
1487 typedef struct LateralJoinInfo
1490 Relids lateral_lhs; /* rels needed to compute a lateral value */
1491 Relids lateral_rhs; /* rel where lateral value is needed */
1495 * Append-relation info.
1497 * When we expand an inheritable table or a UNION-ALL subselect into an
1498 * "append relation" (essentially, a list of child RTEs), we build an
1499 * AppendRelInfo for each child RTE. The list of AppendRelInfos indicates
1500 * which child RTEs must be included when expanding the parent, and each
1501 * node carries information needed to translate Vars referencing the parent
1502 * into Vars referencing that child.
1504 * These structs are kept in the PlannerInfo node's append_rel_list.
1505 * Note that we just throw all the structs into one list, and scan the
1506 * whole list when desiring to expand any one parent. We could have used
1507 * a more complex data structure (eg, one list per parent), but this would
1508 * be harder to update during operations such as pulling up subqueries,
1509 * and not really any easier to scan. Considering that typical queries
1510 * will not have many different append parents, it doesn't seem worthwhile
1511 * to complicate things.
1513 * Note: after completion of the planner prep phase, any given RTE is an
1514 * append parent having entries in append_rel_list if and only if its
1515 * "inh" flag is set. We clear "inh" for plain tables that turn out not
1516 * to have inheritance children, and (in an abuse of the original meaning
1517 * of the flag) we set "inh" for subquery RTEs that turn out to be
1518 * flattenable UNION ALL queries. This lets us avoid useless searches
1519 * of append_rel_list.
1521 * Note: the data structure assumes that append-rel members are single
1522 * baserels. This is OK for inheritance, but it prevents us from pulling
1523 * up a UNION ALL member subquery if it contains a join. While that could
1524 * be fixed with a more complex data structure, at present there's not much
1525 * point because no improvement in the plan could result.
1528 typedef struct AppendRelInfo
1533 * These fields uniquely identify this append relationship. There can be
1534 * (in fact, always should be) multiple AppendRelInfos for the same
1535 * parent_relid, but never more than one per child_relid, since a given
1536 * RTE cannot be a child of more than one append parent.
1538 Index parent_relid; /* RT index of append parent rel */
1539 Index child_relid; /* RT index of append child rel */
1542 * For an inheritance appendrel, the parent and child are both regular
1543 * relations, and we store their rowtype OIDs here for use in translating
1544 * whole-row Vars. For a UNION-ALL appendrel, the parent and child are
1545 * both subqueries with no named rowtype, and we store InvalidOid here.
1547 Oid parent_reltype; /* OID of parent's composite type */
1548 Oid child_reltype; /* OID of child's composite type */
1551 * The N'th element of this list is a Var or expression representing the
1552 * child column corresponding to the N'th column of the parent. This is
1553 * used to translate Vars referencing the parent rel into references to
1554 * the child. A list element is NULL if it corresponds to a dropped
1555 * column of the parent (this is only possible for inheritance cases, not
1556 * UNION ALL). The list elements are always simple Vars for inheritance
1557 * cases, but can be arbitrary expressions in UNION ALL cases.
1559 * Notice we only store entries for user columns (attno > 0). Whole-row
1560 * Vars are special-cased, and system columns (attno < 0) need no special
1561 * translation since their attnos are the same for all tables.
1563 * Caution: the Vars have varlevelsup = 0. Be careful to adjust as needed
1564 * when copying into a subquery.
1566 List *translated_vars; /* Expressions in the child's Vars */
1569 * We store the parent table's OID here for inheritance, or InvalidOid for
1570 * UNION ALL. This is only needed to help in generating error messages if
1571 * an attempt is made to reference a dropped parent column.
1573 Oid parent_reloid; /* OID of parent relation */
1577 * For each distinct placeholder expression generated during planning, we
1578 * store a PlaceHolderInfo node in the PlannerInfo node's placeholder_list.
1579 * This stores info that is needed centrally rather than in each copy of the
1580 * PlaceHolderVar. The phid fields identify which PlaceHolderInfo goes with
1581 * each PlaceHolderVar. Note that phid is unique throughout a planner run,
1582 * not just within a query level --- this is so that we need not reassign ID's
1583 * when pulling a subquery into its parent.
1585 * The idea is to evaluate the expression at (only) the ph_eval_at join level,
1586 * then allow it to bubble up like a Var until the ph_needed join level.
1587 * ph_needed has the same definition as attr_needed for a regular Var.
1589 * The PlaceHolderVar's expression might contain LATERAL references to vars
1590 * coming from outside its syntactic scope. If so, those rels are *not*
1591 * included in ph_eval_at, but they are recorded in ph_lateral.
1593 * Notice that when ph_eval_at is a join rather than a single baserel, the
1594 * PlaceHolderInfo may create constraints on join order: the ph_eval_at join
1595 * has to be formed below any outer joins that should null the PlaceHolderVar.
1597 * We create a PlaceHolderInfo only after determining that the PlaceHolderVar
1598 * is actually referenced in the plan tree, so that unreferenced placeholders
1599 * don't result in unnecessary constraints on join order.
1602 typedef struct PlaceHolderInfo
1606 Index phid; /* ID for PH (unique within planner run) */
1607 PlaceHolderVar *ph_var; /* copy of PlaceHolderVar tree */
1608 Relids ph_eval_at; /* lowest level we can evaluate value at */
1609 Relids ph_lateral; /* relids of contained lateral refs, if any */
1610 Relids ph_needed; /* highest level the value is needed at */
1611 int32 ph_width; /* estimated attribute width */
1615 * For each potentially index-optimizable MIN/MAX aggregate function,
1616 * root->minmax_aggs stores a MinMaxAggInfo describing it.
1618 typedef struct MinMaxAggInfo
1622 Oid aggfnoid; /* pg_proc Oid of the aggregate */
1623 Oid aggsortop; /* Oid of its sort operator */
1624 Expr *target; /* expression we are aggregating on */
1625 PlannerInfo *subroot; /* modified "root" for planning the subquery */
1626 Path *path; /* access path for subquery */
1627 Cost pathcost; /* estimated cost to fetch first row */
1628 Param *param; /* param for subplan's output */
1632 * At runtime, PARAM_EXEC slots are used to pass values around from one plan
1633 * node to another. They can be used to pass values down into subqueries (for
1634 * outer references in subqueries), or up out of subqueries (for the results
1635 * of a subplan), or from a NestLoop plan node into its inner relation (when
1636 * the inner scan is parameterized with values from the outer relation).
1637 * The planner is responsible for assigning nonconflicting PARAM_EXEC IDs to
1638 * the PARAM_EXEC Params it generates.
1640 * Outer references are managed via root->plan_params, which is a list of
1641 * PlannerParamItems. While planning a subquery, each parent query level's
1642 * plan_params contains the values required from it by the current subquery.
1643 * During create_plan(), we use plan_params to track values that must be
1644 * passed from outer to inner sides of NestLoop plan nodes.
1646 * The item a PlannerParamItem represents can be one of three kinds:
1648 * A Var: the slot represents a variable of this level that must be passed
1649 * down because subqueries have outer references to it, or must be passed
1650 * from a NestLoop node to its inner scan. The varlevelsup value in the Var
1651 * will always be zero.
1653 * A PlaceHolderVar: this works much like the Var case, except that the
1654 * entry is a PlaceHolderVar node with a contained expression. The PHV
1655 * will have phlevelsup = 0, and the contained expression is adjusted
1656 * to match in level.
1658 * An Aggref (with an expression tree representing its argument): the slot
1659 * represents an aggregate expression that is an outer reference for some
1660 * subquery. The Aggref itself has agglevelsup = 0, and its argument tree
1661 * is adjusted to match in level.
1663 * Note: we detect duplicate Var and PlaceHolderVar parameters and coalesce
1664 * them into one slot, but we do not bother to do that for Aggrefs.
1665 * The scope of duplicate-elimination only extends across the set of
1666 * parameters passed from one query level into a single subquery, or for
1667 * nestloop parameters across the set of nestloop parameters used in a single
1668 * query level. So there is no possibility of a PARAM_EXEC slot being used
1669 * for conflicting purposes.
1671 * In addition, PARAM_EXEC slots are assigned for Params representing outputs
1672 * from subplans (values that are setParam items for those subplans). These
1673 * IDs need not be tracked via PlannerParamItems, since we do not need any
1674 * duplicate-elimination nor later processing of the represented expressions.
1675 * Instead, we just record the assignment of the slot number by incrementing
1676 * root->glob->nParamExec.
1678 typedef struct PlannerParamItem
1682 Node *item; /* the Var, PlaceHolderVar, or Aggref */
1683 int paramId; /* its assigned PARAM_EXEC slot number */
1687 * When making cost estimates for a SEMI or ANTI join, there are some
1688 * correction factors that are needed in both nestloop and hash joins
1689 * to account for the fact that the executor can stop scanning inner rows
1690 * as soon as it finds a match to the current outer row. These numbers
1691 * depend only on the selected outer and inner join relations, not on the
1692 * particular paths used for them, so it's worthwhile to calculate them
1693 * just once per relation pair not once per considered path. This struct
1694 * is filled by compute_semi_anti_join_factors and must be passed along
1695 * to the join cost estimation functions.
1697 * outer_match_frac is the fraction of the outer tuples that are
1698 * expected to have at least one match.
1699 * match_count is the average number of matches expected for
1700 * outer tuples that have at least one match.
1702 typedef struct SemiAntiJoinFactors
1704 Selectivity outer_match_frac;
1705 Selectivity match_count;
1706 } SemiAntiJoinFactors;
1709 * Struct for extra information passed to subroutines of add_paths_to_joinrel
1711 * restrictlist contains all of the RestrictInfo nodes for restriction
1712 * clauses that apply to this join
1713 * mergeclause_list is a list of RestrictInfo nodes for available
1714 * mergejoin clauses in this join
1715 * sjinfo is extra info about special joins for selectivity estimation
1716 * semifactors is as shown above (only valid for SEMI or ANTI joins)
1717 * param_source_rels are OK targets for parameterization of result paths
1718 * extra_lateral_rels are additional parameterization for result paths
1720 typedef struct JoinPathExtraData
1723 List *mergeclause_list;
1724 SpecialJoinInfo *sjinfo;
1725 SemiAntiJoinFactors semifactors;
1726 Relids param_source_rels;
1727 Relids extra_lateral_rels;
1728 } JoinPathExtraData;
1731 * For speed reasons, cost estimation for join paths is performed in two
1732 * phases: the first phase tries to quickly derive a lower bound for the
1733 * join cost, and then we check if that's sufficient to reject the path.
1734 * If not, we come back for a more refined cost estimate. The first phase
1735 * fills a JoinCostWorkspace struct with its preliminary cost estimates
1736 * and possibly additional intermediate values. The second phase takes
1737 * these values as inputs to avoid repeating work.
1739 * (Ideally we'd declare this in cost.h, but it's also needed in pathnode.h,
1740 * so seems best to put it here.)
1742 typedef struct JoinCostWorkspace
1744 /* Preliminary cost estimates --- must not be larger than final ones! */
1745 Cost startup_cost; /* cost expended before fetching any tuples */
1746 Cost total_cost; /* total cost (assuming all tuples fetched) */
1748 /* Fields below here should be treated as private to costsize.c */
1749 Cost run_cost; /* non-startup cost components */
1751 /* private for cost_nestloop code */
1752 Cost inner_run_cost; /* also used by cost_mergejoin code */
1753 Cost inner_rescan_run_cost;
1755 /* private for cost_mergejoin code */
1758 double outer_skip_rows;
1759 double inner_skip_rows;
1761 /* private for cost_hashjoin code */
1764 } JoinCostWorkspace;
1766 #endif /* RELATION_H */