1 /*-------------------------------------------------------------------------
4 * Definitions for planner's internal data structures.
7 * Portions Copyright (c) 1996-2010, 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 "nodes/bitmapset.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 */
52 * Global information for planning/optimization
54 * PlannerGlobal holds state for an entire planner invocation; this state
55 * is shared across all levels of sub-Queries that exist in the command being
59 typedef struct PlannerGlobal
63 ParamListInfo boundParams; /* Param values provided to planner() */
65 List *paramlist; /* to keep track of cross-level Params */
67 List *subplans; /* Plans for SubPlan nodes */
69 List *subrtables; /* Rangetables for SubPlan nodes */
71 List *subrowmarks; /* PlanRowMarks for SubPlan nodes */
73 Bitmapset *rewindPlanIDs; /* indices of subplans that require REWIND */
75 List *finalrtable; /* "flat" rangetable for executor */
77 List *finalrowmarks; /* "flat" list of PlanRowMarks */
79 List *relationOids; /* OIDs of relations the plan depends on */
81 List *invalItems; /* other dependencies, as PlanInvalItems */
83 Index lastPHId; /* highest PlaceHolderVar ID assigned */
85 bool transientPlan; /* redo plan when TransactionXmin changes? */
88 /* macro for fetching the Plan associated with a SubPlan node */
89 #define planner_subplan_get_plan(root, subplan) \
90 ((Plan *) list_nth((root)->glob->subplans, (subplan)->plan_id - 1))
95 * Per-query information for planning/optimization
97 * This struct is conventionally called "root" in all the planner routines.
98 * It holds links to all of the planner's working state, in addition to the
99 * original Query. Note that at present the planner extensively modifies
100 * the passed-in Query data structure; someday that should stop.
103 typedef struct PlannerInfo
107 Query *parse; /* the Query being planned */
109 PlannerGlobal *glob; /* global info for current planner run */
111 Index query_level; /* 1 at the outermost Query */
113 struct PlannerInfo *parent_root; /* NULL at outermost Query */
116 * simple_rel_array holds pointers to "base rels" and "other rels" (see
117 * comments for RelOptInfo for more info). It is indexed by rangetable
118 * index (so entry 0 is always wasted). Entries can be NULL when an RTE
119 * does not correspond to a base relation, such as a join RTE or an
120 * unreferenced view RTE; or if the RelOptInfo hasn't been made yet.
122 struct RelOptInfo **simple_rel_array; /* All 1-rel RelOptInfos */
123 int simple_rel_array_size; /* allocated size of array */
126 * simple_rte_array is the same length as simple_rel_array and holds
127 * pointers to the associated rangetable entries. This lets us avoid
128 * rt_fetch(), which can be a bit slow once large inheritance sets have
131 RangeTblEntry **simple_rte_array; /* rangetable as an array */
134 * join_rel_list is a list of all join-relation RelOptInfos we have
135 * considered in this planning run. For small problems we just scan the
136 * list to do lookups, but when there are many join relations we build a
137 * hash table for faster lookups. The hash table is present and valid
138 * when join_rel_hash is not NULL. Note that we still maintain the list
139 * even when using the hash table for lookups; this simplifies life for
142 List *join_rel_list; /* list of join-relation RelOptInfos */
143 struct HTAB *join_rel_hash; /* optional hashtable for join relations */
146 * When doing a dynamic-programming-style join search, join_rel_level[k]
147 * is a list of all join-relation RelOptInfos of level k, and
148 * join_cur_level is the current level. New join-relation RelOptInfos are
149 * automatically added to the join_rel_level[join_cur_level] list.
150 * join_rel_level is NULL if not in use.
152 List **join_rel_level; /* lists of join-relation RelOptInfos */
153 int join_cur_level; /* index of list being extended */
155 List *resultRelations; /* integer list of RT indexes, or NIL */
157 List *init_plans; /* init SubPlans for query */
159 List *cte_plan_ids; /* per-CTE-item list of subplan IDs */
161 List *eq_classes; /* list of active EquivalenceClasses */
163 List *canon_pathkeys; /* list of "canonical" PathKeys */
165 List *left_join_clauses; /* list of RestrictInfos for
166 * mergejoinable outer join clauses
167 * w/nonnullable var on left */
169 List *right_join_clauses; /* list of RestrictInfos for
170 * mergejoinable outer join clauses
171 * w/nonnullable var on right */
173 List *full_join_clauses; /* list of RestrictInfos for
174 * mergejoinable full join clauses */
176 List *join_info_list; /* list of SpecialJoinInfos */
178 List *append_rel_list; /* list of AppendRelInfos */
180 List *rowMarks; /* list of PlanRowMarks */
182 List *placeholder_list; /* list of PlaceHolderInfos */
184 List *query_pathkeys; /* desired pathkeys for query_planner(), and
185 * actual pathkeys afterwards */
187 List *group_pathkeys; /* groupClause pathkeys, if any */
188 List *window_pathkeys; /* pathkeys of bottom window, if any */
189 List *distinct_pathkeys; /* distinctClause pathkeys, if any */
190 List *sort_pathkeys; /* sortClause pathkeys, if any */
192 List *minmax_aggs; /* List of MinMaxAggInfos */
194 List *initial_rels; /* RelOptInfos we are now trying to join */
196 MemoryContext planner_cxt; /* context holding PlannerInfo */
198 double total_table_pages; /* # of pages in all tables of query */
200 double tuple_fraction; /* tuple_fraction passed to query_planner */
201 double limit_tuples; /* limit_tuples passed to query_planner */
203 bool hasInheritedTarget; /* true if parse->resultRelation is an
204 * inheritance child rel */
205 bool hasJoinRTEs; /* true if any RTEs are RTE_JOIN kind */
206 bool hasHavingQual; /* true if havingQual was non-null */
207 bool hasPseudoConstantQuals; /* true if any RestrictInfo has
208 * pseudoconstant = true */
209 bool hasRecursion; /* true if planning a recursive WITH item */
211 /* These fields are used only when hasRecursion is true: */
212 int wt_param_id; /* PARAM_EXEC ID for the work table */
213 struct Plan *non_recursive_plan; /* plan for non-recursive term */
215 /* These fields are workspace for createplan.c */
216 Relids curOuterRels; /* outer rels above current node */
217 List *curOuterParams; /* not-yet-assigned NestLoopParams */
219 /* optional private data for join_search_hook, e.g., GEQO */
220 void *join_search_private;
225 * In places where it's known that simple_rte_array[] must have been prepared
226 * already, we just index into it to fetch RTEs. In code that might be
227 * executed before or after entering query_planner(), use this macro.
229 #define planner_rt_fetch(rti, root) \
230 ((root)->simple_rte_array ? (root)->simple_rte_array[rti] : \
231 rt_fetch(rti, (root)->parse->rtable))
236 * Per-relation information for planning/optimization
238 * For planning purposes, a "base rel" is either a plain relation (a table)
239 * or the output of a sub-SELECT or function that appears in the range table.
240 * In either case it is uniquely identified by an RT index. A "joinrel"
241 * is the joining of two or more base rels. A joinrel is identified by
242 * the set of RT indexes for its component baserels. We create RelOptInfo
243 * nodes for each baserel and joinrel, and store them in the PlannerInfo's
244 * simple_rel_array and join_rel_list respectively.
246 * Note that there is only one joinrel for any given set of component
247 * baserels, no matter what order we assemble them in; so an unordered
248 * set is the right datatype to identify it with.
250 * We also have "other rels", which are like base rels in that they refer to
251 * single RT indexes; but they are not part of the join tree, and are given
252 * a different RelOptKind to identify them. Lastly, there is a RelOptKind
253 * for "dead" relations, which are base rels that we have proven we don't
254 * need to join after all.
256 * Currently the only kind of otherrels are those made for member relations
257 * of an "append relation", that is an inheritance set or UNION ALL subquery.
258 * An append relation has a parent RTE that is a base rel, which represents
259 * the entire append relation. The member RTEs are otherrels. The parent
260 * is present in the query join tree but the members are not. The member
261 * RTEs and otherrels are used to plan the scans of the individual tables or
262 * subqueries of the append set; then the parent baserel is given Append
263 * and/or MergeAppend paths comprising the best paths for the individual
264 * member rels. (See comments for AppendRelInfo for more information.)
266 * At one time we also made otherrels to represent join RTEs, for use in
267 * handling join alias Vars. Currently this is not needed because all join
268 * alias Vars are expanded to non-aliased form during preprocess_expression.
270 * Parts of this data structure are specific to various scan and join
271 * mechanisms. It didn't seem worth creating new node types for them.
273 * relids - Set of base-relation identifiers; it is a base relation
274 * if there is just one, a join relation if more than one
275 * rows - estimated number of tuples in the relation after restriction
276 * clauses have been applied (ie, output rows of a plan for it)
277 * width - avg. number of bytes per tuple in the relation after the
278 * appropriate projections have been done (ie, output width)
279 * reltargetlist - List of Var and PlaceHolderVar nodes for the values
280 * we need to output from this relation.
281 * List is in no particular order, but all rels of an
282 * appendrel set must use corresponding orders.
283 * NOTE: in a child relation, may contain RowExpr or
284 * ConvertRowtypeExpr representing a whole-row Var.
285 * pathlist - List of Path nodes, one for each potentially useful
286 * method of generating the relation
287 * cheapest_startup_path - the pathlist member with lowest startup cost
288 * (regardless of its ordering)
289 * cheapest_total_path - the pathlist member with lowest total cost
290 * (regardless of its ordering)
291 * cheapest_unique_path - for caching cheapest path to produce unique
292 * (no duplicates) output from relation
294 * If the relation is a base relation it will have these fields set:
296 * relid - RTE index (this is redundant with the relids field, but
297 * is provided for convenience of access)
298 * rtekind - distinguishes plain relation, subquery, or function RTE
299 * min_attr, max_attr - range of valid AttrNumbers for rel
300 * attr_needed - array of bitmapsets indicating the highest joinrel
301 * in which each attribute is needed; if bit 0 is set then
302 * the attribute is needed as part of final targetlist
303 * attr_widths - cache space for per-attribute width estimates;
304 * zero means not computed yet
305 * indexlist - list of IndexOptInfo nodes for relation's indexes
306 * (always NIL if it's not a table)
307 * pages - number of disk pages in relation (zero if not a table)
308 * tuples - number of tuples in relation (not considering restrictions)
309 * subplan - plan for subquery (NULL if it's not a subquery)
310 * subrtable - rangetable for subquery (NIL if it's not a subquery)
311 * subrowmark - rowmarks for subquery (NIL if it's not a subquery)
313 * Note: for a subquery, tuples and subplan are not set immediately
314 * upon creation of the RelOptInfo object; they are filled in when
315 * set_base_rel_pathlist processes the object.
317 * For otherrels that are appendrel members, these fields are filled
318 * in just as for a baserel.
320 * The presence of the remaining fields depends on the restrictions
321 * and joins that the relation participates in:
323 * baserestrictinfo - List of RestrictInfo nodes, containing info about
324 * each non-join qualification clause in which this relation
325 * participates (only used for base rels)
326 * baserestrictcost - Estimated cost of evaluating the baserestrictinfo
327 * clauses at a single tuple (only used for base rels)
328 * joininfo - List of RestrictInfo nodes, containing info about each
329 * join clause in which this relation participates (but
330 * note this excludes clauses that might be derivable from
331 * EquivalenceClasses)
332 * has_eclass_joins - flag that EquivalenceClass joins are possible
333 * index_outer_relids - only used for base rels; set of outer relids
334 * that participate in indexable joinclauses for this rel
335 * index_inner_paths - only used for base rels; list of InnerIndexscanInfo
336 * nodes showing best indexpaths for various subsets of
337 * index_outer_relids.
339 * Note: Keeping a restrictinfo list in the RelOptInfo is useful only for
340 * base rels, because for a join rel the set of clauses that are treated as
341 * restrict clauses varies depending on which sub-relations we choose to join.
342 * (For example, in a 3-base-rel join, a clause relating rels 1 and 2 must be
343 * treated as a restrictclause if we join {1} and {2 3} to make {1 2 3}; but
344 * if we join {1 2} and {3} then that clause will be a restrictclause in {1 2}
345 * and should not be processed again at the level of {1 2 3}.) Therefore,
346 * the restrictinfo list in the join case appears in individual JoinPaths
347 * (field joinrestrictinfo), not in the parent relation. But it's OK for
348 * the RelOptInfo to store the joininfo list, because that is the same
349 * for a given rel no matter how we form it.
351 * We store baserestrictcost in the RelOptInfo (for base relations) because
352 * we know we will need it at least once (to price the sequential scan)
353 * and may need it multiple times to price index scans.
356 typedef enum RelOptKind
360 RELOPT_OTHER_MEMBER_REL,
364 typedef struct RelOptInfo
368 RelOptKind reloptkind;
370 /* all relations included in this RelOptInfo */
371 Relids relids; /* set of base relids (rangetable indexes) */
373 /* size estimates generated by planner */
374 double rows; /* estimated number of result tuples */
375 int width; /* estimated avg width of result tuples */
377 /* materialization information */
378 List *reltargetlist; /* Vars to be output by scan of relation */
379 List *pathlist; /* Path structures */
380 struct Path *cheapest_startup_path;
381 struct Path *cheapest_total_path;
382 struct Path *cheapest_unique_path;
384 /* information about a base rel (not set for join rels!) */
386 Oid reltablespace; /* containing tablespace */
387 RTEKind rtekind; /* RELATION, SUBQUERY, or FUNCTION */
388 AttrNumber min_attr; /* smallest attrno of rel (often <0) */
389 AttrNumber max_attr; /* largest attrno of rel */
390 Relids *attr_needed; /* array indexed [min_attr .. max_attr] */
391 int32 *attr_widths; /* array indexed [min_attr .. max_attr] */
392 List *indexlist; /* list of IndexOptInfo */
395 struct Plan *subplan; /* if subquery */
396 List *subrtable; /* if subquery */
397 List *subrowmark; /* if subquery */
399 /* used by various scans and joins: */
400 List *baserestrictinfo; /* RestrictInfo structures (if base
402 QualCost baserestrictcost; /* cost of evaluating the above */
403 List *joininfo; /* RestrictInfo structures for join clauses
404 * involving this rel */
405 bool has_eclass_joins; /* T means joininfo is incomplete */
407 /* cached info about inner indexscan paths for relation: */
408 Relids index_outer_relids; /* other relids in indexable join
410 List *index_inner_paths; /* InnerIndexscanInfo nodes */
413 * Inner indexscans are not in the main pathlist because they are not
414 * usable except in specific join contexts. We use the index_inner_paths
415 * list just to avoid recomputing the best inner indexscan repeatedly for
416 * similar outer relations. See comments for InnerIndexscanInfo.
422 * Per-index information for planning/optimization
424 * Prior to Postgres 7.0, RelOptInfo was used to describe both relations
425 * and indexes, but that created confusion without actually doing anything
426 * useful. So now we have a separate IndexOptInfo struct for indexes.
428 * opfamily[], indexkeys[], opcintype[], fwdsortop[], revsortop[],
429 * and nulls_first[] each have ncolumns entries.
431 * Zeroes in the indexkeys[] array indicate index columns that are
432 * expressions; there is one element in indexprs for each such column.
434 * For an unordered index, the sortop arrays contains zeroes. Note that
435 * fwdsortop[] and nulls_first[] describe the sort ordering of a forward
436 * indexscan; we can also consider a backward indexscan, which will
437 * generate sort order described by revsortop/!nulls_first.
439 * The indexprs and indpred expressions have been run through
440 * prepqual.c and eval_const_expressions() for ease of matching to
441 * WHERE clauses. indpred is in implicit-AND form.
443 typedef struct IndexOptInfo
447 Oid indexoid; /* OID of the index relation */
448 Oid reltablespace; /* tablespace of index (not table) */
449 RelOptInfo *rel; /* back-link to index's table */
451 /* statistics from pg_class */
452 BlockNumber pages; /* number of disk pages in index */
453 double tuples; /* number of index tuples in index */
455 /* index descriptor information */
456 int ncolumns; /* number of columns in index */
457 Oid *opfamily; /* OIDs of operator families for columns */
458 int *indexkeys; /* column numbers of index's keys, or 0 */
459 Oid *opcintype; /* OIDs of opclass declared input data types */
460 Oid *fwdsortop; /* OIDs of sort operators for each column */
461 Oid *revsortop; /* OIDs of sort operators for backward scan */
462 bool *nulls_first; /* do NULLs come first in the sort order? */
463 Oid relam; /* OID of the access method (in pg_am) */
465 RegProcedure amcostestimate; /* OID of the access method's cost fcn */
467 List *indexprs; /* expressions for non-simple index columns */
468 List *indpred; /* predicate if a partial index, else NIL */
470 bool predOK; /* true if predicate matches query */
471 bool unique; /* true if a unique index */
472 bool amcanorderbyop; /* does AM support order by operator result? */
473 bool amoptionalkey; /* can query omit key for the first column? */
474 bool amsearchnulls; /* can AM search for NULL/NOT NULL entries? */
475 bool amhasgettuple; /* does AM have amgettuple interface? */
476 bool amhasgetbitmap; /* does AM have amgetbitmap interface? */
483 * Whenever we can determine that a mergejoinable equality clause A = B is
484 * not delayed by any outer join, we create an EquivalenceClass containing
485 * the expressions A and B to record this knowledge. If we later find another
486 * equivalence B = C, we add C to the existing EquivalenceClass; this may
487 * require merging two existing EquivalenceClasses. At the end of the qual
488 * distribution process, we have sets of values that are known all transitively
489 * equal to each other, where "equal" is according to the rules of the btree
490 * operator family(s) shown in ec_opfamilies. (We restrict an EC to contain
491 * only equalities whose operators belong to the same set of opfamilies. This
492 * could probably be relaxed, but for now it's not worth the trouble, since
493 * nearly all equality operators belong to only one btree opclass anyway.)
495 * We also use EquivalenceClasses as the base structure for PathKeys, letting
496 * us represent knowledge about different sort orderings being equivalent.
497 * Since every PathKey must reference an EquivalenceClass, we will end up
498 * with single-member EquivalenceClasses whenever a sort key expression has
499 * not been equivalenced to anything else. It is also possible that such an
500 * EquivalenceClass will contain a volatile expression ("ORDER BY random()"),
501 * which is a case that can't arise otherwise since clauses containing
502 * volatile functions are never considered mergejoinable. We mark such
503 * EquivalenceClasses specially to prevent them from being merged with
504 * ordinary EquivalenceClasses. Also, for volatile expressions we have
505 * to be careful to match the EquivalenceClass to the correct targetlist
506 * entry: consider SELECT random() AS a, random() AS b ... ORDER BY b,a.
507 * So we record the SortGroupRef of the originating sort clause.
509 * We allow equality clauses appearing below the nullable side of an outer join
510 * to form EquivalenceClasses, but these have a slightly different meaning:
511 * the included values might be all NULL rather than all the same non-null
512 * values. See src/backend/optimizer/README for more on that point.
514 * NB: if ec_merged isn't NULL, this class has been merged into another, and
515 * should be ignored in favor of using the pointed-to class.
517 typedef struct EquivalenceClass
521 List *ec_opfamilies; /* btree operator family OIDs */
522 List *ec_members; /* list of EquivalenceMembers */
523 List *ec_sources; /* list of generating RestrictInfos */
524 List *ec_derives; /* list of derived RestrictInfos */
525 Relids ec_relids; /* all relids appearing in ec_members */
526 bool ec_has_const; /* any pseudoconstants in ec_members? */
527 bool ec_has_volatile; /* the (sole) member is a volatile expr */
528 bool ec_below_outer_join; /* equivalence applies below an OJ */
529 bool ec_broken; /* failed to generate needed clauses? */
530 Index ec_sortref; /* originating sortclause label, or 0 */
531 struct EquivalenceClass *ec_merged; /* set if merged into another EC */
535 * If an EC contains a const and isn't below-outer-join, any PathKey depending
536 * on it must be redundant, since there's only one possible value of the key.
538 #define EC_MUST_BE_REDUNDANT(eclass) \
539 ((eclass)->ec_has_const && !(eclass)->ec_below_outer_join)
542 * EquivalenceMember - one member expression of an EquivalenceClass
544 * em_is_child signifies that this element was built by transposing a member
545 * for an inheritance parent relation to represent the corresponding expression
546 * on an inheritance child. These elements are used for constructing
547 * inner-indexscan paths for the child relation (other types of join are
548 * driven from transposed joininfo-list entries) and for constructing
549 * MergeAppend paths for the whole inheritance tree. Note that the EC's
550 * ec_relids field does NOT include the child relation.
552 * em_datatype is usually the same as exprType(em_expr), but can be
553 * different when dealing with a binary-compatible opfamily; in particular
554 * anyarray_ops would never work without this. Use em_datatype when
555 * looking up a specific btree operator to work with this expression.
557 typedef struct EquivalenceMember
561 Expr *em_expr; /* the expression represented */
562 Relids em_relids; /* all relids appearing in em_expr */
563 bool em_is_const; /* expression is pseudoconstant? */
564 bool em_is_child; /* derived version for a child relation? */
565 Oid em_datatype; /* the "nominal type" used by the opfamily */
571 * The sort ordering of a path is represented by a list of PathKey nodes.
572 * An empty list implies no known ordering. Otherwise the first item
573 * represents the primary sort key, the second the first secondary sort key,
574 * etc. The value being sorted is represented by linking to an
575 * EquivalenceClass containing that value and including pk_opfamily among its
576 * ec_opfamilies. This is a convenient method because it makes it trivial
577 * to detect equivalent and closely-related orderings. (See optimizer/README
578 * for more information.)
580 * Note: pk_strategy is either BTLessStrategyNumber (for ASC) or
581 * BTGreaterStrategyNumber (for DESC). We assume that all ordering-capable
582 * index types will use btree-compatible strategy numbers.
585 typedef struct PathKey
589 EquivalenceClass *pk_eclass; /* the value that is ordered */
590 Oid pk_opfamily; /* btree opfamily defining the ordering */
591 int pk_strategy; /* sort direction (ASC or DESC) */
592 bool pk_nulls_first; /* do NULLs come before normal values? */
596 * Type "Path" is used as-is for sequential-scan paths, as well as some other
597 * simple plan types that we don't need any extra information in the path for.
598 * For other path types it is the first component of a larger struct.
600 * Note: "pathtype" is the NodeTag of the Plan node we could build from this
601 * Path. It is partially redundant with the Path's NodeTag, but allows us
602 * to use the same Path type for multiple Plan types where there is no need
603 * to distinguish the Plan type during path processing.
610 NodeTag pathtype; /* tag identifying scan/join method */
612 RelOptInfo *parent; /* the relation this path can build */
614 /* estimated execution costs for path (see costsize.c for more info) */
615 Cost startup_cost; /* cost expended before fetching any tuples */
616 Cost total_cost; /* total cost (assuming all tuples fetched) */
618 List *pathkeys; /* sort ordering of path's output */
619 /* pathkeys is a List of PathKey nodes; see above */
623 * IndexPath represents an index scan over a single index.
625 * 'indexinfo' is the index to be scanned.
627 * 'indexclauses' is a list of index qualification clauses, with implicit
628 * AND semantics across the list. Each clause is a RestrictInfo node from
629 * the query's WHERE or JOIN conditions.
631 * 'indexquals' has the same structure as 'indexclauses', but it contains
632 * the actual indexqual conditions that can be used with the index.
633 * In simple cases this is identical to 'indexclauses', but when special
634 * indexable operators appear in 'indexclauses', they are replaced by the
635 * derived indexscannable conditions in 'indexquals'.
637 * 'isjoininner' is TRUE if the path is a nestloop inner scan (that is,
638 * some of the index conditions are join rather than restriction clauses).
639 * Note that the path costs will be calculated differently from a plain
640 * indexscan in this case, and in addition there's a special 'rows' value
641 * different from the parent RelOptInfo's (see below).
643 * 'indexscandir' is one of:
644 * ForwardScanDirection: forward scan of an ordered index
645 * BackwardScanDirection: backward scan of an ordered index
646 * NoMovementScanDirection: scan of an unordered index, or don't care
647 * (The executor doesn't care whether it gets ForwardScanDirection or
648 * NoMovementScanDirection for an indexscan, but the planner wants to
649 * distinguish ordered from unordered indexes for building pathkeys.)
651 * 'indextotalcost' and 'indexselectivity' are saved in the IndexPath so that
652 * we need not recompute them when considering using the same index in a
653 * bitmap index/heap scan (see BitmapHeapPath). The costs of the IndexPath
654 * itself represent the costs of an IndexScan plan type.
656 * 'rows' is the estimated result tuple count for the indexscan. This
657 * is the same as path.parent->rows for a simple indexscan, but it is
658 * different for a nestloop inner scan, because the additional indexquals
659 * coming from join clauses make the scan more selective than the parent
660 * rel's restrict clauses alone would do.
663 typedef struct IndexPath
666 IndexOptInfo *indexinfo;
670 ScanDirection indexscandir;
672 Selectivity indexselectivity;
673 double rows; /* estimated number of result tuples */
677 * BitmapHeapPath represents one or more indexscans that generate TID bitmaps
678 * instead of directly accessing the heap, followed by AND/OR combinations
679 * to produce a single bitmap, followed by a heap scan that uses the bitmap.
680 * Note that the output is always considered unordered, since it will come
681 * out in physical heap order no matter what the underlying indexes did.
683 * The individual indexscans are represented by IndexPath nodes, and any
684 * logic on top of them is represented by a tree of BitmapAndPath and
685 * BitmapOrPath nodes. Notice that we can use the same IndexPath node both
686 * to represent a regular IndexScan plan, and as the child of a BitmapHeapPath
687 * that represents scanning the same index using a BitmapIndexScan. The
688 * startup_cost and total_cost figures of an IndexPath always represent the
689 * costs to use it as a regular IndexScan. The costs of a BitmapIndexScan
690 * can be computed using the IndexPath's indextotalcost and indexselectivity.
692 * BitmapHeapPaths can be nestloop inner indexscans. The isjoininner and
693 * rows fields serve the same purpose as for plain IndexPaths.
695 typedef struct BitmapHeapPath
698 Path *bitmapqual; /* IndexPath, BitmapAndPath, BitmapOrPath */
699 bool isjoininner; /* T if it's a nestloop inner scan */
700 double rows; /* estimated number of result tuples */
704 * BitmapAndPath represents a BitmapAnd plan node; it can only appear as
705 * part of the substructure of a BitmapHeapPath. The Path structure is
706 * a bit more heavyweight than we really need for this, but for simplicity
707 * we make it a derivative of Path anyway.
709 typedef struct BitmapAndPath
712 List *bitmapquals; /* IndexPaths and BitmapOrPaths */
713 Selectivity bitmapselectivity;
717 * BitmapOrPath represents a BitmapOr plan node; it can only appear as
718 * part of the substructure of a BitmapHeapPath. The Path structure is
719 * a bit more heavyweight than we really need for this, but for simplicity
720 * we make it a derivative of Path anyway.
722 typedef struct BitmapOrPath
725 List *bitmapquals; /* IndexPaths and BitmapAndPaths */
726 Selectivity bitmapselectivity;
730 * TidPath represents a scan by TID
732 * tidquals is an implicitly OR'ed list of qual expressions of the form
733 * "CTID = pseudoconstant" or "CTID = ANY(pseudoconstant_array)".
734 * Note they are bare expressions, not RestrictInfos.
736 typedef struct TidPath
739 List *tidquals; /* qual(s) involving CTID = something */
743 * AppendPath represents an Append plan, ie, successive execution of
744 * several member plans.
746 * Note: it is possible for "subpaths" to contain only one, or even no,
747 * elements. These cases are optimized during create_append_plan.
748 * In particular, an AppendPath with no subpaths is a "dummy" path that
749 * is created to represent the case that a relation is provably empty.
751 typedef struct AppendPath
754 List *subpaths; /* list of component Paths */
757 #define IS_DUMMY_PATH(p) \
758 (IsA((p), AppendPath) && ((AppendPath *) (p))->subpaths == NIL)
761 * MergeAppendPath represents a MergeAppend plan, ie, the merging of sorted
762 * results from several member plans to produce similarly-sorted output.
764 typedef struct MergeAppendPath
767 List *subpaths; /* list of component Paths */
768 double limit_tuples; /* hard limit on output tuples, or -1 */
772 * ResultPath represents use of a Result plan node to compute a variable-free
773 * targetlist with no underlying tables (a "SELECT expressions" query).
774 * The query could have a WHERE clause, too, represented by "quals".
776 * Note that quals is a list of bare clauses, not RestrictInfos.
778 typedef struct ResultPath
785 * MaterialPath represents use of a Material plan node, i.e., caching of
786 * the output of its subpath. This is used when the subpath is expensive
787 * and needs to be scanned repeatedly, or when we need mark/restore ability
788 * and the subpath doesn't have it.
790 typedef struct MaterialPath
797 * UniquePath represents elimination of distinct rows from the output of
800 * This is unlike the other Path nodes in that it can actually generate
801 * different plans: either hash-based or sort-based implementation, or a
802 * no-op if the input path can be proven distinct already. The decision
803 * is sufficiently localized that it's not worth having separate Path node
804 * types. (Note: in the no-op case, we could eliminate the UniquePath node
805 * entirely and just return the subpath; but it's convenient to have a
806 * UniquePath in the path tree to signal upper-level routines that the input
807 * is known distinct.)
811 UNIQUE_PATH_NOOP, /* input is known unique already */
812 UNIQUE_PATH_HASH, /* use hashing */
813 UNIQUE_PATH_SORT /* use sorting */
816 typedef struct UniquePath
820 UniquePathMethod umethod;
821 List *in_operators; /* equality operators of the IN clause */
822 List *uniq_exprs; /* expressions to be made unique */
823 double rows; /* estimated number of result tuples */
827 * All join-type paths share these fields.
830 typedef struct JoinPath
836 Path *outerjoinpath; /* path for the outer side of the join */
837 Path *innerjoinpath; /* path for the inner side of the join */
839 List *joinrestrictinfo; /* RestrictInfos to apply to join */
842 * See the notes for RelOptInfo to understand why joinrestrictinfo is
843 * needed in JoinPath, and can't be merged into the parent RelOptInfo.
848 * A nested-loop path needs no special fields.
851 typedef JoinPath NestPath;
854 * A mergejoin path has these fields.
856 * Unlike other path types, a MergePath node doesn't represent just a single
857 * run-time plan node: it can represent up to four. Aside from the MergeJoin
858 * node itself, there can be a Sort node for the outer input, a Sort node
859 * for the inner input, and/or a Material node for the inner input. We could
860 * represent these nodes by separate path nodes, but considering how many
861 * different merge paths are investigated during a complex join problem,
862 * it seems better to avoid unnecessary palloc overhead.
864 * path_mergeclauses lists the clauses (in the form of RestrictInfos)
865 * that will be used in the merge.
867 * Note that the mergeclauses are a subset of the parent relation's
868 * restriction-clause list. Any join clauses that are not mergejoinable
869 * appear only in the parent's restrict list, and must be checked by a
870 * qpqual at execution time.
872 * outersortkeys (resp. innersortkeys) is NIL if the outer path
873 * (resp. inner path) is already ordered appropriately for the
874 * mergejoin. If it is not NIL then it is a PathKeys list describing
875 * the ordering that must be created by an explicit Sort node.
877 * materialize_inner is TRUE if a Material node should be placed atop the
878 * inner input. This may appear with or without an inner Sort step.
881 typedef struct MergePath
884 List *path_mergeclauses; /* join clauses to be used for merge */
885 List *outersortkeys; /* keys for explicit sort, if any */
886 List *innersortkeys; /* keys for explicit sort, if any */
887 bool materialize_inner; /* add Materialize to inner? */
891 * A hashjoin path has these fields.
893 * The remarks above for mergeclauses apply for hashclauses as well.
895 * Hashjoin does not care what order its inputs appear in, so we have
896 * no need for sortkeys.
899 typedef struct HashPath
902 List *path_hashclauses; /* join clauses used for hashing */
903 int num_batches; /* number of batches expected */
907 * Restriction clause info.
909 * We create one of these for each AND sub-clause of a restriction condition
910 * (WHERE or JOIN/ON clause). Since the restriction clauses are logically
911 * ANDed, we can use any one of them or any subset of them to filter out
912 * tuples, without having to evaluate the rest. The RestrictInfo node itself
913 * stores data used by the optimizer while choosing the best query plan.
915 * If a restriction clause references a single base relation, it will appear
916 * in the baserestrictinfo list of the RelOptInfo for that base rel.
918 * If a restriction clause references more than one base rel, it will
919 * appear in the joininfo list of every RelOptInfo that describes a strict
920 * subset of the base rels mentioned in the clause. The joininfo lists are
921 * used to drive join tree building by selecting plausible join candidates.
922 * The clause cannot actually be applied until we have built a join rel
923 * containing all the base rels it references, however.
925 * When we construct a join rel that includes all the base rels referenced
926 * in a multi-relation restriction clause, we place that clause into the
927 * joinrestrictinfo lists of paths for the join rel, if neither left nor
928 * right sub-path includes all base rels referenced in the clause. The clause
929 * will be applied at that join level, and will not propagate any further up
930 * the join tree. (Note: the "predicate migration" code was once intended to
931 * push restriction clauses up and down the plan tree based on evaluation
932 * costs, but it's dead code and is unlikely to be resurrected in the
933 * foreseeable future.)
935 * Note that in the presence of more than two rels, a multi-rel restriction
936 * might reach different heights in the join tree depending on the join
937 * sequence we use. So, these clauses cannot be associated directly with
938 * the join RelOptInfo, but must be kept track of on a per-join-path basis.
940 * RestrictInfos that represent equivalence conditions (i.e., mergejoinable
941 * equalities that are not outerjoin-delayed) are handled a bit differently.
942 * Initially we attach them to the EquivalenceClasses that are derived from
943 * them. When we construct a scan or join path, we look through all the
944 * EquivalenceClasses and generate derived RestrictInfos representing the
945 * minimal set of conditions that need to be checked for this particular scan
946 * or join to enforce that all members of each EquivalenceClass are in fact
947 * equal in all rows emitted by the scan or join.
949 * When dealing with outer joins we have to be very careful about pushing qual
950 * clauses up and down the tree. An outer join's own JOIN/ON conditions must
951 * be evaluated exactly at that join node, unless they are "degenerate"
952 * conditions that reference only Vars from the nullable side of the join.
953 * Quals appearing in WHERE or in a JOIN above the outer join cannot be pushed
954 * down below the outer join, if they reference any nullable Vars.
955 * RestrictInfo nodes contain a flag to indicate whether a qual has been
956 * pushed down to a lower level than its original syntactic placement in the
957 * join tree would suggest. If an outer join prevents us from pushing a qual
958 * down to its "natural" semantic level (the level associated with just the
959 * base rels used in the qual) then we mark the qual with a "required_relids"
960 * value including more than just the base rels it actually uses. By
961 * pretending that the qual references all the rels required to form the outer
962 * join, we prevent it from being evaluated below the outer join's joinrel.
963 * When we do form the outer join's joinrel, we still need to distinguish
964 * those quals that are actually in that join's JOIN/ON condition from those
965 * that appeared elsewhere in the tree and were pushed down to the join rel
966 * because they used no other rels. That's what the is_pushed_down flag is
967 * for; it tells us that a qual is not an OUTER JOIN qual for the set of base
968 * rels listed in required_relids. A clause that originally came from WHERE
969 * or an INNER JOIN condition will *always* have its is_pushed_down flag set.
970 * It's possible for an OUTER JOIN clause to be marked is_pushed_down too,
971 * if we decide that it can be pushed down into the nullable side of the join.
972 * In that case it acts as a plain filter qual for wherever it gets evaluated.
973 * (In short, is_pushed_down is only false for non-degenerate outer join
974 * conditions. Possibly we should rename it to reflect that meaning?)
976 * RestrictInfo nodes also contain an outerjoin_delayed flag, which is true
977 * if the clause's applicability must be delayed due to any outer joins
978 * appearing below it (ie, it has to be postponed to some join level higher
979 * than the set of relations it actually references). There is also a
980 * nullable_relids field, which is the set of rels it references that can be
981 * forced null by some outer join below the clause. outerjoin_delayed = true
982 * is subtly different from nullable_relids != NULL: a clause might reference
983 * some nullable rels and yet not be outerjoin_delayed because it also
984 * references all the other rels of the outer join(s). A clause that is not
985 * outerjoin_delayed can be enforced anywhere it is computable.
987 * In general, the referenced clause might be arbitrarily complex. The
988 * kinds of clauses we can handle as indexscan quals, mergejoin clauses,
989 * or hashjoin clauses are limited (e.g., no volatile functions). The code
990 * for each kind of path is responsible for identifying the restrict clauses
991 * it can use and ignoring the rest. Clauses not implemented by an indexscan,
992 * mergejoin, or hashjoin will be placed in the plan qual or joinqual field
993 * of the finished Plan node, where they will be enforced by general-purpose
994 * qual-expression-evaluation code. (But we are still entitled to count
995 * their selectivity when estimating the result tuple count, if we
996 * can guess what it is...)
998 * When the referenced clause is an OR clause, we generate a modified copy
999 * in which additional RestrictInfo nodes are inserted below the top-level
1000 * OR/AND structure. This is a convenience for OR indexscan processing:
1001 * indexquals taken from either the top level or an OR subclause will have
1002 * associated RestrictInfo nodes.
1004 * The can_join flag is set true if the clause looks potentially useful as
1005 * a merge or hash join clause, that is if it is a binary opclause with
1006 * nonoverlapping sets of relids referenced in the left and right sides.
1007 * (Whether the operator is actually merge or hash joinable isn't checked,
1010 * The pseudoconstant flag is set true if the clause contains no Vars of
1011 * the current query level and no volatile functions. Such a clause can be
1012 * pulled out and used as a one-time qual in a gating Result node. We keep
1013 * pseudoconstant clauses in the same lists as other RestrictInfos so that
1014 * the regular clause-pushing machinery can assign them to the correct join
1015 * level, but they need to be treated specially for cost and selectivity
1016 * estimates. Note that a pseudoconstant clause can never be an indexqual
1017 * or merge or hash join clause, so it's of no interest to large parts of
1020 * When join clauses are generated from EquivalenceClasses, there may be
1021 * several equally valid ways to enforce join equivalence, of which we need
1022 * apply only one. We mark clauses of this kind by setting parent_ec to
1023 * point to the generating EquivalenceClass. Multiple clauses with the same
1024 * parent_ec in the same join are redundant.
1027 typedef struct RestrictInfo
1031 Expr *clause; /* the represented clause of WHERE or JOIN */
1033 bool is_pushed_down; /* TRUE if clause was pushed down in level */
1035 bool outerjoin_delayed; /* TRUE if delayed by lower outer join */
1037 bool can_join; /* see comment above */
1039 bool pseudoconstant; /* see comment above */
1041 /* The set of relids (varnos) actually referenced in the clause: */
1042 Relids clause_relids;
1044 /* The set of relids required to evaluate the clause: */
1045 Relids required_relids;
1047 /* The relids used in the clause that are nullable by lower outer joins: */
1048 Relids nullable_relids;
1050 /* These fields are set for any binary opclause: */
1051 Relids left_relids; /* relids in left side of clause */
1052 Relids right_relids; /* relids in right side of clause */
1054 /* This field is NULL unless clause is an OR clause: */
1055 Expr *orclause; /* modified clause with RestrictInfos */
1057 /* This field is NULL unless clause is potentially redundant: */
1058 EquivalenceClass *parent_ec; /* generating EquivalenceClass */
1060 /* cache space for cost and selectivity */
1061 QualCost eval_cost; /* eval cost of clause; -1 if not yet set */
1062 Selectivity norm_selec; /* selectivity for "normal" (JOIN_INNER)
1063 * semantics; -1 if not yet set; >1 means a
1064 * redundant clause */
1065 Selectivity outer_selec; /* selectivity for outer join semantics; -1 if
1068 /* valid if clause is mergejoinable, else NIL */
1069 List *mergeopfamilies; /* opfamilies containing clause operator */
1071 /* cache space for mergeclause processing; NULL if not yet set */
1072 EquivalenceClass *left_ec; /* EquivalenceClass containing lefthand */
1073 EquivalenceClass *right_ec; /* EquivalenceClass containing righthand */
1074 EquivalenceMember *left_em; /* EquivalenceMember for lefthand */
1075 EquivalenceMember *right_em; /* EquivalenceMember for righthand */
1076 List *scansel_cache; /* list of MergeScanSelCache structs */
1078 /* transient workspace for use while considering a specific join path */
1079 bool outer_is_left; /* T = outer var on left, F = on right */
1081 /* valid if clause is hashjoinable, else InvalidOid: */
1082 Oid hashjoinoperator; /* copy of clause operator */
1084 /* cache space for hashclause processing; -1 if not yet set */
1085 Selectivity left_bucketsize; /* avg bucketsize of left side */
1086 Selectivity right_bucketsize; /* avg bucketsize of right side */
1090 * Since mergejoinscansel() is a relatively expensive function, and would
1091 * otherwise be invoked many times while planning a large join tree,
1092 * we go out of our way to cache its results. Each mergejoinable
1093 * RestrictInfo carries a list of the specific sort orderings that have
1094 * been considered for use with it, and the resulting selectivities.
1096 typedef struct MergeScanSelCache
1098 /* Ordering details (cache lookup key) */
1099 Oid opfamily; /* btree opfamily defining the ordering */
1100 int strategy; /* sort direction (ASC or DESC) */
1101 bool nulls_first; /* do NULLs come before normal values? */
1103 Selectivity leftstartsel; /* first-join fraction for clause left side */
1104 Selectivity leftendsel; /* last-join fraction for clause left side */
1105 Selectivity rightstartsel; /* first-join fraction for clause right side */
1106 Selectivity rightendsel; /* last-join fraction for clause right side */
1107 } MergeScanSelCache;
1110 * Inner indexscan info.
1112 * An inner indexscan is one that uses one or more joinclauses as index
1113 * conditions (perhaps in addition to plain restriction clauses). So it
1114 * can only be used as the inner path of a nestloop join where the outer
1115 * relation includes all other relids appearing in those joinclauses.
1116 * The set of usable joinclauses, and thus the best inner indexscan,
1117 * thus varies depending on which outer relation we consider; so we have
1118 * to recompute the best such paths for every join. To avoid lots of
1119 * redundant computation, we cache the results of such searches. For
1120 * each relation we compute the set of possible otherrelids (all relids
1121 * appearing in joinquals that could become indexquals for this table).
1122 * Two outer relations whose relids have the same intersection with this
1123 * set will have the same set of available joinclauses and thus the same
1124 * best inner indexscans for the inner relation. By taking the intersection
1125 * before scanning the cache, we avoid recomputing when considering
1126 * join rels that differ only by the inclusion of irrelevant other rels.
1128 * The search key also includes a bool showing whether the join being
1129 * considered is an outer join. Since we constrain the join order for
1130 * outer joins, I believe that this bool can only have one possible value
1131 * for any particular lookup key; but store it anyway to avoid confusion.
1134 typedef struct InnerIndexscanInfo
1137 /* The lookup key: */
1138 Relids other_relids; /* a set of relevant other relids */
1139 bool isouterjoin; /* true if join is outer */
1140 /* Best paths for this lookup key (NULL if no available indexscans): */
1141 Path *cheapest_startup_innerpath; /* cheapest startup cost */
1142 Path *cheapest_total_innerpath; /* cheapest total cost */
1143 } InnerIndexscanInfo;
1146 * Placeholder node for an expression to be evaluated below the top level
1147 * of a plan tree. This is used during planning to represent the contained
1148 * expression. At the end of the planning process it is replaced by either
1149 * the contained expression or a Var referring to a lower-level evaluation of
1150 * the contained expression. Typically the evaluation occurs below an outer
1151 * join, and Var references above the outer join might thereby yield NULL
1152 * instead of the expression value.
1154 * Although the planner treats this as an expression node type, it is not
1155 * recognized by the parser or executor, so we declare it here rather than
1159 typedef struct PlaceHolderVar
1162 Expr *phexpr; /* the represented expression */
1163 Relids phrels; /* base relids syntactically within expr src */
1164 Index phid; /* ID for PHV (unique within planner run) */
1165 Index phlevelsup; /* > 0 if PHV belongs to outer query */
1169 * "Special join" info.
1171 * One-sided outer joins constrain the order of joining partially but not
1172 * completely. We flatten such joins into the planner's top-level list of
1173 * relations to join, but record information about each outer join in a
1174 * SpecialJoinInfo struct. These structs are kept in the PlannerInfo node's
1177 * Similarly, semijoins and antijoins created by flattening IN (subselect)
1178 * and EXISTS(subselect) clauses create partial constraints on join order.
1179 * These are likewise recorded in SpecialJoinInfo structs.
1181 * We make SpecialJoinInfos for FULL JOINs even though there is no flexibility
1182 * of planning for them, because this simplifies make_join_rel()'s API.
1184 * min_lefthand and min_righthand are the sets of base relids that must be
1185 * available on each side when performing the special join. lhs_strict is
1186 * true if the special join's condition cannot succeed when the LHS variables
1187 * are all NULL (this means that an outer join can commute with upper-level
1188 * outer joins even if it appears in their RHS). We don't bother to set
1189 * lhs_strict for FULL JOINs, however.
1191 * It is not valid for either min_lefthand or min_righthand to be empty sets;
1192 * if they were, this would break the logic that enforces join order.
1194 * syn_lefthand and syn_righthand are the sets of base relids that are
1195 * syntactically below this special join. (These are needed to help compute
1196 * min_lefthand and min_righthand for higher joins.)
1198 * delay_upper_joins is set TRUE if we detect a pushed-down clause that has
1199 * to be evaluated after this join is formed (because it references the RHS).
1200 * Any outer joins that have such a clause and this join in their RHS cannot
1201 * commute with this join, because that would leave noplace to check the
1202 * pushed-down clause. (We don't track this for FULL JOINs, either.)
1204 * join_quals is an implicit-AND list of the quals syntactically associated
1205 * with the join (they may or may not end up being applied at the join level).
1206 * This is just a side list and does not drive actual application of quals.
1207 * For JOIN_SEMI joins, this is cleared to NIL in create_unique_path() if
1208 * the join is found not to be suitable for a uniqueify-the-RHS plan.
1210 * jointype is never JOIN_RIGHT; a RIGHT JOIN is handled by switching
1211 * the inputs to make it a LEFT JOIN. So the allowed values of jointype
1212 * in a join_info_list member are only LEFT, FULL, SEMI, or ANTI.
1214 * For purposes of join selectivity estimation, we create transient
1215 * SpecialJoinInfo structures for regular inner joins; so it is possible
1216 * to have jointype == JOIN_INNER in such a structure, even though this is
1217 * not allowed within join_info_list. We also create transient
1218 * SpecialJoinInfos with jointype == JOIN_INNER for outer joins, since for
1219 * cost estimation purposes it is sometimes useful to know the join size under
1220 * plain innerjoin semantics. Note that lhs_strict, delay_upper_joins, and
1221 * join_quals are not set meaningfully within such structs.
1224 typedef struct SpecialJoinInfo
1227 Relids min_lefthand; /* base relids in minimum LHS for join */
1228 Relids min_righthand; /* base relids in minimum RHS for join */
1229 Relids syn_lefthand; /* base relids syntactically within LHS */
1230 Relids syn_righthand; /* base relids syntactically within RHS */
1231 JoinType jointype; /* always INNER, LEFT, FULL, SEMI, or ANTI */
1232 bool lhs_strict; /* joinclause is strict for some LHS rel */
1233 bool delay_upper_joins; /* can't commute with upper RHS */
1234 List *join_quals; /* join quals, in implicit-AND list format */
1238 * Append-relation info.
1240 * When we expand an inheritable table or a UNION-ALL subselect into an
1241 * "append relation" (essentially, a list of child RTEs), we build an
1242 * AppendRelInfo for each child RTE. The list of AppendRelInfos indicates
1243 * which child RTEs must be included when expanding the parent, and each
1244 * node carries information needed to translate Vars referencing the parent
1245 * into Vars referencing that child.
1247 * These structs are kept in the PlannerInfo node's append_rel_list.
1248 * Note that we just throw all the structs into one list, and scan the
1249 * whole list when desiring to expand any one parent. We could have used
1250 * a more complex data structure (eg, one list per parent), but this would
1251 * be harder to update during operations such as pulling up subqueries,
1252 * and not really any easier to scan. Considering that typical queries
1253 * will not have many different append parents, it doesn't seem worthwhile
1254 * to complicate things.
1256 * Note: after completion of the planner prep phase, any given RTE is an
1257 * append parent having entries in append_rel_list if and only if its
1258 * "inh" flag is set. We clear "inh" for plain tables that turn out not
1259 * to have inheritance children, and (in an abuse of the original meaning
1260 * of the flag) we set "inh" for subquery RTEs that turn out to be
1261 * flattenable UNION ALL queries. This lets us avoid useless searches
1262 * of append_rel_list.
1264 * Note: the data structure assumes that append-rel members are single
1265 * baserels. This is OK for inheritance, but it prevents us from pulling
1266 * up a UNION ALL member subquery if it contains a join. While that could
1267 * be fixed with a more complex data structure, at present there's not much
1268 * point because no improvement in the plan could result.
1271 typedef struct AppendRelInfo
1276 * These fields uniquely identify this append relationship. There can be
1277 * (in fact, always should be) multiple AppendRelInfos for the same
1278 * parent_relid, but never more than one per child_relid, since a given
1279 * RTE cannot be a child of more than one append parent.
1281 Index parent_relid; /* RT index of append parent rel */
1282 Index child_relid; /* RT index of append child rel */
1285 * For an inheritance appendrel, the parent and child are both regular
1286 * relations, and we store their rowtype OIDs here for use in translating
1287 * whole-row Vars. For a UNION-ALL appendrel, the parent and child are
1288 * both subqueries with no named rowtype, and we store InvalidOid here.
1290 Oid parent_reltype; /* OID of parent's composite type */
1291 Oid child_reltype; /* OID of child's composite type */
1294 * The N'th element of this list is a Var or expression representing the
1295 * child column corresponding to the N'th column of the parent. This is
1296 * used to translate Vars referencing the parent rel into references to
1297 * the child. A list element is NULL if it corresponds to a dropped
1298 * column of the parent (this is only possible for inheritance cases, not
1299 * UNION ALL). The list elements are always simple Vars for inheritance
1300 * cases, but can be arbitrary expressions in UNION ALL cases.
1302 * Notice we only store entries for user columns (attno > 0). Whole-row
1303 * Vars are special-cased, and system columns (attno < 0) need no special
1304 * translation since their attnos are the same for all tables.
1306 * Caution: the Vars have varlevelsup = 0. Be careful to adjust as needed
1307 * when copying into a subquery.
1309 List *translated_vars; /* Expressions in the child's Vars */
1312 * We store the parent table's OID here for inheritance, or InvalidOid for
1313 * UNION ALL. This is only needed to help in generating error messages if
1314 * an attempt is made to reference a dropped parent column.
1316 Oid parent_reloid; /* OID of parent relation */
1320 * For each distinct placeholder expression generated during planning, we
1321 * store a PlaceHolderInfo node in the PlannerInfo node's placeholder_list.
1322 * This stores info that is needed centrally rather than in each copy of the
1323 * PlaceHolderVar. The phid fields identify which PlaceHolderInfo goes with
1324 * each PlaceHolderVar. Note that phid is unique throughout a planner run,
1325 * not just within a query level --- this is so that we need not reassign ID's
1326 * when pulling a subquery into its parent.
1328 * The idea is to evaluate the expression at (only) the ph_eval_at join level,
1329 * then allow it to bubble up like a Var until the ph_needed join level.
1330 * ph_needed has the same definition as attr_needed for a regular Var.
1332 * ph_may_need is an initial estimate of ph_needed, formed using the
1333 * syntactic locations of references to the PHV. We need this in order to
1334 * determine whether the PHV reference forces a join ordering constraint:
1335 * if the PHV has to be evaluated below the nullable side of an outer join,
1336 * and then used above that outer join, we must constrain join order to ensure
1337 * there's a valid place to evaluate the PHV below the join. The final
1338 * actual ph_needed level might be lower than ph_may_need, but we can't
1339 * determine that until later on. Fortunately this doesn't matter for what
1340 * we need ph_may_need for: if there's a PHV reference syntactically
1341 * above the outer join, it's not going to be allowed to drop below the outer
1342 * join, so we would come to the same conclusions about join order even if
1343 * we had the final ph_needed value to compare to.
1345 * We create a PlaceHolderInfo only after determining that the PlaceHolderVar
1346 * is actually referenced in the plan tree, so that unreferenced placeholders
1347 * don't result in unnecessary constraints on join order.
1350 typedef struct PlaceHolderInfo
1354 Index phid; /* ID for PH (unique within planner run) */
1355 PlaceHolderVar *ph_var; /* copy of PlaceHolderVar tree */
1356 Relids ph_eval_at; /* lowest level we can evaluate value at */
1357 Relids ph_needed; /* highest level the value is needed at */
1358 Relids ph_may_need; /* highest level it might be needed at */
1359 int32 ph_width; /* estimated attribute width */
1363 * For each potentially index-optimizable MIN/MAX aggregate function,
1364 * root->minmax_aggs stores a MinMaxAggInfo describing it.
1366 * Note: a MIN/MAX agg doesn't really care about the nulls_first property,
1367 * so the pathkey's nulls_first flag should be ignored.
1369 typedef struct MinMaxAggInfo
1373 Oid aggfnoid; /* pg_proc Oid of the aggregate */
1374 Oid aggsortop; /* Oid of its sort operator */
1375 Expr *target; /* expression we are aggregating on */
1376 List *pathkeys; /* pathkeys representing needed sort order */
1380 * glob->paramlist keeps track of the PARAM_EXEC slots that we have decided
1381 * we need for the query. At runtime these slots are used to pass values
1382 * around from one plan node to another. They can be used to pass values
1383 * down into subqueries (for outer references in subqueries), or up out of
1384 * subqueries (for the results of a subplan), or from a NestLoop plan node
1385 * into its inner relation (when the inner scan is parameterized with values
1386 * from the outer relation). The n'th entry in the list (n counts from 0)
1387 * corresponds to Param->paramid = n.
1389 * Each paramlist item shows the absolute query level it is associated with,
1390 * where the outermost query is level 1 and nested subqueries have higher
1391 * numbers. The item the parameter slot represents can be one of three kinds:
1393 * A Var: the slot represents a variable of that level that must be passed
1394 * down because subqueries have outer references to it, or must be passed
1395 * from a NestLoop node of that level to its inner scan. The varlevelsup
1396 * value in the Var will always be zero.
1398 * An Aggref (with an expression tree representing its argument): the slot
1399 * represents an aggregate expression that is an outer reference for some
1400 * subquery. The Aggref itself has agglevelsup = 0, and its argument tree
1401 * is adjusted to match in level.
1403 * A Param: the slot holds the result of a subplan (it is a setParam item
1404 * for that subplan). The absolute level shown for such items corresponds
1405 * to the parent query of the subplan.
1407 * Note: we detect duplicate Var parameters and coalesce them into one slot,
1408 * but we do not bother to do this for Aggrefs, and it would be incorrect
1409 * to do so for Param slots. Duplicate detection is actually *necessary*
1410 * in the case of NestLoop parameters since it serves to match up the usage
1411 * of a Param (in the inner scan) with the assignment of the value (in the
1412 * NestLoop node). This might result in the same PARAM_EXEC slot being used
1413 * by multiple NestLoop nodes or SubPlan nodes, but no harm is done since
1414 * the same value would be assigned anyway.
1416 typedef struct PlannerParamItem
1420 Node *item; /* the Var, Aggref, or Param */
1421 Index abslevel; /* its absolute query level */
1424 #endif /* RELATION_H */