1 /*-------------------------------------------------------------------------
4 * Definitions for planner's internal data structures.
7 * Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
8 * Portions Copyright (c) 1994, Regents of the University of California
10 * $PostgreSQL: pgsql/src/include/nodes/relation.h,v 1.122 2005/12/20 02:30:36 tgl Exp $
12 *-------------------------------------------------------------------------
17 #include "access/sdir.h"
18 #include "nodes/bitmapset.h"
19 #include "nodes/parsenodes.h"
20 #include "storage/block.h"
25 * Set of relation identifiers (indexes into the rangetable).
27 typedef Bitmapset *Relids;
30 * When looking for a "cheapest path", this enum specifies whether we want
31 * cheapest startup cost or cheapest total cost.
33 typedef enum CostSelector
35 STARTUP_COST, TOTAL_COST
39 * The cost estimate produced by cost_qual_eval() includes both a one-time
40 * (startup) cost, and a per-tuple cost.
42 typedef struct QualCost
44 Cost startup; /* one-time cost */
45 Cost per_tuple; /* per-evaluation cost */
51 * Per-query information for planning/optimization
53 * This struct is conventionally called "root" in all the planner routines.
54 * It holds links to all of the planner's working state, in addition to the
55 * original Query. Note that at present the planner extensively manipulates
56 * the passed-in Query data structure; someday that should stop.
59 typedef struct PlannerInfo
63 Query *parse; /* the Query being planned */
66 * base_rel_array holds pointers to "base rels" and "other rels" (see
67 * comments for RelOptInfo for more info). It is indexed by rangetable
68 * index (so entry 0 is always wasted). Entries can be NULL when an RTE
69 * does not correspond to a base relation. Note that the array may be
70 * enlarged on-the-fly.
72 struct RelOptInfo **base_rel_array; /* All one-relation RelOptInfos */
73 int base_rel_array_size; /* current allocated array len */
76 * join_rel_list is a list of all join-relation RelOptInfos we have
77 * considered in this planning run. For small problems we just scan the
78 * list to do lookups, but when there are many join relations we build a
79 * hash table for faster lookups. The hash table is present and valid
80 * when join_rel_hash is not NULL. Note that we still maintain the list
81 * even when using the hash table for lookups; this simplifies life for
84 List *join_rel_list; /* list of join-relation RelOptInfos */
85 struct HTAB *join_rel_hash; /* optional hashtable for join relations */
87 List *equi_key_list; /* list of lists of equijoined PathKeyItems */
89 List *left_join_clauses; /* list of RestrictInfos for outer
90 * join clauses w/nonnullable var on
93 List *right_join_clauses; /* list of RestrictInfos for outer
94 * join clauses w/nonnullable var on
97 List *full_join_clauses; /* list of RestrictInfos for full
98 * outer join clauses */
100 List *oj_info_list; /* list of OuterJoinInfos */
102 List *in_info_list; /* list of InClauseInfos */
104 List *query_pathkeys; /* desired pathkeys for query_planner(), and
105 * actual pathkeys afterwards */
107 List *group_pathkeys; /* groupClause pathkeys, if any */
108 List *sort_pathkeys; /* sortClause pathkeys, if any */
110 double tuple_fraction; /* tuple_fraction passed to query_planner */
112 bool hasJoinRTEs; /* true if any RTEs are RTE_JOIN kind */
113 bool hasOuterJoins; /* true if any RTEs are outer joins */
114 bool hasHavingQual; /* true if havingQual was non-null */
120 * Per-relation information for planning/optimization
122 * For planning purposes, a "base rel" is either a plain relation (a table)
123 * or the output of a sub-SELECT or function that appears in the range table.
124 * In either case it is uniquely identified by an RT index. A "joinrel"
125 * is the joining of two or more base rels. A joinrel is identified by
126 * the set of RT indexes for its component baserels. We create RelOptInfo
127 * nodes for each baserel and joinrel, and store them in the PlannerInfo's
128 * base_rel_array and join_rel_list respectively.
130 * Note that there is only one joinrel for any given set of component
131 * baserels, no matter what order we assemble them in; so an unordered
132 * set is the right datatype to identify it with.
134 * We also have "other rels", which are like base rels in that they refer to
135 * single RT indexes; but they are not part of the join tree, and are given
136 * a different RelOptKind to identify them.
138 * Currently the only kind of otherrels are those made for child relations
139 * of an inheritance scan (SELECT FROM foo*). The parent table's RTE and
140 * corresponding baserel represent the whole result of the inheritance scan.
141 * The planner creates separate RTEs and associated RelOptInfos for each child
142 * table (including the parent table, in its capacity as a member of the
143 * inheritance set). These RelOptInfos are physically identical to baserels,
144 * but are otherrels because they are not in the main join tree. These added
145 * RTEs and otherrels are used to plan the scans of the individual tables in
146 * the inheritance set; then the parent baserel is given an Append plan
147 * comprising the best plans for the individual child tables.
149 * At one time we also made otherrels to represent join RTEs, for use in
150 * handling join alias Vars. Currently this is not needed because all join
151 * alias Vars are expanded to non-aliased form during preprocess_expression.
153 * Parts of this data structure are specific to various scan and join
154 * mechanisms. It didn't seem worth creating new node types for them.
156 * relids - Set of base-relation identifiers; it is a base relation
157 * if there is just one, a join relation if more than one
158 * rows - estimated number of tuples in the relation after restriction
159 * clauses have been applied (ie, output rows of a plan for it)
160 * width - avg. number of bytes per tuple in the relation after the
161 * appropriate projections have been done (ie, output width)
162 * reltargetlist - List of Var nodes for the attributes we need to
163 * output from this relation (in no particular order)
164 * NOTE: in a child relation, may contain RowExprs
165 * pathlist - List of Path nodes, one for each potentially useful
166 * method of generating the relation
167 * cheapest_startup_path - the pathlist member with lowest startup cost
168 * (regardless of its ordering)
169 * cheapest_total_path - the pathlist member with lowest total cost
170 * (regardless of its ordering)
171 * cheapest_unique_path - for caching cheapest path to produce unique
172 * (no duplicates) output from relation
174 * If the relation is a base relation it will have these fields set:
176 * relid - RTE index (this is redundant with the relids field, but
177 * is provided for convenience of access)
178 * rtekind - distinguishes plain relation, subquery, or function RTE
179 * min_attr, max_attr - range of valid AttrNumbers for rel
180 * attr_needed - array of bitmapsets indicating the highest joinrel
181 * in which each attribute is needed; if bit 0 is set then
182 * the attribute is needed as part of final targetlist
183 * attr_widths - cache space for per-attribute width estimates;
184 * zero means not computed yet
185 * indexlist - list of IndexOptInfo nodes for relation's indexes
186 * (always NIL if it's not a table)
187 * pages - number of disk pages in relation (zero if not a table)
188 * tuples - number of tuples in relation (not considering restrictions)
189 * subplan - plan for subquery (NULL if it's not a subquery)
191 * Note: for a subquery, tuples and subplan are not set immediately
192 * upon creation of the RelOptInfo object; they are filled in when
193 * set_base_rel_pathlist processes the object.
195 * For otherrels that are inheritance children, these fields are filled
196 * in just as for a baserel.
198 * The presence of the remaining fields depends on the restrictions
199 * and joins that the relation participates in:
201 * baserestrictinfo - List of RestrictInfo nodes, containing info about
202 * each non-join qualification clause in which this relation
203 * participates (only used for base rels)
204 * baserestrictcost - Estimated cost of evaluating the baserestrictinfo
205 * clauses at a single tuple (only used for base rels)
206 * joininfo - List of RestrictInfo nodes, containing info about each
207 * join clause in which this relation participates
208 * index_outer_relids - only used for base rels; set of outer relids
209 * that participate in indexable joinclauses for this rel
210 * index_inner_paths - only used for base rels; list of InnerIndexscanInfo
211 * nodes showing best indexpaths for various subsets of
212 * index_outer_relids.
214 * Note: Keeping a restrictinfo list in the RelOptInfo is useful only for
215 * base rels, because for a join rel the set of clauses that are treated as
216 * restrict clauses varies depending on which sub-relations we choose to join.
217 * (For example, in a 3-base-rel join, a clause relating rels 1 and 2 must be
218 * treated as a restrictclause if we join {1} and {2 3} to make {1 2 3}; but
219 * if we join {1 2} and {3} then that clause will be a restrictclause in {1 2}
220 * and should not be processed again at the level of {1 2 3}.) Therefore,
221 * the restrictinfo list in the join case appears in individual JoinPaths
222 * (field joinrestrictinfo), not in the parent relation. But it's OK for
223 * the RelOptInfo to store the joininfo list, because that is the same
224 * for a given rel no matter how we form it.
226 * We store baserestrictcost in the RelOptInfo (for base relations) because
227 * we know we will need it at least once (to price the sequential scan)
228 * and may need it multiple times to price index scans.
231 typedef enum RelOptKind
235 RELOPT_OTHER_CHILD_REL
238 typedef struct RelOptInfo
242 RelOptKind reloptkind;
244 /* all relations included in this RelOptInfo */
245 Relids relids; /* set of base relids (rangetable indexes) */
247 /* size estimates generated by planner */
248 double rows; /* estimated number of result tuples */
249 int width; /* estimated avg width of result tuples */
251 /* materialization information */
252 List *reltargetlist; /* needed Vars */
253 List *pathlist; /* Path structures */
254 struct Path *cheapest_startup_path;
255 struct Path *cheapest_total_path;
256 struct Path *cheapest_unique_path;
258 /* information about a base rel (not set for join rels!) */
260 RTEKind rtekind; /* RELATION, SUBQUERY, or FUNCTION */
261 AttrNumber min_attr; /* smallest attrno of rel (often <0) */
262 AttrNumber max_attr; /* largest attrno of rel */
263 Relids *attr_needed; /* array indexed [min_attr .. max_attr] */
264 int32 *attr_widths; /* array indexed [min_attr .. max_attr] */
268 struct Plan *subplan; /* if subquery */
270 /* used by various scans and joins: */
271 List *baserestrictinfo; /* RestrictInfo structures (if base
273 QualCost baserestrictcost; /* cost of evaluating the above */
274 List *joininfo; /* RestrictInfo structures for join clauses
275 * involving this rel */
277 /* cached info about inner indexscan paths for relation: */
278 Relids index_outer_relids; /* other relids in indexable join
280 List *index_inner_paths; /* InnerIndexscanInfo nodes */
283 * Inner indexscans are not in the main pathlist because they are not
284 * usable except in specific join contexts. We use the index_inner_paths
285 * list just to avoid recomputing the best inner indexscan repeatedly for
286 * similar outer relations. See comments for InnerIndexscanInfo.
292 * Per-index information for planning/optimization
294 * Prior to Postgres 7.0, RelOptInfo was used to describe both relations
295 * and indexes, but that created confusion without actually doing anything
296 * useful. So now we have a separate IndexOptInfo struct for indexes.
298 * classlist[], indexkeys[], and ordering[] have ncolumns entries.
299 * Zeroes in the indexkeys[] array indicate index columns that are
300 * expressions; there is one element in indexprs for each such column.
302 * Note: for historical reasons, the classlist and ordering arrays have
303 * an extra entry that is always zero. Some code scans until it sees a
304 * zero entry, rather than looking at ncolumns.
306 * The indexprs and indpred expressions have been run through
307 * prepqual.c and eval_const_expressions() for ease of matching to
308 * WHERE clauses. indpred is in implicit-AND form.
311 typedef struct IndexOptInfo
315 Oid indexoid; /* OID of the index relation */
316 RelOptInfo *rel; /* back-link to index's table */
318 /* statistics from pg_class */
319 BlockNumber pages; /* number of disk pages in index */
320 double tuples; /* number of index tuples in index */
322 /* index descriptor information */
323 int ncolumns; /* number of columns in index */
324 Oid *classlist; /* OIDs of operator classes for columns */
325 int *indexkeys; /* column numbers of index's keys, or 0 */
326 Oid *ordering; /* OIDs of sort operators for each column */
327 Oid relam; /* OID of the access method (in pg_am) */
329 RegProcedure amcostestimate; /* OID of the access method's cost fcn */
331 List *indexprs; /* expressions for non-simple index columns */
332 List *indpred; /* predicate if a partial index, else NIL */
334 bool predOK; /* true if predicate matches query */
335 bool unique; /* true if a unique index */
336 bool amoptionalkey; /* can query omit key for the first column? */
343 * The sort ordering of a path is represented by a list of sublists of
344 * PathKeyItem nodes. An empty list implies no known ordering. Otherwise
345 * the first sublist represents the primary sort key, the second the
346 * first secondary sort key, etc. Each sublist contains one or more
347 * PathKeyItem nodes, each of which can be taken as the attribute that
348 * appears at that sort position. (See optimizer/README for more
352 typedef struct PathKeyItem
356 Node *key; /* the item that is ordered */
357 Oid sortop; /* the ordering operator ('<' op) */
360 * key typically points to a Var node, ie a relation attribute, but it can
361 * also point to an arbitrary expression representing the value indexed by
362 * an index expression.
367 * Type "Path" is used as-is for sequential-scan paths. For other
368 * path types it is the first component of a larger struct.
370 * Note: "pathtype" is the NodeTag of the Plan node we could build from this
371 * Path. It is partially redundant with the Path's NodeTag, but allows us
372 * to use the same Path type for multiple Plan types where there is no need
373 * to distinguish the Plan type during path processing.
380 NodeTag pathtype; /* tag identifying scan/join method */
382 RelOptInfo *parent; /* the relation this path can build */
384 /* estimated execution costs for path (see costsize.c for more info) */
385 Cost startup_cost; /* cost expended before fetching any tuples */
386 Cost total_cost; /* total cost (assuming all tuples fetched) */
388 List *pathkeys; /* sort ordering of path's output */
389 /* pathkeys is a List of Lists of PathKeyItem nodes; see above */
393 * IndexPath represents an index scan over a single index.
395 * 'indexinfo' is the index to be scanned.
397 * 'indexclauses' is a list of index qualification clauses, with implicit
398 * AND semantics across the list. Each clause is a RestrictInfo node from
399 * the query's WHERE or JOIN conditions.
401 * 'indexquals' has the same structure as 'indexclauses', but it contains
402 * the actual indexqual conditions that can be used with the index.
403 * In simple cases this is identical to 'indexclauses', but when special
404 * indexable operators appear in 'indexclauses', they are replaced by the
405 * derived indexscannable conditions in 'indexquals'.
407 * 'isjoininner' is TRUE if the path is a nestloop inner scan (that is,
408 * some of the index conditions are join rather than restriction clauses).
410 * 'indexscandir' is one of:
411 * ForwardScanDirection: forward scan of an ordered index
412 * BackwardScanDirection: backward scan of an ordered index
413 * NoMovementScanDirection: scan of an unordered index, or don't care
414 * (The executor doesn't care whether it gets ForwardScanDirection or
415 * NoMovementScanDirection for an indexscan, but the planner wants to
416 * distinguish ordered from unordered indexes for building pathkeys.)
418 * 'indextotalcost' and 'indexselectivity' are saved in the IndexPath so that
419 * we need not recompute them when considering using the same index in a
420 * bitmap index/heap scan (see BitmapHeapPath). The costs of the IndexPath
421 * itself represent the costs of an IndexScan plan type.
423 * 'rows' is the estimated result tuple count for the indexscan. This
424 * is the same as path.parent->rows for a simple indexscan, but it is
425 * different for a nestloop inner scan, because the additional indexquals
426 * coming from join clauses make the scan more selective than the parent
427 * rel's restrict clauses alone would do.
430 typedef struct IndexPath
433 IndexOptInfo *indexinfo;
437 ScanDirection indexscandir;
439 Selectivity indexselectivity;
440 double rows; /* estimated number of result tuples */
444 * BitmapHeapPath represents one or more indexscans that generate TID bitmaps
445 * instead of directly accessing the heap, followed by AND/OR combinations
446 * to produce a single bitmap, followed by a heap scan that uses the bitmap.
447 * Note that the output is always considered unordered, since it will come
448 * out in physical heap order no matter what the underlying indexes did.
450 * The individual indexscans are represented by IndexPath nodes, and any
451 * logic on top of them is represented by a tree of BitmapAndPath and
452 * BitmapOrPath nodes. Notice that we can use the same IndexPath node both
453 * to represent a regular IndexScan plan, and as the child of a BitmapHeapPath
454 * that represents scanning the same index using a BitmapIndexScan. The
455 * startup_cost and total_cost figures of an IndexPath always represent the
456 * costs to use it as a regular IndexScan. The costs of a BitmapIndexScan
457 * can be computed using the IndexPath's indextotalcost and indexselectivity.
459 * BitmapHeapPaths can be nestloop inner indexscans. The isjoininner and
460 * rows fields serve the same purpose as for plain IndexPaths.
462 typedef struct BitmapHeapPath
465 Path *bitmapqual; /* IndexPath, BitmapAndPath, BitmapOrPath */
466 bool isjoininner; /* T if it's a nestloop inner scan */
467 double rows; /* estimated number of result tuples */
471 * BitmapAndPath represents a BitmapAnd plan node; it can only appear as
472 * part of the substructure of a BitmapHeapPath. The Path structure is
473 * a bit more heavyweight than we really need for this, but for simplicity
474 * we make it a derivative of Path anyway.
476 typedef struct BitmapAndPath
479 List *bitmapquals; /* IndexPaths and BitmapOrPaths */
480 Selectivity bitmapselectivity;
484 * BitmapOrPath represents a BitmapOr plan node; it can only appear as
485 * part of the substructure of a BitmapHeapPath. The Path structure is
486 * a bit more heavyweight than we really need for this, but for simplicity
487 * we make it a derivative of Path anyway.
489 typedef struct BitmapOrPath
492 List *bitmapquals; /* IndexPaths and BitmapAndPaths */
493 Selectivity bitmapselectivity;
497 * TidPath represents a scan by TID
499 * tidquals is an implicitly OR'ed list of qual expressions of the form
500 * "CTID = pseudoconstant" or "CTID = ANY(pseudoconstant_array)".
501 * Note they are bare expressions, not RestrictInfos.
503 typedef struct TidPath
506 List *tidquals; /* qual(s) involving CTID = something */
510 * AppendPath represents an Append plan, ie, successive execution of
511 * several member plans. Currently it is only used to handle expansion
512 * of inheritance trees.
514 * Note: it is possible for "subpaths" to contain only one, or even no,
515 * elements. These cases are optimized during create_append_plan.
517 typedef struct AppendPath
520 List *subpaths; /* list of component Paths */
524 * ResultPath represents use of a Result plan node. There are several
525 * applications for this:
526 * * To compute a variable-free targetlist (a "SELECT expressions" query).
527 * In this case subpath and path.parent will both be NULL. constantqual
528 * might or might not be empty ("SELECT expressions WHERE something").
529 * * To gate execution of a subplan with a one-time (variable-free) qual
530 * condition. path.parent is copied from the subpath.
531 * * To substitute for a scan plan when we have proven that no rows in
532 * a table will satisfy the query. subpath is NULL but path.parent
533 * references the not-to-be-scanned relation, and constantqual is
536 * Note that constantqual is a list of bare clauses, not RestrictInfos.
538 typedef struct ResultPath
546 * MaterialPath represents use of a Material plan node, i.e., caching of
547 * the output of its subpath. This is used when the subpath is expensive
548 * and needs to be scanned repeatedly, or when we need mark/restore ability
549 * and the subpath doesn't have it.
551 typedef struct MaterialPath
558 * UniquePath represents elimination of distinct rows from the output of
561 * This is unlike the other Path nodes in that it can actually generate
562 * different plans: either hash-based or sort-based implementation, or a
563 * no-op if the input path can be proven distinct already. The decision
564 * is sufficiently localized that it's not worth having separate Path node
565 * types. (Note: in the no-op case, we could eliminate the UniquePath node
566 * entirely and just return the subpath; but it's convenient to have a
567 * UniquePath in the path tree to signal upper-level routines that the input
568 * is known distinct.)
572 UNIQUE_PATH_NOOP, /* input is known unique already */
573 UNIQUE_PATH_HASH, /* use hashing */
574 UNIQUE_PATH_SORT /* use sorting */
577 typedef struct UniquePath
581 UniquePathMethod umethod;
582 double rows; /* estimated number of result tuples */
586 * All join-type paths share these fields.
589 typedef struct JoinPath
595 Path *outerjoinpath; /* path for the outer side of the join */
596 Path *innerjoinpath; /* path for the inner side of the join */
598 List *joinrestrictinfo; /* RestrictInfos to apply to join */
601 * See the notes for RelOptInfo to understand why joinrestrictinfo is
602 * needed in JoinPath, and can't be merged into the parent RelOptInfo.
607 * A nested-loop path needs no special fields.
610 typedef JoinPath NestPath;
613 * A mergejoin path has these fields.
615 * path_mergeclauses lists the clauses (in the form of RestrictInfos)
616 * that will be used in the merge.
618 * Note that the mergeclauses are a subset of the parent relation's
619 * restriction-clause list. Any join clauses that are not mergejoinable
620 * appear only in the parent's restrict list, and must be checked by a
621 * qpqual at execution time.
623 * outersortkeys (resp. innersortkeys) is NIL if the outer path
624 * (resp. inner path) is already ordered appropriately for the
625 * mergejoin. If it is not NIL then it is a PathKeys list describing
626 * the ordering that must be created by an explicit sort step.
629 typedef struct MergePath
632 List *path_mergeclauses; /* join clauses to be used for merge */
633 List *outersortkeys; /* keys for explicit sort, if any */
634 List *innersortkeys; /* keys for explicit sort, if any */
638 * A hashjoin path has these fields.
640 * The remarks above for mergeclauses apply for hashclauses as well.
642 * Hashjoin does not care what order its inputs appear in, so we have
643 * no need for sortkeys.
646 typedef struct HashPath
649 List *path_hashclauses; /* join clauses used for hashing */
653 * Restriction clause info.
655 * We create one of these for each AND sub-clause of a restriction condition
656 * (WHERE or JOIN/ON clause). Since the restriction clauses are logically
657 * ANDed, we can use any one of them or any subset of them to filter out
658 * tuples, without having to evaluate the rest. The RestrictInfo node itself
659 * stores data used by the optimizer while choosing the best query plan.
661 * If a restriction clause references a single base relation, it will appear
662 * in the baserestrictinfo list of the RelOptInfo for that base rel.
664 * If a restriction clause references more than one base rel, it will
665 * appear in the joininfo list of every RelOptInfo that describes a strict
666 * subset of the base rels mentioned in the clause. The joininfo lists are
667 * used to drive join tree building by selecting plausible join candidates.
668 * The clause cannot actually be applied until we have built a join rel
669 * containing all the base rels it references, however.
671 * When we construct a join rel that includes all the base rels referenced
672 * in a multi-relation restriction clause, we place that clause into the
673 * joinrestrictinfo lists of paths for the join rel, if neither left nor
674 * right sub-path includes all base rels referenced in the clause. The clause
675 * will be applied at that join level, and will not propagate any further up
676 * the join tree. (Note: the "predicate migration" code was once intended to
677 * push restriction clauses up and down the plan tree based on evaluation
678 * costs, but it's dead code and is unlikely to be resurrected in the
679 * foreseeable future.)
681 * Note that in the presence of more than two rels, a multi-rel restriction
682 * might reach different heights in the join tree depending on the join
683 * sequence we use. So, these clauses cannot be associated directly with
684 * the join RelOptInfo, but must be kept track of on a per-join-path basis.
686 * When dealing with outer joins we have to be very careful about pushing qual
687 * clauses up and down the tree. An outer join's own JOIN/ON conditions must
688 * be evaluated exactly at that join node, and any quals appearing in WHERE or
689 * in a JOIN above the outer join cannot be pushed down below the outer join.
690 * Otherwise the outer join will produce wrong results because it will see the
691 * wrong sets of input rows. All quals are stored as RestrictInfo nodes
692 * during planning, but there's a flag to indicate whether a qual has been
693 * pushed down to a lower level than its original syntactic placement in the
694 * join tree would suggest. If an outer join prevents us from pushing a qual
695 * down to its "natural" semantic level (the level associated with just the
696 * base rels used in the qual) then we mark the qual with a "required_relids"
697 * value including more than just the base rels it actually uses. By
698 * pretending that the qual references all the rels appearing in the outer
699 * join, we prevent it from being evaluated below the outer join's joinrel.
700 * When we do form the outer join's joinrel, we still need to distinguish
701 * those quals that are actually in that join's JOIN/ON condition from those
702 * that appeared higher in the tree and were pushed down to the join rel
703 * because they used no other rels. That's what the is_pushed_down flag is
704 * for; it tells us that a qual came from a point above the join of the
705 * set of base rels listed in required_relids. A clause that originally came
706 * from WHERE will *always* have its is_pushed_down flag set; a clause that
707 * came from an INNER JOIN condition, but doesn't use all the rels being
708 * joined, will also have is_pushed_down set because it will get attached to
709 * some lower joinrel.
711 * When application of a qual must be delayed by outer join, we also mark it
712 * with outerjoin_delayed = true. This isn't redundant with required_relids
713 * because that might equal clause_relids whether or not it's an outer-join
716 * In general, the referenced clause might be arbitrarily complex. The
717 * kinds of clauses we can handle as indexscan quals, mergejoin clauses,
718 * or hashjoin clauses are fairly limited --- the code for each kind of
719 * path is responsible for identifying the restrict clauses it can use
720 * and ignoring the rest. Clauses not implemented by an indexscan,
721 * mergejoin, or hashjoin will be placed in the plan qual or joinqual field
722 * of the finished Plan node, where they will be enforced by general-purpose
723 * qual-expression-evaluation code. (But we are still entitled to count
724 * their selectivity when estimating the result tuple count, if we
725 * can guess what it is...)
727 * When the referenced clause is an OR clause, we generate a modified copy
728 * in which additional RestrictInfo nodes are inserted below the top-level
729 * OR/AND structure. This is a convenience for OR indexscan processing:
730 * indexquals taken from either the top level or an OR subclause will have
731 * associated RestrictInfo nodes.
734 typedef struct RestrictInfo
738 Expr *clause; /* the represented clause of WHERE or JOIN */
740 bool is_pushed_down; /* TRUE if clause was pushed down in level */
742 bool outerjoin_delayed; /* TRUE if delayed by outer join */
745 * This flag is set true if the clause looks potentially useful as a merge
746 * or hash join clause, that is if it is a binary opclause with
747 * nonoverlapping sets of relids referenced in the left and right sides.
748 * (Whether the operator is actually merge or hash joinable isn't checked,
753 /* The set of relids (varnos) actually referenced in the clause: */
754 Relids clause_relids;
756 /* The set of relids required to evaluate the clause: */
757 Relids required_relids;
759 /* These fields are set for any binary opclause: */
760 Relids left_relids; /* relids in left side of clause */
761 Relids right_relids; /* relids in right side of clause */
763 /* This field is NULL unless clause is an OR clause: */
764 Expr *orclause; /* modified clause with RestrictInfos */
766 /* cache space for cost and selectivity */
767 QualCost eval_cost; /* eval cost of clause; -1 if not yet set */
768 Selectivity this_selec; /* selectivity; -1 if not yet set */
770 /* valid if clause is mergejoinable, else InvalidOid: */
771 Oid mergejoinoperator; /* copy of clause operator */
772 Oid left_sortop; /* leftside sortop needed for mergejoin */
773 Oid right_sortop; /* rightside sortop needed for mergejoin */
775 /* cache space for mergeclause processing; NIL if not yet set */
776 List *left_pathkey; /* canonical pathkey for left side */
777 List *right_pathkey; /* canonical pathkey for right side */
779 /* cache space for mergeclause processing; -1 if not yet set */
780 Selectivity left_mergescansel; /* fraction of left side to scan */
781 Selectivity right_mergescansel; /* fraction of right side to scan */
783 /* valid if clause is hashjoinable, else InvalidOid: */
784 Oid hashjoinoperator; /* copy of clause operator */
786 /* cache space for hashclause processing; -1 if not yet set */
787 Selectivity left_bucketsize; /* avg bucketsize of left side */
788 Selectivity right_bucketsize; /* avg bucketsize of right side */
792 * Inner indexscan info.
794 * An inner indexscan is one that uses one or more joinclauses as index
795 * conditions (perhaps in addition to plain restriction clauses). So it
796 * can only be used as the inner path of a nestloop join where the outer
797 * relation includes all other relids appearing in those joinclauses.
798 * The set of usable joinclauses, and thus the best inner indexscan,
799 * thus varies depending on which outer relation we consider; so we have
800 * to recompute the best such path for every join. To avoid lots of
801 * redundant computation, we cache the results of such searches. For
802 * each relation we compute the set of possible otherrelids (all relids
803 * appearing in joinquals that could become indexquals for this table).
804 * Two outer relations whose relids have the same intersection with this
805 * set will have the same set of available joinclauses and thus the same
806 * best inner indexscan for the inner relation. By taking the intersection
807 * before scanning the cache, we avoid recomputing when considering
808 * join rels that differ only by the inclusion of irrelevant other rels.
810 * The search key also includes a bool showing whether the join being
811 * considered is an outer join. Since we constrain the join order for
812 * outer joins, I believe that this bool can only have one possible value
813 * for any particular base relation; but store it anyway to avoid confusion.
816 typedef struct InnerIndexscanInfo
819 /* The lookup key: */
820 Relids other_relids; /* a set of relevant other relids */
821 bool isouterjoin; /* true if join is outer */
822 /* Best path for this lookup key: */
823 Path *best_innerpath; /* best inner indexscan, or NULL if none */
824 } InnerIndexscanInfo;
829 * One-sided outer joins constrain the order of joining partially but not
830 * completely. We flatten such joins into the planner's top-level list of
831 * relations to join, but record information about each outer join in an
832 * OuterJoinInfo struct. These structs are kept in the PlannerInfo node's
835 * min_lefthand and min_righthand are the sets of base relids that must be
836 * available on each side when performing the outer join. lhs_strict is
837 * true if the outer join's condition cannot succeed when the LHS variables
838 * are all NULL (this means that the outer join can commute with upper-level
839 * outer joins even if it appears in their RHS). We don't bother to set
840 * lhs_strict for FULL JOINs, however.
842 * It is not valid for either min_lefthand or min_righthand to be empty sets;
843 * if they were, this would break the logic that enforces join order.
845 * Note: OuterJoinInfo directly represents only LEFT JOIN and FULL JOIN;
846 * RIGHT JOIN is handled by switching the inputs to make it a LEFT JOIN.
847 * We make an OuterJoinInfo for FULL JOINs even though there is no flexibility
848 * of planning for them, because this simplifies make_join_rel()'s API.
851 typedef struct OuterJoinInfo
854 Relids min_lefthand; /* base relids in minimum LHS for join */
855 Relids min_righthand; /* base relids in minimum RHS for join */
856 bool is_full_join; /* it's a FULL OUTER JOIN */
857 bool lhs_strict; /* joinclause is strict for some LHS rel */
863 * When we convert top-level IN quals into join operations, we must restrict
864 * the order of joining and use special join methods at some join points.
865 * We record information about each such IN clause in an InClauseInfo struct.
866 * These structs are kept in the PlannerInfo node's in_info_list.
869 typedef struct InClauseInfo
872 Relids lefthand; /* base relids in lefthand expressions */
873 Relids righthand; /* base relids coming from the subselect */
874 List *sub_targetlist; /* targetlist of original RHS subquery */
877 * Note: sub_targetlist is just a list of Vars or expressions; it does not
878 * contain TargetEntry nodes.
882 #endif /* RELATION_H */