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 1.118 2005/08/27 22:13:43 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
69 * an RTE does not correspond to a base relation. Note that the array
70 * may be 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
78 * the list to do lookups, but when there are many join relations we
79 * build a hash table for faster lookups. The hash table is present
80 * and valid when join_rel_hash is not NULL. Note that we still maintain
81 * the list even when using the hash table for lookups; this simplifies
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
90 List *left_join_clauses; /* list of RestrictInfos for outer join
91 * clauses w/nonnullable var on left */
93 List *right_join_clauses; /* list of RestrictInfos for outer join
94 * clauses w/nonnullable var on right */
96 List *full_join_clauses; /* list of RestrictInfos for full outer
99 List *in_info_list; /* list of InClauseInfos */
101 List *query_pathkeys; /* desired pathkeys for query_planner(),
102 * and actual pathkeys afterwards */
104 List *group_pathkeys; /* groupClause pathkeys, if any */
105 List *sort_pathkeys; /* sortClause pathkeys, if any */
107 double tuple_fraction; /* tuple_fraction passed to query_planner */
109 bool hasJoinRTEs; /* true if any RTEs are RTE_JOIN kind */
110 bool hasOuterJoins; /* true if any RTEs are outer joins */
111 bool hasHavingQual; /* true if havingQual was non-null */
117 * Per-relation information for planning/optimization
119 * For planning purposes, a "base rel" is either a plain relation (a table)
120 * or the output of a sub-SELECT or function that appears in the range table.
121 * In either case it is uniquely identified by an RT index. A "joinrel"
122 * is the joining of two or more base rels. A joinrel is identified by
123 * the set of RT indexes for its component baserels. We create RelOptInfo
124 * nodes for each baserel and joinrel, and store them in the PlannerInfo's
125 * base_rel_array and join_rel_list respectively.
127 * Note that there is only one joinrel for any given set of component
128 * baserels, no matter what order we assemble them in; so an unordered
129 * set is the right datatype to identify it with.
131 * We also have "other rels", which are like base rels in that they refer to
132 * single RT indexes; but they are not part of the join tree, and are given
133 * a different RelOptKind to identify them.
135 * Currently the only kind of otherrels are those made for child relations
136 * of an inheritance scan (SELECT FROM foo*). The parent table's RTE and
137 * corresponding baserel represent the whole result of the inheritance scan.
138 * The planner creates separate RTEs and associated RelOptInfos for each child
139 * table (including the parent table, in its capacity as a member of the
140 * inheritance set). These RelOptInfos are physically identical to baserels,
141 * but are otherrels because they are not in the main join tree. These added
142 * RTEs and otherrels are used to plan the scans of the individual tables in
143 * the inheritance set; then the parent baserel is given an Append plan
144 * comprising the best plans for the individual child tables.
146 * At one time we also made otherrels to represent join RTEs, for use in
147 * handling join alias Vars. Currently this is not needed because all join
148 * alias Vars are expanded to non-aliased form during preprocess_expression.
150 * Parts of this data structure are specific to various scan and join
151 * mechanisms. It didn't seem worth creating new node types for them.
153 * relids - Set of base-relation identifiers; it is a base relation
154 * if there is just one, a join relation if more than one
155 * rows - estimated number of tuples in the relation after restriction
156 * clauses have been applied (ie, output rows of a plan for it)
157 * width - avg. number of bytes per tuple in the relation after the
158 * appropriate projections have been done (ie, output width)
159 * reltargetlist - List of Var nodes for the attributes we need to
160 * output from this relation (in no particular order)
161 * NOTE: in a child relation, may contain RowExprs
162 * pathlist - List of Path nodes, one for each potentially useful
163 * method of generating the relation
164 * cheapest_startup_path - the pathlist member with lowest startup cost
165 * (regardless of its ordering)
166 * cheapest_total_path - the pathlist member with lowest total cost
167 * (regardless of its ordering)
168 * cheapest_unique_path - for caching cheapest path to produce unique
169 * (no duplicates) output from relation
171 * If the relation is a base relation it will have these fields set:
173 * relid - RTE index (this is redundant with the relids field, but
174 * is provided for convenience of access)
175 * rtekind - distinguishes plain relation, subquery, or function RTE
176 * min_attr, max_attr - range of valid AttrNumbers for rel
177 * attr_needed - array of bitmapsets indicating the highest joinrel
178 * in which each attribute is needed; if bit 0 is set then
179 * the attribute is needed as part of final targetlist
180 * attr_widths - cache space for per-attribute width estimates;
181 * zero means not computed yet
182 * indexlist - list of IndexOptInfo nodes for relation's indexes
183 * (always NIL if it's not a table)
184 * pages - number of disk pages in relation (zero if not a table)
185 * tuples - number of tuples in relation (not considering restrictions)
186 * subplan - plan for subquery (NULL if it's not a subquery)
188 * Note: for a subquery, tuples and subplan are not set immediately
189 * upon creation of the RelOptInfo object; they are filled in when
190 * set_base_rel_pathlist processes the object.
192 * For otherrels that are inheritance children, these fields are filled
193 * in just as for a baserel.
195 * The presence of the remaining fields depends on the restrictions
196 * and joins that the relation participates in:
198 * baserestrictinfo - List of RestrictInfo nodes, containing info about
199 * each non-join qualification clause in which this relation
200 * participates (only used for base rels)
201 * baserestrictcost - Estimated cost of evaluating the baserestrictinfo
202 * clauses at a single tuple (only used for base rels)
203 * outerjoinset - For a base rel: if the rel appears within the nullable
204 * side of an outer join, the set of all relids
205 * participating in the highest such outer join; else NULL.
207 * joininfo - List of RestrictInfo nodes, containing info about each
208 * join clause in which this relation participates
209 * index_outer_relids - only used for base rels; set of outer relids
210 * that participate in indexable joinclauses for this rel
211 * index_inner_paths - only used for base rels; list of InnerIndexscanInfo
212 * nodes showing best indexpaths for various subsets of
213 * index_outer_relids.
215 * Note: Keeping a restrictinfo list in the RelOptInfo is useful only for
216 * base rels, because for a join rel the set of clauses that are treated as
217 * restrict clauses varies depending on which sub-relations we choose to join.
218 * (For example, in a 3-base-rel join, a clause relating rels 1 and 2 must be
219 * treated as a restrictclause if we join {1} and {2 3} to make {1 2 3}; but
220 * if we join {1 2} and {3} then that clause will be a restrictclause in {1 2}
221 * and should not be processed again at the level of {1 2 3}.) Therefore,
222 * the restrictinfo list in the join case appears in individual JoinPaths
223 * (field joinrestrictinfo), not in the parent relation. But it's OK for
224 * the RelOptInfo to store the joininfo list, because that is the same
225 * for a given rel no matter how we form it.
227 * We store baserestrictcost in the RelOptInfo (for base relations) because
228 * we know we will need it at least once (to price the sequential scan)
229 * and may need it multiple times to price index scans.
231 * outerjoinset is used to ensure correct placement of WHERE clauses that
232 * apply to outer-joined relations; we must not apply such WHERE clauses
233 * until after the outer join is performed.
236 typedef enum RelOptKind
240 RELOPT_OTHER_CHILD_REL
243 typedef struct RelOptInfo
247 RelOptKind reloptkind;
249 /* all relations included in this RelOptInfo */
250 Relids relids; /* set of base relids (rangetable indexes) */
252 /* size estimates generated by planner */
253 double rows; /* estimated number of result tuples */
254 int width; /* estimated avg width of result tuples */
256 /* materialization information */
257 List *reltargetlist; /* needed Vars */
258 List *pathlist; /* Path structures */
259 struct Path *cheapest_startup_path;
260 struct Path *cheapest_total_path;
261 struct Path *cheapest_unique_path;
263 /* information about a base rel (not set for join rels!) */
265 RTEKind rtekind; /* RELATION, SUBQUERY, or FUNCTION */
266 AttrNumber min_attr; /* smallest attrno of rel (often <0) */
267 AttrNumber max_attr; /* largest attrno of rel */
268 Relids *attr_needed; /* array indexed [min_attr .. max_attr] */
269 int32 *attr_widths; /* array indexed [min_attr .. max_attr] */
273 struct Plan *subplan; /* if subquery */
275 /* used by various scans and joins: */
276 List *baserestrictinfo; /* RestrictInfo structures (if
278 QualCost baserestrictcost; /* cost of evaluating the above */
279 Relids outerjoinset; /* set of base relids */
280 List *joininfo; /* RestrictInfo structures for join clauses
281 * involving this rel */
283 /* cached info about inner indexscan paths for relation: */
284 Relids index_outer_relids; /* other relids in indexable join
286 List *index_inner_paths; /* InnerIndexscanInfo nodes */
289 * Inner indexscans are not in the main pathlist because they are not
290 * usable except in specific join contexts. We use the
291 * index_inner_paths list just to avoid recomputing the best inner
292 * indexscan repeatedly for similar outer relations. See comments for
293 * InnerIndexscanInfo.
299 * Per-index information for planning/optimization
301 * Prior to Postgres 7.0, RelOptInfo was used to describe both relations
302 * and indexes, but that created confusion without actually doing anything
303 * useful. So now we have a separate IndexOptInfo struct for indexes.
305 * classlist[], indexkeys[], and ordering[] have ncolumns entries.
306 * Zeroes in the indexkeys[] array indicate index columns that are
307 * expressions; there is one element in indexprs for each such column.
309 * Note: for historical reasons, the classlist and ordering arrays have
310 * an extra entry that is always zero. Some code scans until it sees a
311 * zero entry, rather than looking at ncolumns.
313 * The indexprs and indpred expressions have been run through
314 * prepqual.c and eval_const_expressions() for ease of matching to
315 * WHERE clauses. indpred is in implicit-AND form.
318 typedef struct IndexOptInfo
322 Oid indexoid; /* OID of the index relation */
323 RelOptInfo *rel; /* back-link to index's table */
325 /* statistics from pg_class */
326 BlockNumber pages; /* number of disk pages in index */
327 double tuples; /* number of index tuples in index */
329 /* index descriptor information */
330 int ncolumns; /* number of columns in index */
331 Oid *classlist; /* OIDs of operator classes for columns */
332 int *indexkeys; /* column numbers of index's keys, or 0 */
333 Oid *ordering; /* OIDs of sort operators for each column */
334 Oid relam; /* OID of the access method (in pg_am) */
336 RegProcedure amcostestimate; /* OID of the access method's cost fcn */
338 List *indexprs; /* expressions for non-simple index
340 List *indpred; /* predicate if a partial index, else NIL */
342 bool predOK; /* true if predicate matches query */
343 bool unique; /* true if a unique index */
344 bool amoptionalkey; /* can query omit key for the first column? */
351 * The sort ordering of a path is represented by a list of sublists of
352 * PathKeyItem nodes. An empty list implies no known ordering. Otherwise
353 * the first sublist represents the primary sort key, the second the
354 * first secondary sort key, etc. Each sublist contains one or more
355 * PathKeyItem nodes, each of which can be taken as the attribute that
356 * appears at that sort position. (See optimizer/README for more
360 typedef struct PathKeyItem
364 Node *key; /* the item that is ordered */
365 Oid sortop; /* the ordering operator ('<' op) */
368 * key typically points to a Var node, ie a relation attribute, but it
369 * can also point to an arbitrary expression representing the value
370 * indexed by an index expression.
375 * Type "Path" is used as-is for sequential-scan paths. For other
376 * path types it is the first component of a larger struct.
378 * Note: "pathtype" is the NodeTag of the Plan node we could build from this
379 * Path. It is partially redundant with the Path's NodeTag, but allows us
380 * to use the same Path type for multiple Plan types where there is no need
381 * to distinguish the Plan type during path processing.
388 NodeTag pathtype; /* tag identifying scan/join method */
390 RelOptInfo *parent; /* the relation this path can build */
392 /* estimated execution costs for path (see costsize.c for more info) */
393 Cost startup_cost; /* cost expended before fetching any
395 Cost total_cost; /* total cost (assuming all tuples
398 List *pathkeys; /* sort ordering of path's output */
399 /* pathkeys is a List of Lists of PathKeyItem nodes; see above */
403 * IndexPath represents an index scan over a single index.
405 * 'indexinfo' is the index to be scanned.
407 * 'indexclauses' is a list of index qualification clauses, with implicit
408 * AND semantics across the list. Each clause is a RestrictInfo node from
409 * the query's WHERE or JOIN conditions.
411 * 'indexquals' has the same structure as 'indexclauses', but it contains
412 * the actual indexqual conditions that can be used with the index.
413 * In simple cases this is identical to 'indexclauses', but when special
414 * indexable operators appear in 'indexclauses', they are replaced by the
415 * derived indexscannable conditions in 'indexquals'.
417 * 'isjoininner' is TRUE if the path is a nestloop inner scan (that is,
418 * some of the index conditions are join rather than restriction clauses).
420 * 'indexscandir' is one of:
421 * ForwardScanDirection: forward scan of an ordered index
422 * BackwardScanDirection: backward scan of an ordered index
423 * NoMovementScanDirection: scan of an unordered index, or don't care
424 * (The executor doesn't care whether it gets ForwardScanDirection or
425 * NoMovementScanDirection for an indexscan, but the planner wants to
426 * distinguish ordered from unordered indexes for building pathkeys.)
428 * 'indextotalcost' and 'indexselectivity' are saved in the IndexPath so that
429 * we need not recompute them when considering using the same index in a
430 * bitmap index/heap scan (see BitmapHeapPath). The costs of the IndexPath
431 * itself represent the costs of an IndexScan plan type.
433 * 'rows' is the estimated result tuple count for the indexscan. This
434 * is the same as path.parent->rows for a simple indexscan, but it is
435 * different for a nestloop inner scan, because the additional indexquals
436 * coming from join clauses make the scan more selective than the parent
437 * rel's restrict clauses alone would do.
440 typedef struct IndexPath
443 IndexOptInfo *indexinfo;
447 ScanDirection indexscandir;
449 Selectivity indexselectivity;
450 double rows; /* estimated number of result tuples */
454 * BitmapHeapPath represents one or more indexscans that generate TID bitmaps
455 * instead of directly accessing the heap, followed by AND/OR combinations
456 * to produce a single bitmap, followed by a heap scan that uses the bitmap.
457 * Note that the output is always considered unordered, since it will come
458 * out in physical heap order no matter what the underlying indexes did.
460 * The individual indexscans are represented by IndexPath nodes, and any
461 * logic on top of them is represented by a tree of BitmapAndPath and
462 * BitmapOrPath nodes. Notice that we can use the same IndexPath node both
463 * to represent a regular IndexScan plan, and as the child of a BitmapHeapPath
464 * that represents scanning the same index using a BitmapIndexScan. The
465 * startup_cost and total_cost figures of an IndexPath always represent the
466 * costs to use it as a regular IndexScan. The costs of a BitmapIndexScan
467 * can be computed using the IndexPath's indextotalcost and indexselectivity.
469 * BitmapHeapPaths can be nestloop inner indexscans. The isjoininner and
470 * rows fields serve the same purpose as for plain IndexPaths.
472 typedef struct BitmapHeapPath
475 Path *bitmapqual; /* IndexPath, BitmapAndPath, BitmapOrPath */
476 bool isjoininner; /* T if it's a nestloop inner scan */
477 double rows; /* estimated number of result tuples */
481 * BitmapAndPath represents a BitmapAnd plan node; it can only appear as
482 * part of the substructure of a BitmapHeapPath. The Path structure is
483 * a bit more heavyweight than we really need for this, but for simplicity
484 * we make it a derivative of Path anyway.
486 typedef struct BitmapAndPath
489 List *bitmapquals; /* IndexPaths and BitmapOrPaths */
490 Selectivity bitmapselectivity;
494 * BitmapOrPath represents a BitmapOr plan node; it can only appear as
495 * part of the substructure of a BitmapHeapPath. The Path structure is
496 * a bit more heavyweight than we really need for this, but for simplicity
497 * we make it a derivative of Path anyway.
499 typedef struct BitmapOrPath
502 List *bitmapquals; /* IndexPaths and BitmapAndPaths */
503 Selectivity bitmapselectivity;
507 * TidPath represents a scan by TID
509 * tideval is an implicitly OR'ed list of quals of the form CTID = something.
510 * Note they are bare quals, not RestrictInfos.
512 typedef struct TidPath
515 List *tideval; /* qual(s) involving CTID = something */
519 * AppendPath represents an Append plan, ie, successive execution of
520 * several member plans. Currently it is only used to handle expansion
521 * of inheritance trees.
523 * Note: it is possible for "subpaths" to contain only one, or even no,
524 * elements. These cases are optimized during create_append_plan.
526 typedef struct AppendPath
529 List *subpaths; /* list of component Paths */
533 * ResultPath represents use of a Result plan node. There are several
534 * applications for this:
535 * * To compute a variable-free targetlist (a "SELECT expressions" query).
536 * In this case subpath and path.parent will both be NULL. constantqual
537 * might or might not be empty ("SELECT expressions WHERE something").
538 * * To gate execution of a subplan with a one-time (variable-free) qual
539 * condition. path.parent is copied from the subpath.
540 * * To substitute for a scan plan when we have proven that no rows in
541 * a table will satisfy the query. subpath is NULL but path.parent
542 * references the not-to-be-scanned relation, and constantqual is
545 * Note that constantqual is a list of bare clauses, not RestrictInfos.
547 typedef struct ResultPath
555 * MaterialPath represents use of a Material plan node, i.e., caching of
556 * the output of its subpath. This is used when the subpath is expensive
557 * and needs to be scanned repeatedly, or when we need mark/restore ability
558 * and the subpath doesn't have it.
560 typedef struct MaterialPath
567 * UniquePath represents elimination of distinct rows from the output of
570 * This is unlike the other Path nodes in that it can actually generate
571 * different plans: either hash-based or sort-based implementation, or a
572 * no-op if the input path can be proven distinct already. The decision
573 * is sufficiently localized that it's not worth having separate Path node
574 * types. (Note: in the no-op case, we could eliminate the UniquePath node
575 * entirely and just return the subpath; but it's convenient to have a
576 * UniquePath in the path tree to signal upper-level routines that the input
577 * is known distinct.)
581 UNIQUE_PATH_NOOP, /* input is known unique already */
582 UNIQUE_PATH_HASH, /* use hashing */
583 UNIQUE_PATH_SORT /* use sorting */
586 typedef struct UniquePath
590 UniquePathMethod umethod;
591 double rows; /* estimated number of result tuples */
595 * All join-type paths share these fields.
598 typedef struct JoinPath
604 Path *outerjoinpath; /* path for the outer side of the join */
605 Path *innerjoinpath; /* path for the inner side of the join */
607 List *joinrestrictinfo; /* RestrictInfos to apply to join */
610 * See the notes for RelOptInfo to understand why joinrestrictinfo is
611 * needed in JoinPath, and can't be merged into the parent RelOptInfo.
616 * A nested-loop path needs no special fields.
619 typedef JoinPath NestPath;
622 * A mergejoin path has these fields.
624 * path_mergeclauses lists the clauses (in the form of RestrictInfos)
625 * that will be used in the merge.
627 * Note that the mergeclauses are a subset of the parent relation's
628 * restriction-clause list. Any join clauses that are not mergejoinable
629 * appear only in the parent's restrict list, and must be checked by a
630 * qpqual at execution time.
632 * outersortkeys (resp. innersortkeys) is NIL if the outer path
633 * (resp. inner path) is already ordered appropriately for the
634 * mergejoin. If it is not NIL then it is a PathKeys list describing
635 * the ordering that must be created by an explicit sort step.
638 typedef struct MergePath
641 List *path_mergeclauses; /* join clauses to be used for
643 List *outersortkeys; /* keys for explicit sort, if any */
644 List *innersortkeys; /* keys for explicit sort, if any */
648 * A hashjoin path has these fields.
650 * The remarks above for mergeclauses apply for hashclauses as well.
652 * Hashjoin does not care what order its inputs appear in, so we have
653 * no need for sortkeys.
656 typedef struct HashPath
659 List *path_hashclauses; /* join clauses used for hashing */
663 * Restriction clause info.
665 * We create one of these for each AND sub-clause of a restriction condition
666 * (WHERE or JOIN/ON clause). Since the restriction clauses are logically
667 * ANDed, we can use any one of them or any subset of them to filter out
668 * tuples, without having to evaluate the rest. The RestrictInfo node itself
669 * stores data used by the optimizer while choosing the best query plan.
671 * If a restriction clause references a single base relation, it will appear
672 * in the baserestrictinfo list of the RelOptInfo for that base rel.
674 * If a restriction clause references more than one base rel, it will
675 * appear in the joininfo list of every RelOptInfo that describes a strict
676 * subset of the base rels mentioned in the clause. The joininfo lists are
677 * used to drive join tree building by selecting plausible join candidates.
678 * The clause cannot actually be applied until we have built a join rel
679 * containing all the base rels it references, however.
681 * When we construct a join rel that includes all the base rels referenced
682 * in a multi-relation restriction clause, we place that clause into the
683 * joinrestrictinfo lists of paths for the join rel, if neither left nor
684 * right sub-path includes all base rels referenced in the clause. The clause
685 * will be applied at that join level, and will not propagate any further up
686 * the join tree. (Note: the "predicate migration" code was once intended to
687 * push restriction clauses up and down the plan tree based on evaluation
688 * costs, but it's dead code and is unlikely to be resurrected in the
689 * foreseeable future.)
691 * Note that in the presence of more than two rels, a multi-rel restriction
692 * might reach different heights in the join tree depending on the join
693 * sequence we use. So, these clauses cannot be associated directly with
694 * the join RelOptInfo, but must be kept track of on a per-join-path basis.
696 * When dealing with outer joins we have to be very careful about pushing qual
697 * clauses up and down the tree. An outer join's own JOIN/ON conditions must
698 * be evaluated exactly at that join node, and any quals appearing in WHERE or
699 * in a JOIN above the outer join cannot be pushed down below the outer join.
700 * Otherwise the outer join will produce wrong results because it will see the
701 * wrong sets of input rows. All quals are stored as RestrictInfo nodes
702 * during planning, but there's a flag to indicate whether a qual has been
703 * pushed down to a lower level than its original syntactic placement in the
704 * join tree would suggest. If an outer join prevents us from pushing a qual
705 * down to its "natural" semantic level (the level associated with just the
706 * base rels used in the qual) then we mark the qual with a "required_relids"
707 * value including more than just the base rels it actually uses. By
708 * pretending that the qual references all the rels appearing in the outer
709 * join, we prevent it from being evaluated below the outer join's joinrel.
710 * When we do form the outer join's joinrel, we still need to distinguish
711 * those quals that are actually in that join's JOIN/ON condition from those
712 * that appeared higher in the tree and were pushed down to the join rel
713 * because they used no other rels. That's what the is_pushed_down flag is
714 * for; it tells us that a qual came from a point above the join of the
715 * set of base rels listed in required_relids. A clause that originally came
716 * from WHERE will *always* have its is_pushed_down flag set; a clause that
717 * came from an INNER JOIN condition, but doesn't use all the rels being
718 * joined, will also have is_pushed_down set because it will get attached to
719 * some lower joinrel.
721 * In general, the referenced clause might be arbitrarily complex. The
722 * kinds of clauses we can handle as indexscan quals, mergejoin clauses,
723 * or hashjoin clauses are fairly limited --- the code for each kind of
724 * path is responsible for identifying the restrict clauses it can use
725 * and ignoring the rest. Clauses not implemented by an indexscan,
726 * mergejoin, or hashjoin will be placed in the plan qual or joinqual field
727 * of the finished Plan node, where they will be enforced by general-purpose
728 * qual-expression-evaluation code. (But we are still entitled to count
729 * their selectivity when estimating the result tuple count, if we
730 * can guess what it is...)
732 * When the referenced clause is an OR clause, we generate a modified copy
733 * in which additional RestrictInfo nodes are inserted below the top-level
734 * OR/AND structure. This is a convenience for OR indexscan processing:
735 * indexquals taken from either the top level or an OR subclause will have
736 * associated RestrictInfo nodes.
739 typedef struct RestrictInfo
743 Expr *clause; /* the represented clause of WHERE or JOIN */
745 bool is_pushed_down; /* TRUE if clause was pushed down in level */
748 * This flag is set true if the clause looks potentially useful as a
749 * merge or hash join clause, that is if it is a binary opclause with
750 * nonoverlapping sets of relids referenced in the left and right
751 * sides. (Whether the operator is actually merge or hash joinable
752 * isn't checked, however.)
756 /* The set of relids (varnos) actually referenced in the clause: */
757 Relids clause_relids;
759 /* The set of relids required to evaluate the clause: */
760 Relids required_relids;
762 /* These fields are set for any binary opclause: */
763 Relids left_relids; /* relids in left side of clause */
764 Relids right_relids; /* relids in right side of clause */
766 /* This field is NULL unless clause is an OR clause: */
767 Expr *orclause; /* modified clause with RestrictInfos */
769 /* cache space for cost and selectivity */
770 QualCost eval_cost; /* eval cost of clause; -1 if not yet set */
771 Selectivity this_selec; /* selectivity; -1 if not yet set */
773 /* valid if clause is mergejoinable, else InvalidOid: */
774 Oid mergejoinoperator; /* copy of clause operator */
775 Oid left_sortop; /* leftside sortop needed for mergejoin */
776 Oid right_sortop; /* rightside sortop needed for mergejoin */
778 /* cache space for mergeclause processing; NIL if not yet set */
779 List *left_pathkey; /* canonical pathkey for left side */
780 List *right_pathkey; /* canonical pathkey for right side */
782 /* cache space for mergeclause processing; -1 if not yet set */
783 Selectivity left_mergescansel; /* fraction of left side to scan */
784 Selectivity right_mergescansel; /* fraction of right side to scan */
786 /* valid if clause is hashjoinable, else InvalidOid: */
787 Oid hashjoinoperator; /* copy of clause operator */
789 /* cache space for hashclause processing; -1 if not yet set */
790 Selectivity left_bucketsize; /* avg bucketsize of left side */
791 Selectivity right_bucketsize; /* avg bucketsize of right side */
795 * Inner indexscan info.
797 * An inner indexscan is one that uses one or more joinclauses as index
798 * conditions (perhaps in addition to plain restriction clauses). So it
799 * can only be used as the inner path of a nestloop join where the outer
800 * relation includes all other relids appearing in those joinclauses.
801 * The set of usable joinclauses, and thus the best inner indexscan,
802 * thus varies depending on which outer relation we consider; so we have
803 * to recompute the best such path for every join. To avoid lots of
804 * redundant computation, we cache the results of such searches. For
805 * each relation we compute the set of possible otherrelids (all relids
806 * appearing in joinquals that could become indexquals for this table).
807 * Two outer relations whose relids have the same intersection with this
808 * set will have the same set of available joinclauses and thus the same
809 * best inner indexscan for the inner relation. By taking the intersection
810 * before scanning the cache, we avoid recomputing when considering
811 * join rels that differ only by the inclusion of irrelevant other rels.
813 * The search key also includes a bool showing whether the join being
814 * considered is an outer join. Since we constrain the join order for
815 * outer joins, I believe that this bool can only have one possible value
816 * for any particular base relation; but store it anyway to avoid confusion.
819 typedef struct InnerIndexscanInfo
822 /* The lookup key: */
823 Relids other_relids; /* a set of relevant other relids */
824 bool isouterjoin; /* true if join is outer */
825 /* Best path for this lookup key: */
826 Path *best_innerpath; /* best inner indexscan, or NULL if none */
827 } InnerIndexscanInfo;
832 * When we convert top-level IN quals into join operations, we must restrict
833 * the order of joining and use special join methods at some join points.
834 * We record information about each such IN clause in an InClauseInfo struct.
835 * These structs are kept in the PlannerInfo node's in_info_list.
838 typedef struct InClauseInfo
841 Relids lefthand; /* base relids in lefthand expressions */
842 Relids righthand; /* base relids coming from the subselect */
843 List *sub_targetlist; /* targetlist of original RHS subquery */
846 * Note: sub_targetlist is just a list of Vars or expressions; it does
847 * not contain TargetEntry nodes.
851 #endif /* RELATION_H */