* initsplan.c
* Target list, qualification, joininfo initialization routines
*
- * Copyright (c) 1994, Regents of the University of California
+ * Portions Copyright (c) 1996-2005, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
*
*
* IDENTIFICATION
- * $Header: /cvsroot/pgsql/src/backend/optimizer/plan/initsplan.c,v 1.32 1999/07/15 15:19:22 momjian Exp $
+ * $PostgreSQL: pgsql/src/backend/optimizer/plan/initsplan.c,v 1.106 2005/06/05 22:32:55 tgl Exp $
*
*-------------------------------------------------------------------------
*/
-#include <sys/types.h>
-
#include "postgres.h"
-#include "nodes/pg_list.h"
-#include "nodes/plannodes.h"
-#include "nodes/parsenodes.h"
-#include "nodes/relation.h"
-#include "nodes/makefuncs.h"
-
-#include "access/htup.h"
-
+#include "catalog/pg_operator.h"
#include "catalog/pg_type.h"
-
-#include "utils/lsyscache.h"
-
-#include "optimizer/internal.h"
-#include "optimizer/planmain.h"
+#include "nodes/makefuncs.h"
+#include "optimizer/clauses.h"
+#include "optimizer/cost.h"
#include "optimizer/joininfo.h"
#include "optimizer/pathnode.h"
+#include "optimizer/paths.h"
+#include "optimizer/planmain.h"
+#include "optimizer/restrictinfo.h"
#include "optimizer/tlist.h"
#include "optimizer/var.h"
-#include "optimizer/clauses.h"
-#include "optimizer/cost.h"
-
-extern int Quiet;
+#include "parser/parsetree.h"
+#include "parser/parse_expr.h"
+#include "parser/parse_oper.h"
+#include "utils/builtins.h"
+#include "utils/lsyscache.h"
+#include "utils/syscache.h"
-static void add_restrict_and_join_to_rel(Query *root, List *clause);
-static void add_join_info_to_rels(Query *root, RestrictInfo *restrictinfo,
- Relids join_relids);
-static void add_vars_to_targetlist(Query *root, List *vars, Relids join_relids);
-static MergeOrder *mergejoinop(Expr *clause);
-static Oid hashjoinop(Expr *clause);
+static void mark_baserels_for_outer_join(PlannerInfo *root, Relids rels,
+ Relids outerrels);
+static void distribute_qual_to_rels(PlannerInfo *root, Node *clause,
+ bool is_pushed_down,
+ bool isdeduced,
+ Relids outerjoin_nonnullable,
+ Relids qualscope);
+static void add_vars_to_targetlist(PlannerInfo *root, List *vars,
+ Relids where_needed);
+static bool qual_is_redundant(PlannerInfo *root, RestrictInfo *restrictinfo,
+ List *restrictlist);
+static void check_mergejoinable(RestrictInfo *restrictinfo);
+static void check_hashjoinable(RestrictInfo *restrictinfo);
/*****************************************************************************
*
- * TARGET LISTS
+ * JOIN TREES
*
*****************************************************************************/
/*
- * make_var_only_tlist
- * Creates rel nodes for every relation mentioned in the target list
- * 'tlist' (if a node hasn't already been created) and adds them to
- * *query_relation_list*. Creates targetlist entries for each member of
- * 'tlist' and adds them to the tlist field of the appropriate rel node.
+ * add_base_rels_to_query
+ *
+ * Scan the query's jointree and create baserel RelOptInfos for all
+ * the base relations (ie, table, subquery, and function RTEs)
+ * appearing in the jointree.
+ *
+ * At the end of this process, there should be one baserel RelOptInfo for
+ * every non-join RTE that is used in the query. Therefore, this routine
+ * is the only place that should call build_base_rel. But build_other_rel
+ * will be used later to build rels for inheritance children.
*/
void
-make_var_only_tlist(Query *root, List *tlist)
+add_base_rels_to_query(PlannerInfo *root, Node *jtnode)
{
- List *tlist_vars = NIL;
- List *l = NIL;
- List *tvar = NIL;
-
- foreach(l, tlist)
+ if (jtnode == NULL)
+ return;
+ if (IsA(jtnode, RangeTblRef))
{
- TargetEntry *entry = (TargetEntry *) lfirst(l);
+ int varno = ((RangeTblRef *) jtnode)->rtindex;
- tlist_vars = nconc(tlist_vars, pull_var_clause(entry->expr));
+ build_base_rel(root, varno);
}
-
- /* now, the target list only contains Var nodes */
- foreach(tvar, tlist_vars)
+ else if (IsA(jtnode, FromExpr))
{
- Var *var = (Var *) lfirst(tvar);
- Index varno;
- RelOptInfo *result;
+ FromExpr *f = (FromExpr *) jtnode;
+ ListCell *l;
- varno = var->varno;
- result = get_base_rel(root, varno);
+ foreach(l, f->fromlist)
+ add_base_rels_to_query(root, lfirst(l));
+ }
+ else if (IsA(jtnode, JoinExpr))
+ {
+ JoinExpr *j = (JoinExpr *) jtnode;
- add_var_to_tlist(result, var);
+ add_base_rels_to_query(root, j->larg);
+ add_base_rels_to_query(root, j->rarg);
}
+ else
+ elog(ERROR, "unrecognized node type: %d",
+ (int) nodeTag(jtnode));
}
+
+/*****************************************************************************
+ *
+ * TARGET LISTS
+ *
+ *****************************************************************************/
+
/*
- * add_missing_vars_to_tlist
- * If we have range variable(s) in the FROM clause that does not appear
- * in the target list nor qualifications, we add it to the base relation
- * list. For instance, "select f.x from foo f, foo f2" is a join of f and
- * f2. Note that if we have "select foo.x from foo f", it also gets turned
- * into a join.
+ * build_base_rel_tlists
+ * Add targetlist entries for each var needed in the query's final tlist
+ * to the appropriate base relations.
+ *
+ * We mark such vars as needed by "relation 0" to ensure that they will
+ * propagate up through all join plan steps.
*/
void
-add_missing_vars_to_tlist(Query *root, List *tlist)
+build_base_rel_tlists(PlannerInfo *root, List *final_tlist)
{
- List *l;
- int varno;
+ List *tlist_vars = pull_var_clause((Node *) final_tlist, false);
- varno = 1;
- foreach(l, root->rtable)
+ if (tlist_vars != NIL)
{
- RangeTblEntry *rte = (RangeTblEntry *) lfirst(l);
- Relids relids;
- RelOptInfo *result;
- Var *var;
+ add_vars_to_targetlist(root, tlist_vars, bms_make_singleton(0));
+ list_free(tlist_vars);
+ }
+}
+
+/*
+ * add_vars_to_targetlist
+ * For each variable appearing in the list, add it to the owning
+ * relation's targetlist if not already present, and mark the variable
+ * as being needed for the indicated join (or for final output if
+ * where_needed includes "relation 0").
+ */
+static void
+add_vars_to_targetlist(PlannerInfo *root, List *vars, Relids where_needed)
+{
+ ListCell *temp;
+
+ Assert(!bms_is_empty(where_needed));
- relids = lconsi(varno, NIL);
- if (rte->inFromCl && !rel_member(relids, root->base_rel_list))
+ foreach(temp, vars)
+ {
+ Var *var = (Var *) lfirst(temp);
+ RelOptInfo *rel = find_base_rel(root, var->varno);
+ int attrno = var->varattno;
+
+ Assert(attrno >= rel->min_attr && attrno <= rel->max_attr);
+ attrno -= rel->min_attr;
+ if (bms_is_empty(rel->attr_needed[attrno]))
{
- var = makeVar(varno, ObjectIdAttributeNumber,
- OIDOID, -1, 0, varno, ObjectIdAttributeNumber);
- /* add it to base_rel_list */
- result = get_base_rel(root, varno);
- add_var_to_tlist(result, var);
+ /* Variable not yet requested, so add to reltargetlist */
+ /* XXX is copyObject necessary here? */
+ rel->reltargetlist = lappend(rel->reltargetlist, copyObject(var));
}
- pfree(relids);
- varno++;
+ rel->attr_needed[attrno] = bms_add_members(rel->attr_needed[attrno],
+ where_needed);
}
-
- return;
}
+
/*****************************************************************************
*
* QUALIFICATIONS
*****************************************************************************/
-
/*
- * add_restrict_and_join_to_rels-
- * Initializes RestrictInfo and JoinInfo fields of relation entries for all
- * relations appearing within clauses. Creates new relation entries if
- * necessary, adding them to *query_relation_list*.
+ * distribute_quals_to_rels
+ * Recursively scan the query's join tree for WHERE and JOIN/ON qual
+ * clauses, and add these to the appropriate RestrictInfo and JoinInfo
+ * lists belonging to base RelOptInfos. Also, base RelOptInfos are marked
+ * with outerjoinset information, to aid in proper positioning of qual
+ * clauses that appear above outer joins.
*
- * Returns nothing of interest.
+ * NOTE: when dealing with inner joins, it is appropriate to let a qual clause
+ * be evaluated at the lowest level where all the variables it mentions are
+ * available. However, we cannot push a qual down into the nullable side(s)
+ * of an outer join since the qual might eliminate matching rows and cause a
+ * NULL row to be incorrectly emitted by the join. Therefore, rels appearing
+ * within the nullable side(s) of an outer join are marked with
+ * outerjoinset = set of Relids used at the outer join node.
+ * This set will be added to the set of rels referenced by quals using such
+ * a rel, thereby forcing them up the join tree to the right level.
+ *
+ * To ease the calculation of these values, distribute_quals_to_rels() returns
+ * the set of base Relids involved in its own level of join. This is just an
+ * internal convenience; no outside callers pay attention to the result.
*/
-void
-add_restrict_and_join_to_rels(Query *root, List *clauses)
+Relids
+distribute_quals_to_rels(PlannerInfo *root, Node *jtnode)
+{
+ Relids result = NULL;
+
+ if (jtnode == NULL)
+ return result;
+ if (IsA(jtnode, RangeTblRef))
+ {
+ int varno = ((RangeTblRef *) jtnode)->rtindex;
+
+ /* No quals to deal with, just return correct result */
+ result = bms_make_singleton(varno);
+ }
+ else if (IsA(jtnode, FromExpr))
+ {
+ FromExpr *f = (FromExpr *) jtnode;
+ ListCell *l;
+
+ /*
+ * First, recurse to handle child joins.
+ */
+ foreach(l, f->fromlist)
+ {
+ result = bms_add_members(result,
+ distribute_quals_to_rels(root,
+ lfirst(l)));
+ }
+
+ /*
+ * Now process the top-level quals. These are always marked as
+ * "pushed down", since they clearly didn't come from a JOIN expr.
+ */
+ foreach(l, (List *) f->quals)
+ distribute_qual_to_rels(root, (Node *) lfirst(l),
+ true, false, NULL, result);
+ }
+ else if (IsA(jtnode, JoinExpr))
+ {
+ JoinExpr *j = (JoinExpr *) jtnode;
+ Relids leftids,
+ rightids,
+ nonnullable_rels,
+ nullable_rels;
+ ListCell *qual;
+
+ /*
+ * Order of operations here is subtle and critical. First we
+ * recurse to handle sub-JOINs. Their join quals will be placed
+ * without regard for whether this level is an outer join, which
+ * is correct. Then we place our own join quals, which are
+ * restricted by lower outer joins in any case, and are forced to
+ * this level if this is an outer join and they mention the outer
+ * side. Finally, if this is an outer join, we mark baserels
+ * contained within the inner side(s) with our own rel set; this
+ * will prevent quals above us in the join tree that use those
+ * rels from being pushed down below this level. (It's okay for
+ * upper quals to be pushed down to the outer side, however.)
+ */
+ leftids = distribute_quals_to_rels(root, j->larg);
+ rightids = distribute_quals_to_rels(root, j->rarg);
+
+ result = bms_union(leftids, rightids);
+
+ nonnullable_rels = nullable_rels = NULL;
+ switch (j->jointype)
+ {
+ case JOIN_INNER:
+ /* Inner join adds no restrictions for quals */
+ break;
+ case JOIN_LEFT:
+ nonnullable_rels = leftids;
+ nullable_rels = rightids;
+ break;
+ case JOIN_FULL:
+ /* each side is both outer and inner */
+ nonnullable_rels = result;
+ nullable_rels = result;
+ break;
+ case JOIN_RIGHT:
+ nonnullable_rels = rightids;
+ nullable_rels = leftids;
+ break;
+ case JOIN_UNION:
+
+ /*
+ * This is where we fail if upper levels of planner
+ * haven't rewritten UNION JOIN as an Append ...
+ */
+ ereport(ERROR,
+ (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+ errmsg("UNION JOIN is not implemented")));
+ break;
+ default:
+ elog(ERROR, "unrecognized join type: %d",
+ (int) j->jointype);
+ break;
+ }
+
+ foreach(qual, (List *) j->quals)
+ distribute_qual_to_rels(root, (Node *) lfirst(qual),
+ false, false,
+ nonnullable_rels, result);
+
+ if (nullable_rels != NULL)
+ mark_baserels_for_outer_join(root, nullable_rels, result);
+ }
+ else
+ elog(ERROR, "unrecognized node type: %d",
+ (int) nodeTag(jtnode));
+ return result;
+}
+
+/*
+ * mark_baserels_for_outer_join
+ * Mark all base rels listed in 'rels' as having the given outerjoinset.
+ */
+static void
+mark_baserels_for_outer_join(PlannerInfo *root, Relids rels, Relids outerrels)
{
- List *clause;
+ Relids tmprelids;
+ int relno;
- foreach(clause, clauses)
- add_restrict_and_join_to_rel(root, lfirst(clause));
- return;
+ tmprelids = bms_copy(rels);
+ while ((relno = bms_first_member(tmprelids)) >= 0)
+ {
+ RelOptInfo *rel = find_base_rel(root, relno);
+
+ /*
+ * Since we do this bottom-up, any outer-rels previously marked
+ * should be within the new outer join set.
+ */
+ Assert(bms_is_subset(rel->outerjoinset, outerrels));
+
+ /*
+ * Presently the executor cannot support FOR UPDATE/SHARE marking of
+ * rels appearing on the nullable side of an outer join. (It's
+ * somewhat unclear what that would mean, anyway: what should we
+ * mark when a result row is generated from no element of the
+ * nullable relation?) So, complain if target rel is FOR UPDATE/SHARE.
+ * It's sufficient to make this check once per rel, so do it only
+ * if rel wasn't already known nullable.
+ */
+ if (rel->outerjoinset == NULL)
+ {
+ if (list_member_int(root->parse->rowMarks, relno))
+ ereport(ERROR,
+ (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+ errmsg("SELECT FOR UPDATE/SHARE cannot be applied to the nullable side of an outer join")));
+ }
+
+ rel->outerjoinset = outerrels;
+ }
+ bms_free(tmprelids);
}
/*
- * add_restrict_and_join_to_rel-
+ * distribute_qual_to_rels
* Add clause information to either the 'RestrictInfo' or 'JoinInfo' field
- * of a relation entry(depending on whether or not the clause is a join)
- * by creating a new RestrictInfo node and setting appropriate fields
- * within the nodes.
+ * (depending on whether the clause is a join) of each base relation
+ * mentioned in the clause. A RestrictInfo node is created and added to
+ * the appropriate list for each rel. Also, if the clause uses a
+ * mergejoinable operator and is not delayed by outer-join rules, enter
+ * the left- and right-side expressions into the query's lists of
+ * equijoined vars.
*
- * Returns nothing of interest.
+ * 'clause': the qual clause to be distributed
+ * 'is_pushed_down': if TRUE, force the clause to be marked 'is_pushed_down'
+ * (this indicates the clause came from a FromExpr, not a JoinExpr)
+ * 'isdeduced': TRUE if the qual came from implied-equality deduction
+ * 'outerjoin_nonnullable': NULL if not an outer-join qual, else the set of
+ * baserels appearing on the outer (nonnullable) side of the join
+ * 'qualscope': set of baserels the qual's syntactic scope covers
+ *
+ * 'qualscope' identifies what level of JOIN the qual came from. For a top
+ * level qual (WHERE qual), qualscope lists all baserel ids and in addition
+ * 'is_pushed_down' will be TRUE.
*/
static void
-add_restrict_and_join_to_rel(Query *root, List *clause)
+distribute_qual_to_rels(PlannerInfo *root, Node *clause,
+ bool is_pushed_down,
+ bool isdeduced,
+ Relids outerjoin_nonnullable,
+ Relids qualscope)
{
Relids relids;
+ bool valid_everywhere;
+ bool can_be_equijoin;
+ RestrictInfo *restrictinfo;
+ RelOptInfo *rel;
List *vars;
- RestrictInfo *restrictinfo = makeNode(RestrictInfo);
/*
- * Retrieve all relids and vars contained within the clause.
+ * Retrieve all relids mentioned within the clause.
*/
- clause_get_relids_vars((Node *) clause, &relids, &vars);
+ relids = pull_varnos(clause);
- restrictinfo->clause = (Expr *) clause;
- restrictinfo->notclause = contains_not((Node *) clause);
- restrictinfo->selectivity = 0;
- restrictinfo->indexids = NIL;
- restrictinfo->mergejoinorder = (MergeOrder *) NULL;
- restrictinfo->hashjoinoperator = (Oid) 0;
+ /*
+ * Cross-check: clause should contain no relids not within its scope.
+ * Otherwise the parser messed up.
+ */
+ if (!bms_is_subset(relids, qualscope))
+ elog(ERROR, "JOIN qualification may not refer to other relations");
- if (length(relids) == 1)
- {
+ /*
+ * If the clause is variable-free, we force it to be evaluated at its
+ * original syntactic level. Note that this should not happen for
+ * top-level clauses, because query_planner() special-cases them. But
+ * it will happen for variable-free JOIN/ON clauses. We don't have to
+ * be real smart about such a case, we just have to be correct.
+ */
+ if (bms_is_empty(relids))
+ relids = qualscope;
+ /*
+ * Check to see if clause application must be delayed by outer-join
+ * considerations.
+ */
+ if (isdeduced)
+ {
/*
- * There is only one relation participating in 'clause', so
- * 'clause' must be a restriction clause.
+ * If the qual came from implied-equality deduction, we can
+ * evaluate the qual at its natural semantic level. It is not
+ * affected by any outer-join rules (else we'd not have decided
+ * the vars were equal).
*/
- RelOptInfo *rel = get_base_rel(root, lfirsti(relids));
-
+ Assert(bms_equal(relids, qualscope));
+ valid_everywhere = true;
+ can_be_equijoin = true;
+ }
+ else if (bms_overlap(relids, outerjoin_nonnullable))
+ {
/*
- * The selectivity of the clause must be computed regardless of
- * whether it's a restriction or a join clause
+ * The qual is attached to an outer join and mentions (some of
+ * the) rels on the nonnullable side. Force the qual to be
+ * evaluated exactly at the level of joining corresponding to the
+ * outer join. We cannot let it get pushed down into the
+ * nonnullable side, since then we'd produce no output rows,
+ * rather than the intended single null-extended row, for any
+ * nonnullable-side rows failing the qual.
+ *
+ * Note: an outer-join qual that mentions only nullable-side rels can
+ * be pushed down into the nullable side without changing the join
+ * result, so we treat it the same as an ordinary inner-join qual.
*/
- if (is_funcclause((Node *) clause))
+ relids = qualscope;
+ valid_everywhere = false;
+ can_be_equijoin = false;
+ }
+ else
+ {
+ /*
+ * For a non-outer-join qual, we can evaluate the qual as soon as
+ * (1) we have all the rels it mentions, and (2) we are at or
+ * above any outer joins that can null any of these rels and are
+ * below the syntactic location of the given qual. To enforce the
+ * latter, scan the base rels listed in relids, and merge their
+ * outer-join sets into the clause's own reference list. At the
+ * time we are called, the outerjoinset of each baserel will show
+ * exactly those outer joins that are below the qual in the join
+ * tree.
+ *
+ * We also need to determine whether the qual is "valid everywhere",
+ * which is true if the qual mentions no variables that are
+ * involved in lower-level outer joins (this may be an overly
+ * strong test).
+ */
+ Relids addrelids = NULL;
+ Relids tmprelids;
+ int relno;
- /*
- * XXX If we have a func clause set selectivity to 1/3, really
- * need a true selectivity function.
- */
- restrictinfo->selectivity = (Cost) 0.3333333;
+ valid_everywhere = true;
+ tmprelids = bms_copy(relids);
+ while ((relno = bms_first_member(tmprelids)) >= 0)
+ {
+ RelOptInfo *rel = find_base_rel(root, relno);
+
+ if (rel->outerjoinset != NULL)
+ {
+ addrelids = bms_add_members(addrelids, rel->outerjoinset);
+ valid_everywhere = false;
+ }
+ }
+ bms_free(tmprelids);
+
+ if (bms_is_subset(addrelids, relids))
+ {
+ /* Qual is not affected by any outer-join restriction */
+ can_be_equijoin = true;
+ }
else
- restrictinfo->selectivity = compute_clause_selec(root, (Node *) clause, NIL);
+ {
+ relids = bms_union(relids, addrelids);
+ /* Should still be a subset of current scope ... */
+ Assert(bms_is_subset(relids, qualscope));
- rel->restrictinfo = lcons(restrictinfo, rel->restrictinfo);
+ /*
+ * Because application of the qual will be delayed by outer
+ * join, we mustn't assume its vars are equal everywhere.
+ */
+ can_be_equijoin = false;
+ }
+ bms_free(addrelids);
}
- else
+
+ /*
+ * Mark the qual as "pushed down" if it can be applied at a level
+ * below its original syntactic level. This allows us to distinguish
+ * original JOIN/ON quals from higher-level quals pushed down to the
+ * same joinrel. A qual originating from WHERE is always considered
+ * "pushed down".
+ */
+ if (!is_pushed_down)
+ is_pushed_down = !bms_equal(relids, qualscope);
+
+ /*
+ * Build the RestrictInfo node itself.
+ */
+ restrictinfo = make_restrictinfo((Expr *) clause,
+ is_pushed_down,
+ valid_everywhere);
+
+ /*
+ * Figure out where to attach it.
+ */
+ switch (bms_membership(relids))
{
+ case BMS_SINGLETON:
- /*
- * 'clause' is a join clause, since there is more than one atom in
- * the relid list.
- */
- if (is_funcclause((Node *) clause))
+ /*
+ * There is only one relation participating in 'clause', so
+ * 'clause' is a restriction clause for that relation.
+ */
+ rel = find_base_rel(root, bms_singleton_member(relids));
/*
- * XXX If we have a func clause set selectivity to 1/3, really
- * need a true selectivity function.
+ * Check for a "mergejoinable" clause even though it's not a
+ * join clause. This is so that we can recognize that "a.x =
+ * a.y" makes x and y eligible to be considered equal, even
+ * when they belong to the same rel. Without this, we would
+ * not recognize that "a.x = a.y AND a.x = b.z AND a.y = c.q"
+ * allows us to consider z and q equal after their rels are
+ * joined.
*/
- restrictinfo->selectivity = (Cost) 0.3333333;
- else
- restrictinfo->selectivity = compute_clause_selec(root, (Node *) clause, NIL);
+ if (can_be_equijoin)
+ check_mergejoinable(restrictinfo);
+
+ /*
+ * If the clause was deduced from implied equality, check to
+ * see whether it is redundant with restriction clauses we
+ * already have for this rel. Note we cannot apply this check
+ * to user-written clauses, since we haven't found the
+ * canonical pathkey sets yet while processing user clauses.
+ * (NB: no comparable check is done in the join-clause case;
+ * redundancy will be detected when the join clause is moved
+ * into a join rel's restriction list.)
+ */
+ if (!isdeduced ||
+ !qual_is_redundant(root, restrictinfo,
+ rel->baserestrictinfo))
+ {
+ /* Add clause to rel's restriction list */
+ rel->baserestrictinfo = lappend(rel->baserestrictinfo,
+ restrictinfo);
+ }
+ break;
+ case BMS_MULTIPLE:
- add_join_info_to_rels(root, restrictinfo, relids);
- /* we are going to be doing a join, so add var to targetlist */
- add_vars_to_targetlist(root, vars, relids);
+ /*
+ * 'clause' is a join clause, since there is more than one rel
+ * in the relid set.
+ */
+
+ /*
+ * Check for hash or mergejoinable operators.
+ *
+ * We don't bother setting the hashjoin info if we're not going
+ * to need it. We do want to know about mergejoinable ops in
+ * all cases, however, because we use mergejoinable ops for
+ * other purposes such as detecting redundant clauses.
+ */
+ check_mergejoinable(restrictinfo);
+ if (enable_hashjoin)
+ check_hashjoinable(restrictinfo);
+
+ /*
+ * Add clause to the join lists of all the relevant relations.
+ */
+ add_join_clause_to_rels(root, restrictinfo, relids);
+
+ /*
+ * Add vars used in the join clause to targetlists of their
+ * relations, so that they will be emitted by the plan nodes
+ * that scan those relations (else they won't be available at
+ * the join node!).
+ */
+ vars = pull_var_clause(clause, false);
+ add_vars_to_targetlist(root, vars, relids);
+ list_free(vars);
+ break;
+ default:
+
+ /*
+ * 'clause' references no rels, and therefore we have no place
+ * to attach it. Shouldn't get here if callers are working
+ * properly.
+ */
+ elog(ERROR, "cannot cope with variable-free clause");
+ break;
}
+
+ /*
+ * If the clause has a mergejoinable operator, and is not an
+ * outer-join qualification nor bubbled up due to an outer join, then
+ * the two sides represent equivalent PathKeyItems for path keys: any
+ * path that is sorted by one side will also be sorted by the other
+ * (as soon as the two rels are joined, that is). Record the key
+ * equivalence for future use. (We can skip this for a deduced
+ * clause, since the keys are already known equivalent in that case.)
+ */
+ if (can_be_equijoin &&
+ restrictinfo->mergejoinoperator != InvalidOid &&
+ !isdeduced)
+ add_equijoined_keys(root, restrictinfo);
}
/*
- * add_join_info_to_rels
- * For every relation participating in a join clause, add 'restrictinfo' to
- * the appropriate joininfo node(creating a new one and adding it to the
- * appropriate rel node if necessary).
- *
- * 'restrictinfo' describes the join clause
- * 'join_relids' is the list of relations participating in the join clause
+ * process_implied_equality
+ * Check to see whether we already have a restrictinfo item that says
+ * item1 = item2, and create one if not; or if delete_it is true,
+ * remove any such restrictinfo item.
*
+ * This processing is a consequence of transitivity of mergejoin equality:
+ * if we have mergejoinable clauses A = B and B = C, we can deduce A = C
+ * (where = is an appropriate mergejoinable operator). See path/pathkeys.c
+ * for more details.
*/
-static void
-add_join_info_to_rels(Query *root, RestrictInfo *restrictinfo,
- Relids join_relids)
+void
+process_implied_equality(PlannerInfo *root,
+ Node *item1, Node *item2,
+ Oid sortop1, Oid sortop2,
+ Relids item1_relids, Relids item2_relids,
+ bool delete_it)
{
- List *join_relid;
+ Relids relids;
+ BMS_Membership membership;
+ RelOptInfo *rel1;
+ List *restrictlist;
+ ListCell *itm;
+ Oid ltype,
+ rtype;
+ Operator eq_operator;
+ Form_pg_operator pgopform;
+ Expr *clause;
+
+ /* Get set of relids referenced in the two expressions */
+ relids = bms_union(item1_relids, item2_relids);
+ membership = bms_membership(relids);
+
+ /*
+ * generate_implied_equalities() shouldn't call me on two constants.
+ */
+ Assert(membership != BMS_EMPTY_SET);
- /* For every relid, find the rel, and add the proper join entries */
- foreach(join_relid, join_relids)
+ /*
+ * If the exprs involve a single rel, we need to look at that rel's
+ * baserestrictinfo list. If multiple rels, any one will have a
+ * joininfo node for the rest, and we can scan any of 'em.
+ */
+ if (membership == BMS_SINGLETON)
+ {
+ rel1 = find_base_rel(root, bms_singleton_member(relids));
+ restrictlist = rel1->baserestrictinfo;
+ }
+ else
{
+ Relids other_rels;
+ int first_rel;
JoinInfo *joininfo;
- Relids unjoined_relids = NIL;
- List *rel;
- /* Get the relids not equal to the current relid */
- foreach(rel, join_relids)
+ /* Copy relids, find and remove one member */
+ other_rels = bms_copy(relids);
+ first_rel = bms_first_member(other_rels);
+
+ rel1 = find_base_rel(root, first_rel);
+
+ /* use remaining members to find join node */
+ joininfo = find_joininfo_node(rel1, other_rels);
+
+ restrictlist = joininfo ? joininfo->jinfo_restrictinfo : NIL;
+
+ bms_free(other_rels);
+ }
+
+ /*
+ * Scan to see if equality is already known. If so, we're done in the
+ * add case, and done after removing it in the delete case.
+ */
+ foreach(itm, restrictlist)
+ {
+ RestrictInfo *restrictinfo = (RestrictInfo *) lfirst(itm);
+ Node *left,
+ *right;
+
+ if (restrictinfo->mergejoinoperator == InvalidOid)
+ continue; /* ignore non-mergejoinable clauses */
+ /* We now know the restrictinfo clause is a binary opclause */
+ left = get_leftop(restrictinfo->clause);
+ right = get_rightop(restrictinfo->clause);
+ if ((equal(item1, left) && equal(item2, right)) ||
+ (equal(item2, left) && equal(item1, right)))
{
- if (lfirsti(rel) != lfirsti(join_relid))
- unjoined_relids = lappendi(unjoined_relids, lfirsti(rel));
+ /* found a matching clause */
+ if (delete_it)
+ {
+ if (membership == BMS_SINGLETON)
+ {
+ /* delete it from local restrictinfo list */
+ rel1->baserestrictinfo = list_delete_ptr(rel1->baserestrictinfo,
+ restrictinfo);
+ }
+ else
+ {
+ /* let joininfo.c do it */
+ remove_join_clause_from_rels(root, restrictinfo, relids);
+ }
+ }
+ return; /* done */
}
+ }
+
+ /* Didn't find it. Done if deletion requested */
+ if (delete_it)
+ return;
- joininfo = find_joininfo_node(get_base_rel(root, lfirsti(join_relid)),
- unjoined_relids);
- joininfo->jinfo_restrictinfo = lcons(copyObject((void *) restrictinfo),
- joininfo->jinfo_restrictinfo);
+ /*
+ * This equality is new information, so construct a clause
+ * representing it to add to the query data structures.
+ */
+ ltype = exprType(item1);
+ rtype = exprType(item2);
+ eq_operator = compatible_oper(list_make1(makeString("=")),
+ ltype, rtype, true);
+ if (!HeapTupleIsValid(eq_operator))
+ {
+ /*
+ * Would it be safe to just not add the equality to the query if
+ * we have no suitable equality operator for the combination of
+ * datatypes? NO, because sortkey selection may screw up anyway.
+ */
+ ereport(ERROR,
+ (errcode(ERRCODE_UNDEFINED_FUNCTION),
+ errmsg("could not identify an equality operator for types %s and %s",
+ format_type_be(ltype), format_type_be(rtype))));
}
+ pgopform = (Form_pg_operator) GETSTRUCT(eq_operator);
+
+ /*
+ * Let's just make sure this appears to be a compatible operator.
+ */
+ if (pgopform->oprlsortop != sortop1 ||
+ pgopform->oprrsortop != sortop2 ||
+ pgopform->oprresult != BOOLOID)
+ ereport(ERROR,
+ (errcode(ERRCODE_INVALID_FUNCTION_DEFINITION),
+ errmsg("equality operator for types %s and %s should be merge-joinable, but isn't",
+ format_type_be(ltype), format_type_be(rtype))));
+
+ /*
+ * Now we can build the new clause. Copy to ensure it shares no
+ * substructure with original (this is necessary in case there are
+ * subselects in there...)
+ */
+ clause = make_opclause(oprid(eq_operator), /* opno */
+ BOOLOID, /* opresulttype */
+ false, /* opretset */
+ (Expr *) copyObject(item1),
+ (Expr *) copyObject(item2));
+
+ ReleaseSysCache(eq_operator);
+
+ /*
+ * Push the new clause into all the appropriate restrictinfo lists.
+ *
+ * Note: we mark the qual "pushed down" to ensure that it can never be
+ * taken for an original JOIN/ON clause.
+ */
+ distribute_qual_to_rels(root, (Node *) clause,
+ true, true, NULL, relids);
}
/*
- * add_vars_to_targetlist
- * For each variable appearing in a clause,
- * (1) If a targetlist entry for the variable is not already present in
- * the appropriate relation's target list, add one.
- * (2) If a targetlist entry is already present, but the var is part of a
- * join clause, add the relids of the join relations to the JoinList
- * entry of the targetlist entry.
+ * qual_is_redundant
+ * Detect whether an implied-equality qual that turns out to be a
+ * restriction clause for a single base relation is redundant with
+ * already-known restriction clauses for that rel. This occurs with,
+ * for example,
+ * SELECT * FROM tab WHERE f1 = f2 AND f2 = f3;
+ * We need to suppress the redundant condition to avoid computing
+ * too-small selectivity, not to mention wasting time at execution.
*
- * 'vars' is the list of var nodes
- * 'join_relids' is the list of relids appearing in the join clause
- * (if this is a join clause)
- *
- * Returns nothing.
+ * Note: quals of the form "var = const" are never considered redundant,
+ * only those of the form "var = var". This is needed because when we
+ * have constants in an implied-equality set, we use a different strategy
+ * that suppresses all "var = var" deductions. We must therefore keep
+ * all the "var = const" quals.
*/
-static void
-add_vars_to_targetlist(Query *root, List *vars, Relids join_relids)
+static bool
+qual_is_redundant(PlannerInfo *root,
+ RestrictInfo *restrictinfo,
+ List *restrictlist)
{
- Var *var;
- List *temp = NIL;
- RelOptInfo *rel = (RelOptInfo *) NULL;
- TargetEntry *tlistentry;
+ Node *newleft;
+ Node *newright;
+ List *oldquals;
+ ListCell *olditem;
+ List *equalexprs;
+ bool someadded;
- foreach(temp, vars)
+ /* Never redundant unless vars appear on both sides */
+ if (bms_is_empty(restrictinfo->left_relids) ||
+ bms_is_empty(restrictinfo->right_relids))
+ return false;
+
+ newleft = get_leftop(restrictinfo->clause);
+ newright = get_rightop(restrictinfo->clause);
+
+ /*
+ * Set cached pathkeys. NB: it is okay to do this now because this
+ * routine is only invoked while we are generating implied equalities.
+ * Therefore, the equi_key_list is already complete and so we can
+ * correctly determine canonical pathkeys.
+ */
+ cache_mergeclause_pathkeys(root, restrictinfo);
+ /* If different, say "not redundant" (should never happen) */
+ if (restrictinfo->left_pathkey != restrictinfo->right_pathkey)
+ return false;
+
+ /*
+ * Scan existing quals to find those referencing same pathkeys.
+ * Usually there will be few, if any, so build a list of just the
+ * interesting ones.
+ */
+ oldquals = NIL;
+ foreach(olditem, restrictlist)
{
- var = (Var *) lfirst(temp);
- rel = get_base_rel(root, var->varno);
- tlistentry = tlistentry_member(var, rel->targetlist);
- if (tlistentry == NULL)
- /* add a new entry */
- add_var_to_tlist(rel, var);
+ RestrictInfo *oldrinfo = (RestrictInfo *) lfirst(olditem);
+
+ if (oldrinfo->mergejoinoperator != InvalidOid)
+ {
+ cache_mergeclause_pathkeys(root, oldrinfo);
+ if (restrictinfo->left_pathkey == oldrinfo->left_pathkey &&
+ restrictinfo->right_pathkey == oldrinfo->right_pathkey)
+ oldquals = lcons(oldrinfo, oldquals);
+ }
}
+ if (oldquals == NIL)
+ return false;
+
+ /*
+ * Now, we want to develop a list of exprs that are known equal to the
+ * left side of the new qual. We traverse the old-quals list
+ * repeatedly to transitively expand the exprs list. If at any point
+ * we find we can reach the right-side expr of the new qual, we are
+ * done. We give up when we can't expand the equalexprs list any
+ * more.
+ */
+ equalexprs = list_make1(newleft);
+ do
+ {
+ someadded = false;
+ /* cannot use foreach here because of possible list_delete */
+ olditem = list_head(oldquals);
+ while (olditem)
+ {
+ RestrictInfo *oldrinfo = (RestrictInfo *) lfirst(olditem);
+ Node *oldleft = get_leftop(oldrinfo->clause);
+ Node *oldright = get_rightop(oldrinfo->clause);
+ Node *newguy = NULL;
+
+ /* must advance olditem before list_delete possibly pfree's it */
+ olditem = lnext(olditem);
+
+ if (list_member(equalexprs, oldleft))
+ newguy = oldright;
+ else if (list_member(equalexprs, oldright))
+ newguy = oldleft;
+ else
+ continue;
+ if (equal(newguy, newright))
+ return true; /* we proved new clause is redundant */
+ equalexprs = lcons(newguy, equalexprs);
+ someadded = true;
+
+ /*
+ * Remove this qual from list, since we don't need it anymore.
+ */
+ oldquals = list_delete_ptr(oldquals, oldrinfo);
+ }
+ } while (someadded);
+
+ return false; /* it's not redundant */
}
+
/*****************************************************************************
*
- * JOININFO
+ * CHECKS FOR MERGEJOINABLE AND HASHJOINABLE CLAUSES
*
*****************************************************************************/
/*
- * set_joininfo_mergeable_hashable
- * Set the MergeJoinable or HashJoinable field for every joininfo node
- * (within a rel node) and the mergejoinorder or hashjoinop field for
- * each restrictinfo node(within a joininfo node) for all relations in a
- * query.
+ * check_mergejoinable
+ * If the restrictinfo's clause is mergejoinable, set the mergejoin
+ * info fields in the restrictinfo.
*
- * Returns nothing.
- */
-void
-set_joininfo_mergeable_hashable(List *rel_list)
-{
- List *x,
- *y,
- *z;
- RelOptInfo *rel;
- JoinInfo *joininfo;
- RestrictInfo *restrictinfo;
- Expr *clause;
-
- foreach(x, rel_list)
- {
- rel = (RelOptInfo *) lfirst(x);
- foreach(y, rel->joininfo)
- {
- joininfo = (JoinInfo *) lfirst(y);
- foreach(z, joininfo->jinfo_restrictinfo)
- {
- restrictinfo = (RestrictInfo *) lfirst(z);
- clause = restrictinfo->clause;
- if (is_joinable((Node *) clause))
- {
- MergeOrder *sortop = (MergeOrder *) NULL;
- Oid hashop = (Oid) NULL;
-
- if (_enable_mergejoin_)
- sortop = mergejoinop(clause);
- if (sortop)
- {
- restrictinfo->mergejoinorder = sortop;
- joininfo->mergejoinable = true;
- }
-
- if (_enable_hashjoin_)
- hashop = hashjoinop(clause);
- if (hashop)
- {
- restrictinfo->hashjoinoperator = hashop;
- joininfo->hashjoinable = true;
- }
- }
- }
- }
- }
-}
-
-/*
- * mergejoinop
- * Returns the mergejoin operator of an operator iff 'clause' is
- * mergejoinable, i.e., both operands are single vars and the operator is
- * a mergejoinable operator.
+ * Currently, we support mergejoin for binary opclauses where
+ * the operator is a mergejoinable operator. The arguments can be
+ * anything --- as long as there are no volatile functions in them.
*/
-static MergeOrder *
-mergejoinop(Expr *clause)
+static void
+check_mergejoinable(RestrictInfo *restrictinfo)
{
- Var *left,
- *right;
+ Expr *clause = restrictinfo->clause;
Oid opno,
leftOp,
rightOp;
- bool sortable;
-
- if (!is_opclause((Node *) clause))
- return NULL;
-
- left = get_leftop(clause);
- right = get_rightop(clause);
-
- /* caution: is_opclause accepts more than I do, so check it */
- if (!right)
- return NULL; /* unary opclauses need not apply */
- if (!IsA(left, Var) ||!IsA(right, Var))
- return NULL;
- opno = ((Oper *) clause->oper)->opno;
+ if (!is_opclause(clause))
+ return;
+ if (list_length(((OpExpr *) clause)->args) != 2)
+ return;
- sortable = op_mergejoinable(opno,
- left->vartype,
- right->vartype,
- &leftOp,
- &rightOp);
+ opno = ((OpExpr *) clause)->opno;
- if (sortable)
+ if (op_mergejoinable(opno,
+ &leftOp,
+ &rightOp) &&
+ !contain_volatile_functions((Node *) clause))
{
- MergeOrder *morder = makeNode(MergeOrder);
-
- morder->join_operator = opno;
- morder->left_operator = leftOp;
- morder->right_operator = rightOp;
- morder->left_type = left->vartype;
- morder->right_type = right->vartype;
- return morder;
+ restrictinfo->mergejoinoperator = opno;
+ restrictinfo->left_sortop = leftOp;
+ restrictinfo->right_sortop = rightOp;
}
- else
- return NULL;
}
/*
- * hashjoinop
- * Returns the hashjoin operator of an operator iff 'clause' is
- * hashjoinable, i.e., both operands are single vars and the operator is
- * a hashjoinable operator.
+ * check_hashjoinable
+ * If the restrictinfo's clause is hashjoinable, set the hashjoin
+ * info fields in the restrictinfo.
+ *
+ * Currently, we support hashjoin for binary opclauses where
+ * the operator is a hashjoinable operator. The arguments can be
+ * anything --- as long as there are no volatile functions in them.
*/
-static Oid
-hashjoinop(Expr *clause)
+static void
+check_hashjoinable(RestrictInfo *restrictinfo)
{
- Var *left,
- *right;
-
- if (!is_opclause((Node *) clause))
- return InvalidOid;
+ Expr *clause = restrictinfo->clause;
+ Oid opno;
- left = get_leftop(clause);
- right = get_rightop(clause);
+ if (!is_opclause(clause))
+ return;
+ if (list_length(((OpExpr *) clause)->args) != 2)
+ return;
- /* caution: is_opclause accepts more than I do, so check it */
- if (!right)
- return InvalidOid; /* unary opclauses need not apply */
- if (!IsA(left, Var) ||!IsA(right, Var))
- return InvalidOid;
+ opno = ((OpExpr *) clause)->opno;
- return op_hashjoinable(((Oper *) clause->oper)->opno,
- left->vartype,
- right->vartype);
+ if (op_hashjoinable(opno) &&
+ !contain_volatile_functions((Node *) clause))
+ restrictinfo->hashjoinoperator = opno;
}
projects / postgresql / blobdiff