subselect.c

PostgreSQL 8.1.4的源码 适用于Linux下的开源数据库系统

			if (use_material)
				node->plan = plan = materialize_finished_plan(plan);
		}

		/* Convert the lefthand exprs and oper OIDs into executable exprs */
		node->exprs = convert_sublink_opers(lefthand,
											slink->operOids,
											plan->targetlist,
											0,
											&node->paramIds);

		/*
		 * Make node->args from parParam.
		 */
		args = NIL;
		foreach(l, node->parParam)
		{
			PlannerParamItem *pitem = list_nth(PlannerParamList, lfirst_int(l));

			/*
			 * The Var or Aggref has already been adjusted to have the correct
			 * varlevelsup or agglevelsup.	We probably don't even need to
			 * copy it again, but be safe.
			 */
			args = lappend(args, copyObject(pitem->item));
		}
		node->args = args;

		result = (Node *) node;
	}

	return result;
}

/*
 * convert_sublink_opers: given a lefthand-expressions list and a list of
 * operator OIDs, build a list of actually executable expressions.	The
 * righthand sides of the expressions are Params or Vars representing the
 * results of the sub-select.
 *
 * If rtindex is 0, we build Params to represent the sub-select outputs.
 * The paramids of the Params created are returned in the *righthandIds list.
 *
 * If rtindex is not 0, we build Vars using that rtindex as varno.	Copies
 * of the Var nodes are returned in *righthandIds (this is a bit of a type
 * cheat, but we can get away with it).
 */
static List *
convert_sublink_opers(List *lefthand, List *operOids,
					  List *targetlist, int rtindex,
					  List **righthandIds)
{
	List	   *result = NIL;
	ListCell   *l,
			   *lefthand_item,
			   *tlist_item;

	*righthandIds = NIL;
	lefthand_item = list_head(lefthand);
	tlist_item = list_head(targetlist);

	foreach(l, operOids)
	{
		Oid			opid = lfirst_oid(l);
		Node	   *leftop = (Node *) lfirst(lefthand_item);
		TargetEntry *te = (TargetEntry *) lfirst(tlist_item);
		Node	   *rightop;
		Operator	tup;

		Assert(!te->resjunk);

		if (rtindex)
		{
			/* Make the Var node representing the subplan's result */
			rightop = (Node *) makeVar(rtindex,
									   te->resno,
									   exprType((Node *) te->expr),
									   exprTypmod((Node *) te->expr),
									   0);

			/*
			 * Copy it for caller.	NB: we need a copy to avoid having
			 * doubly-linked substructure in the modified parse tree.
			 */
			*righthandIds = lappend(*righthandIds, copyObject(rightop));
		}
		else
		{
			/* Make the Param node representing the subplan's result */
			Param	   *prm;

			prm = generate_new_param(exprType((Node *) te->expr),
									 exprTypmod((Node *) te->expr));
			/* Record its ID */
			*righthandIds = lappend_int(*righthandIds, prm->paramid);
			rightop = (Node *) prm;
		}

		/* Look up the operator to pass to make_op_expr */
		tup = SearchSysCache(OPEROID,
							 ObjectIdGetDatum(opid),
							 0, 0, 0);
		if (!HeapTupleIsValid(tup))
			elog(ERROR, "cache lookup failed for operator %u", opid);

		/*
		 * Make the expression node.
		 *
		 * Note: we use make_op_expr in case runtime type conversion function
		 * calls must be inserted for this operator!  (But we are not
		 * expecting to have to resolve unknown Params, so it's okay to pass a
		 * null pstate.)
		 */
		result = lappend(result,
						 make_op_expr(NULL,
									  tup,
									  leftop,
									  rightop,
									  exprType(leftop),
									  exprType((Node *) te->expr)));

		ReleaseSysCache(tup);

		lefthand_item = lnext(lefthand_item);
		tlist_item = lnext(tlist_item);
	}

	return result;
}

/*
 * subplan_is_hashable: decide whether we can implement a subplan by hashing
 *
 * Caution: the SubPlan node is not completely filled in yet.  We can rely
 * on its plan and parParam fields, however.
 */
static bool
subplan_is_hashable(SubLink *slink, SubPlan *node)
{
	double		subquery_size;
	ListCell   *l;

	/*
	 * The sublink type must be "= ANY" --- that is, an IN operator. (We
	 * require the operator name to be unqualified, which may be overly
	 * paranoid, or may not be.)  XXX since we also check that the operators
	 * are hashable, the test on operator name may be redundant?
	 */
	if (slink->subLinkType != ANY_SUBLINK)
		return false;
	if (list_length(slink->operName) != 1 ||
		strcmp(strVal(linitial(slink->operName)), "=") != 0)
		return false;

	/*
	 * The subplan must not have any direct correlation vars --- else we'd
	 * have to recompute its output each time, so that the hashtable wouldn't
	 * gain anything.
	 */
	if (node->parParam != NIL)
		return false;

	/*
	 * The estimated size of the subquery result must fit in work_mem. (XXX
	 * what about hashtable overhead?)
	 */
	subquery_size = node->plan->plan_rows *
		(MAXALIGN(node->plan->plan_width) + MAXALIGN(sizeof(HeapTupleData)));
	if (subquery_size > work_mem * 1024L)
		return false;

	/*
	 * The combining operators must be hashable, strict, and self-commutative.
	 * The need for hashability is obvious, since we want to use hashing.
	 * Without strictness, behavior in the presence of nulls is too
	 * unpredictable.  (We actually must assume even more than plain
	 * strictness, see nodeSubplan.c for details.)	And commutativity ensures
	 * that the left and right datatypes are the same; this allows us to
	 * assume that the combining operators are equality for the righthand
	 * datatype, so that they can be used to compare righthand tuples as well
	 * as comparing lefthand to righthand tuples.  (This last restriction
	 * could be relaxed by using two different sets of operators with the hash
	 * table, but there is no obvious usefulness to that at present.)
	 */
	foreach(l, slink->operOids)
	{
		Oid			opid = lfirst_oid(l);
		HeapTuple	tup;
		Form_pg_operator optup;

		tup = SearchSysCache(OPEROID,
							 ObjectIdGetDatum(opid),
							 0, 0, 0);
		if (!HeapTupleIsValid(tup))
			elog(ERROR, "cache lookup failed for operator %u", opid);
		optup = (Form_pg_operator) GETSTRUCT(tup);
		if (!optup->oprcanhash || optup->oprcom != opid ||
			!func_strict(optup->oprcode))
		{
			ReleaseSysCache(tup);
			return false;
		}
		ReleaseSysCache(tup);
	}
	return true;
}

/*
 * convert_IN_to_join: can we convert an IN SubLink to join style?
 *
 * The caller has found a SubLink at the top level of WHERE, but has not
 * checked the properties of the SubLink at all.  Decide whether it is
 * appropriate to process this SubLink in join style.  If not, return NULL.
 * If so, build the qual clause(s) to replace the SubLink, and return them.
 *
 * Side effects of a successful conversion include adding the SubLink's
 * subselect to the query's rangetable and adding an InClauseInfo node to
 * its in_info_list.
 */
Node *
convert_IN_to_join(PlannerInfo *root, SubLink *sublink)
{
	Query	   *parse = root->parse;
	Query	   *subselect = (Query *) sublink->subselect;
	Relids		left_varnos;
	int			rtindex;
	RangeTblEntry *rte;
	RangeTblRef *rtr;
	InClauseInfo *ininfo;
	List	   *exprs;

	/*
	 * The sublink type must be "= ANY" --- that is, an IN operator. (We
	 * require the operator name to be unqualified, which may be overly
	 * paranoid, or may not be.)
	 */
	if (sublink->subLinkType != ANY_SUBLINK)
		return NULL;
	if (list_length(sublink->operName) != 1 ||
		strcmp(strVal(linitial(sublink->operName)), "=") != 0)
		return NULL;

	/*
	 * The sub-select must not refer to any Vars of the parent query. (Vars of
	 * higher levels should be okay, though.)
	 */
	if (contain_vars_of_level((Node *) subselect, 1))
		return NULL;

	/*
	 * The left-hand expressions must contain some Vars of the current query,
	 * else it's not gonna be a join.
	 */
	left_varnos = pull_varnos((Node *) sublink->lefthand);
	if (bms_is_empty(left_varnos))
		return NULL;

	/*
	 * The left-hand expressions mustn't be volatile.  (Perhaps we should test
	 * the combining operators, too?  We'd only need to point the function
	 * directly at the sublink ...)
	 */
	if (contain_volatile_functions((Node *) sublink->lefthand))
		return NULL;

	/*
	 * Okay, pull up the sub-select into top range table and jointree.
	 *
	 * We rely here on the assumption that the outer query has no references
	 * to the inner (necessarily true, other than the Vars that we build
	 * below). Therefore this is a lot easier than what pull_up_subqueries has
	 * to go through.
	 */
	rte = addRangeTableEntryForSubquery(NULL,
										subselect,
										makeAlias("IN_subquery", NIL),
										false);
	parse->rtable = lappend(parse->rtable, rte);
	rtindex = list_length(parse->rtable);
	rtr = makeNode(RangeTblRef);
	rtr->rtindex = rtindex;
	parse->jointree->fromlist = lappend(parse->jointree->fromlist, rtr);

	/*
	 * Now build the InClauseInfo node.
	 */
	ininfo = makeNode(InClauseInfo);
	ininfo->lefthand = left_varnos;
	ininfo->righthand = bms_make_singleton(rtindex);
	root->in_info_list = lappend(root->in_info_list, ininfo);

	/*
	 * Build the result qual expressions.  As a side effect,
	 * ininfo->sub_targetlist is filled with a list of Vars representing the
	 * subselect outputs.
	 */
	exprs = convert_sublink_opers(sublink->lefthand,
								  sublink->operOids,
								  subselect->targetList,
								  rtindex,
								  &ininfo->sub_targetlist);
	return (Node *) make_ands_explicit(exprs);
}

/*
 * Replace correlation vars (uplevel vars) with Params.
 *
 * Uplevel aggregates are replaced, too.
 *
 * Note: it is critical that this runs immediately after SS_process_sublinks.
 * Since we do not recurse into the arguments of uplevel aggregates, they will
 * get copied to the appropriate subplan args list in the parent query with
 * uplevel vars not replaced by Params, but only adjusted in level (see
 * replace_outer_agg).	That's exactly what we want for the vars of the parent
 * level --- but if an aggregate's argument contains any further-up variables,
 * they have to be replaced with Params in their turn.	That will happen when
 * the parent level runs SS_replace_correlation_vars.  Therefore it must do
 * so after expanding its sublinks to subplans.  And we don't want any steps
 * in between, else those steps would never get applied to the aggregate
 * argument expressions, either in the parent or the child level.
 */
Node *
SS_replace_correlation_vars(Node *expr)
{
	/* No setup needed for tree walk, so away we go */
	return replace_correlation_vars_mutator(expr, NULL);
}

static Node *
replace_correlation_vars_mutator(Node *node, void *context)
{
	if (node == NULL)
		return NULL;
	if (IsA(node, Var))
	{
		if (((Var *) node)->varlevelsup > 0)
			return (Node *) replace_outer_var((Var *) node);
	}
	if (IsA(node, Aggref))
	{
		if (((Aggref *) node)->agglevelsup > 0)
			return (Node *) replace_outer_agg((Aggref *) node);
	}
	return expression_tree_mutator(node,
								   replace_correlation_vars_mutator,
								   context);
}

/*
 * Expand SubLinks to SubPlans in the given expression.
 *
 * The isQual argument tells whether or not this expression is a WHERE/HAVING
 * qualifier expression.  If it is, any sublinks appearing at top level need
 * not distinguish FALSE from UNKNOWN return values.
 */
Node *
SS_process_sublinks(Node *expr, bool isQual)
{
	/* The only context needed is the initial are-we-in-a-qual flag */
	return process_sublinks_mutator(expr, &isQual);
}

static Node *
process_sublinks_mutator(Node *node, bool *isTopQual)
{
	bool		locTopQual;

	if (node == NULL)
		return NULL;
	if (IsA(node, SubLink))
	{
		SubLink    *sublink = (SubLink *) node;
		List	   *lefthand;

		/*
		 * First, recursively process the lefthand-side expressions, if any.
		 */
		locTopQual = false;
		lefthand = (List *)
			process_sublinks_mutator((Node *) sublink->lefthand, &locTopQual);

		/*
		 * Now build the SubPlan node and make the expr to return.
		 */
		return make_subplan(sublink, lefthand, *isTopQual);
	}

	/*
	 * We should never see a SubPlan expression in the input (since this is
	 * the very routine that creates 'em to begin with).  We shouldn't find
	 * ourselves invoked directly on a Query, either.
	 */
	Assert(!is_subplan(node));
	Assert(!IsA(node, Query));

	/*
	 * Because make_subplan() could return an AND or OR clause, we have to
	 * take steps to preserve AND/OR flatness of a qual.  We assume the input
	 * has been AND/OR flattened and so we need no recursion here.
	 *
	 * If we recurse down through anything other than an AND node, we are
	 * definitely not at top qual level anymore.  (Due to the coding here, we
	 * will not get called on the List subnodes of an AND, so no check is
	 * needed for List.)
	 */
	if (and_clause(node))
	{
		List	   *newargs = NIL;
		ListCell   *l;

		/* Still at qual top-level */
		locTopQual = *isTopQual;

		foreach(l, ((BoolExpr *) node)->args)
		{
			Node	   *newarg;

			newarg = process_sublinks_mutator(lfirst(l),
											  (void *) &locTopQual);
			if (and_clause(newarg))
				newargs = list_concat(newargs, ((BoolExpr *) newarg)->args);
			else
				newargs = lappend(newargs, newarg);
		}
		return (Node *) make_andclause(newargs);
	}

	/* otherwise not at qual top-level */
	locTopQual = false;

	if (or_clause(node))
	{