subselect.c

来自「postgresql8.3.4源码,开源数据库」· C语言 代码 · 共 1,470 行 · 第 1/3 页

												   splan->plan_id);

	return result;
}

/*
 * generate_subquery_params: build a list of Params representing the output
 * columns of a sublink's sub-select, given the sub-select's targetlist.
 *
 * We also return an integer list of the paramids of the Params.
 */
static List *
generate_subquery_params(PlannerInfo *root, List *tlist, List **paramIds)
{
	List	   *result;
	List	   *ids;
	ListCell   *lc;

	result = ids = NIL;
	foreach(lc, tlist)
	{
		TargetEntry *tent = (TargetEntry *) lfirst(lc);
		Param	   *param;

		if (tent->resjunk)
			continue;

		param = generate_new_param(root,
								   exprType((Node *) tent->expr),
								   exprTypmod((Node *) tent->expr));
		result = lappend(result, param);
		ids = lappend_int(ids, param->paramid);
	}

	*paramIds = ids;
	return result;
}

/*
 * generate_subquery_vars: build a list of Vars representing the output
 * columns of a sublink's sub-select, given the sub-select's targetlist.
 * The Vars have the specified varno (RTE index).
 */
static List *
generate_subquery_vars(PlannerInfo *root, List *tlist, Index varno)
{
	List	   *result;
	ListCell   *lc;

	result = NIL;
	foreach(lc, tlist)
	{
		TargetEntry *tent = (TargetEntry *) lfirst(lc);
		Var		   *var;

		if (tent->resjunk)
			continue;

		var = makeVar(varno,
					  tent->resno,
					  exprType((Node *) tent->expr),
					  exprTypmod((Node *) tent->expr),
					  0);
		result = lappend(result, var);
	}

	return result;
}

/*
 * convert_testexpr: convert the testexpr given by the parser into
 * actually executable form.  This entails replacing PARAM_SUBLINK Params
 * with Params or Vars representing the results of the sub-select.  The
 * nodes to be substituted are passed in as the List result from
 * generate_subquery_params or generate_subquery_vars.
 *
 * The given testexpr has already been recursively processed by
 * process_sublinks_mutator.  Hence it can no longer contain any
 * PARAM_SUBLINK Params for lower SubLink nodes; we can safely assume that
 * any we find are for our own level of SubLink.
 */
static Node *
convert_testexpr(PlannerInfo *root,
				 Node *testexpr,
				 List *subst_nodes)
{
	convert_testexpr_context context;

	context.root = root;
	context.subst_nodes = subst_nodes;
	return convert_testexpr_mutator(testexpr, &context);
}

static Node *
convert_testexpr_mutator(Node *node,
						 convert_testexpr_context *context)
{
	if (node == NULL)
		return NULL;
	if (IsA(node, Param))
	{
		Param	   *param = (Param *) node;

		if (param->paramkind == PARAM_SUBLINK)
		{
			if (param->paramid <= 0 ||
				param->paramid > list_length(context->subst_nodes))
				elog(ERROR, "unexpected PARAM_SUBLINK ID: %d", param->paramid);

			/*
			 * We copy the list item to avoid having doubly-linked
			 * substructure in the modified parse tree.  This is probably
			 * unnecessary when it's a Param, but be safe.
			 */
			return (Node *) copyObject(list_nth(context->subst_nodes,
												param->paramid - 1));
		}
	}
	return expression_tree_mutator(node,
								   convert_testexpr_mutator,
								   (void *) context);
}

/*
 * 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, Plan *plan)
{
	double		subquery_size;
	ListCell   *l;

	/*
	 * The sublink type must be "= ANY" --- that is, an IN operator.  We
	 * expect that the test expression will be either a single OpExpr, or an
	 * AND-clause containing OpExprs.  (If it's anything else then the parser
	 * must have determined that the operators have non-equality-like
	 * semantics.  In the OpExpr case we can't be sure what the operator's
	 * semantics are like, but the test below for hashability will reject
	 * anything that's not equality.)
	 */
	if (slink->subLinkType != ANY_SUBLINK)
		return false;
	if (slink->testexpr == NULL ||
		(!IsA(slink->testexpr, OpExpr) &&
		 !and_clause(slink->testexpr)))
		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. (Note:
	 * we use sizeof(HeapTupleHeaderData) here even though the tuples will
	 * actually be stored as MinimalTuples; this provides some fudge factor
	 * for hashtable overhead.)
	 */
	subquery_size = plan->plan_rows *
		(MAXALIGN(plan->plan_width) + MAXALIGN(sizeof(HeapTupleHeaderData)));
	if (subquery_size > work_mem * 1024L)
		return false;

	/*
	 * The combining operators must be hashable and strict. 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: they can't yield
	 * NULL for non-null inputs, either (see nodeSubplan.c).  However, hash
	 * indexes and hash joins assume that too.
	 */
	if (IsA(slink->testexpr, OpExpr))
	{
		if (!hash_ok_operator((OpExpr *) slink->testexpr))
			return false;
	}
	else
	{
		foreach(l, ((BoolExpr *) slink->testexpr)->args)
		{
			Node	   *andarg = (Node *) lfirst(l);

			if (!IsA(andarg, OpExpr))
				return false;	/* probably can't happen */
			if (!hash_ok_operator((OpExpr *) andarg))
				return false;
		}
	}

	return true;
}

static bool
hash_ok_operator(OpExpr *expr)
{
	Oid			opid = expr->opno;
	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 || !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;
	List	   *in_operators;
	List	   *left_exprs;
	List	   *right_exprs;
	Relids		left_varnos;
	int			rtindex;
	RangeTblEntry *rte;
	RangeTblRef *rtr;
	List	   *subquery_vars;
	InClauseInfo *ininfo;
	Node	   *result;

	/*
	 * The sublink type must be "= ANY" --- that is, an IN operator.  We
	 * expect that the test expression will be either a single OpExpr, or an
	 * AND-clause containing OpExprs.  (If it's anything else then the parser
	 * must have determined that the operators have non-equality-like
	 * semantics.  In the OpExpr case we can't be sure what the operator's
	 * semantics are like, and must check for ourselves.)
	 */
	if (sublink->subLinkType != ANY_SUBLINK)
		return NULL;
	if (sublink->testexpr && IsA(sublink->testexpr, OpExpr))
	{
		OpExpr	   *op = (OpExpr *) sublink->testexpr;
		Oid			opno = op->opno;
		List	   *opfamilies;
		List	   *opstrats;

		if (list_length(op->args) != 2)
			return NULL;				/* not binary operator? */
		get_op_btree_interpretation(opno, &opfamilies, &opstrats);
		if (!list_member_int(opstrats, ROWCOMPARE_EQ))
			return NULL;
		in_operators = list_make1_oid(opno);
		left_exprs = list_make1(linitial(op->args));
		right_exprs = list_make1(lsecond(op->args));
	}
	else if (and_clause(sublink->testexpr))
	{
		ListCell   *lc;

		/* OK, but we need to extract the per-column info */
		in_operators = left_exprs = right_exprs = NIL;
		foreach(lc, ((BoolExpr *) sublink->testexpr)->args)
		{
			OpExpr	   *op = (OpExpr *) lfirst(lc);

			if (!IsA(op, OpExpr))		/* probably shouldn't happen */
				return NULL;
			if (list_length(op->args) != 2)
				return NULL;			/* not binary operator? */
			in_operators = lappend_oid(in_operators, op->opno);
			left_exprs = lappend(left_exprs, linitial(op->args));
			right_exprs = lappend(right_exprs, lsecond(op->args));
		}
	}
	else
		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 *) left_exprs);
	if (bms_is_empty(left_varnos))
		return NULL;

	/* ... and the right-hand expressions better not contain Vars at all */
	Assert(!contain_var_clause((Node *) right_exprs));

	/*
	 * The combining operators and left-hand expressions mustn't be volatile.
	 */
	if (contain_volatile_functions(sublink->testexpr))
		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);

	/*
	 * Build a list of Vars representing the subselect outputs.
	 */
	subquery_vars = generate_subquery_vars(root,
										   subselect->targetList,
										   rtindex);

	/*
	 * Build the result qual expression, replacing Params with these Vars.
	 */
	result = convert_testexpr(root,
							  sublink->testexpr,
							  subquery_vars);

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

	/*
	 * ininfo->sub_targetlist must be filled with a list of expressions that
	 * would need to be unique-ified if we try to implement the IN using a
	 * regular join to unique-ified subquery output.  This is most easily done
	 * by applying convert_testexpr to just the RHS inputs of the testexpr
	 * operators.  That handles cases like type coercions of the subquery
	 * outputs, clauses dropped due to const-simplification, etc.
	 */
	ininfo->sub_targetlist = (List *) convert_testexpr(root,
													   (Node *) right_exprs,
													   subquery_vars);

	/* Add the completed node to the query's list */
	root->in_info_list = lappend(root->in_info_list, ininfo);

	return result;
}

/*
 * 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(PlannerInfo *root, Node *expr)
{
	/* No setup needed for tree walk, so away we go */
	return replace_correlation_vars_mutator(expr, root);
}

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

/*
 * 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(PlannerInfo *root, Node *expr, bool isQual)
{
	process_sublinks_context context;

	context.root = root;
	context.isTopQual = isQual;
	return process_sublinks_mutator(expr, &context);
}

static Node *
process_sublinks_mutator(Node *node, process_sublinks_context *context)
{
	process_sublinks_context locContext;

	locContext.root = context->root;

	if (node == NULL)
		return NULL;
	if (IsA(node, SubLink))
	{
		SubLink    *sublink = (SubLink *) node;
		Node	   *testexpr;

		/*
		 * First, recursively process the lefthand-side expressions, if any.
		 * They're not top-level anymore.
		 */
		locContext.isTopQual = false;
		testexpr = process_sublinks_mutator(sublink->testexpr, &locContext);

		/*
		 * Now build the SubPlan node and make the expr to return.
		 */
		return make_subplan(context->root,
							sublink,
							testexpr,
							context->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))
	{