1 /*-------------------------------------------------------------------------
2  *
3  * parse_clause.c
4  *	  handle clauses in parser
5  *
6  * Portions Copyright (c) 1996-2021, PostgreSQL Global Development Group
7  * Portions Copyright (c) 1994, Regents of the University of California
8  *
9  *
10  * IDENTIFICATION
11  *	  src/backend/parser/parse_clause.c
12  *
13  *-------------------------------------------------------------------------
14  */
15 
16 #include "postgres.h"
17 
18 #include "access/htup_details.h"
19 #include "access/nbtree.h"
20 #include "access/table.h"
21 #include "access/tsmapi.h"
22 #include "catalog/catalog.h"
23 #include "catalog/heap.h"
24 #include "catalog/pg_am.h"
25 #include "catalog/pg_amproc.h"
26 #include "catalog/pg_collation.h"
27 #include "catalog/pg_constraint.h"
28 #include "catalog/pg_type.h"
29 #include "commands/defrem.h"
30 #include "miscadmin.h"
31 #include "nodes/makefuncs.h"
32 #include "nodes/nodeFuncs.h"
33 #include "optimizer/optimizer.h"
34 #include "parser/analyze.h"
35 #include "parser/parse_clause.h"
36 #include "parser/parse_coerce.h"
37 #include "parser/parse_collate.h"
38 #include "parser/parse_expr.h"
39 #include "parser/parse_func.h"
40 #include "parser/parse_oper.h"
41 #include "parser/parse_relation.h"
42 #include "parser/parse_target.h"
43 #include "parser/parse_type.h"
44 #include "parser/parser.h"
45 #include "parser/parsetree.h"
46 #include "rewrite/rewriteManip.h"
47 #include "utils/builtins.h"
48 #include "utils/catcache.h"
49 #include "utils/guc.h"
50 #include "utils/lsyscache.h"
51 #include "utils/rel.h"
52 #include "utils/syscache.h"
53 
54 
55 static int	extractRemainingColumns(ParseNamespaceColumn *src_nscolumns,
56 									List *src_colnames,
57 									List **src_colnos,
58 									List **res_colnames, List **res_colvars,
59 									ParseNamespaceColumn *res_nscolumns);
60 static Node *transformJoinUsingClause(ParseState *pstate,
61 									  List *leftVars, List *rightVars);
62 static Node *transformJoinOnClause(ParseState *pstate, JoinExpr *j,
63 								   List *namespace);
64 static ParseNamespaceItem *transformTableEntry(ParseState *pstate, RangeVar *r);
65 static ParseNamespaceItem *transformRangeSubselect(ParseState *pstate,
66 												   RangeSubselect *r);
67 static ParseNamespaceItem *transformRangeFunction(ParseState *pstate,
68 												  RangeFunction *r);
69 static ParseNamespaceItem *transformRangeTableFunc(ParseState *pstate,
70 												   RangeTableFunc *t);
71 static TableSampleClause *transformRangeTableSample(ParseState *pstate,
72 													RangeTableSample *rts);
73 static ParseNamespaceItem *getNSItemForSpecialRelationTypes(ParseState *pstate,
74 															RangeVar *rv);
75 static Node *transformFromClauseItem(ParseState *pstate, Node *n,
76 									 ParseNamespaceItem **top_nsitem,
77 									 List **namespace);
78 static Var *buildVarFromNSColumn(ParseNamespaceColumn *nscol);
79 static Node *buildMergedJoinVar(ParseState *pstate, JoinType jointype,
80 								Var *l_colvar, Var *r_colvar);
81 static void setNamespaceColumnVisibility(List *namespace, bool cols_visible);
82 static void setNamespaceLateralState(List *namespace,
83 									 bool lateral_only, bool lateral_ok);
84 static void checkExprIsVarFree(ParseState *pstate, Node *n,
85 							   const char *constructName);
86 static TargetEntry *findTargetlistEntrySQL92(ParseState *pstate, Node *node,
87 											 List **tlist, ParseExprKind exprKind);
88 static TargetEntry *findTargetlistEntrySQL99(ParseState *pstate, Node *node,
89 											 List **tlist, ParseExprKind exprKind);
90 static int	get_matching_location(int sortgroupref,
91 								  List *sortgrouprefs, List *exprs);
92 static List *resolve_unique_index_expr(ParseState *pstate, InferClause *infer,
93 									   Relation heapRel);
94 static List *addTargetToGroupList(ParseState *pstate, TargetEntry *tle,
95 								  List *grouplist, List *targetlist, int location);
96 static WindowClause *findWindowClause(List *wclist, const char *name);
97 static Node *transformFrameOffset(ParseState *pstate, int frameOptions,
98 								  Oid rangeopfamily, Oid rangeopcintype, Oid *inRangeFunc,
99 								  Node *clause);
100 
101 
102 /*
103  * transformFromClause -
104  *	  Process the FROM clause and add items to the query's range table,
105  *	  joinlist, and namespace.
106  *
107  * Note: we assume that the pstate's p_rtable, p_joinlist, and p_namespace
108  * lists were initialized to NIL when the pstate was created.
109  * We will add onto any entries already present --- this is needed for rule
110  * processing, as well as for UPDATE and DELETE.
111  */
112 void
transformFromClause(ParseState * pstate,List * frmList)113 transformFromClause(ParseState *pstate, List *frmList)
114 {
115 	ListCell   *fl;
116 
117 	/*
118 	 * The grammar will have produced a list of RangeVars, RangeSubselects,
119 	 * RangeFunctions, and/or JoinExprs. Transform each one (possibly adding
120 	 * entries to the rtable), check for duplicate refnames, and then add it
121 	 * to the joinlist and namespace.
122 	 *
123 	 * Note we must process the items left-to-right for proper handling of
124 	 * LATERAL references.
125 	 */
126 	foreach(fl, frmList)
127 	{
128 		Node	   *n = lfirst(fl);
129 		ParseNamespaceItem *nsitem;
130 		List	   *namespace;
131 
132 		n = transformFromClauseItem(pstate, n,
133 									&nsitem,
134 									&namespace);
135 
136 		checkNameSpaceConflicts(pstate, pstate->p_namespace, namespace);
137 
138 		/* Mark the new namespace items as visible only to LATERAL */
139 		setNamespaceLateralState(namespace, true, true);
140 
141 		pstate->p_joinlist = lappend(pstate->p_joinlist, n);
142 		pstate->p_namespace = list_concat(pstate->p_namespace, namespace);
143 	}
144 
145 	/*
146 	 * We're done parsing the FROM list, so make all namespace items
147 	 * unconditionally visible.  Note that this will also reset lateral_only
148 	 * for any namespace items that were already present when we were called;
149 	 * but those should have been that way already.
150 	 */
151 	setNamespaceLateralState(pstate->p_namespace, false, true);
152 }
153 
154 /*
155  * setTargetTable
156  *	  Add the target relation of INSERT/UPDATE/DELETE to the range table,
157  *	  and make the special links to it in the ParseState.
158  *
159  *	  We also open the target relation and acquire a write lock on it.
160  *	  This must be done before processing the FROM list, in case the target
161  *	  is also mentioned as a source relation --- we want to be sure to grab
162  *	  the write lock before any read lock.
163  *
164  *	  If alsoSource is true, add the target to the query's joinlist and
165  *	  namespace.  For INSERT, we don't want the target to be joined to;
166  *	  it's a destination of tuples, not a source.   For UPDATE/DELETE,
167  *	  we do need to scan or join the target.  (NOTE: we do not bother
168  *	  to check for namespace conflict; we assume that the namespace was
169  *	  initially empty in these cases.)
170  *
171  *	  Finally, we mark the relation as requiring the permissions specified
172  *	  by requiredPerms.
173  *
174  *	  Returns the rangetable index of the target relation.
175  */
176 int
setTargetTable(ParseState * pstate,RangeVar * relation,bool inh,bool alsoSource,AclMode requiredPerms)177 setTargetTable(ParseState *pstate, RangeVar *relation,
178 			   bool inh, bool alsoSource, AclMode requiredPerms)
179 {
180 	ParseNamespaceItem *nsitem;
181 
182 	/*
183 	 * ENRs hide tables of the same name, so we need to check for them first.
184 	 * In contrast, CTEs don't hide tables (for this purpose).
185 	 */
186 	if (relation->schemaname == NULL &&
187 		scanNameSpaceForENR(pstate, relation->relname))
188 		ereport(ERROR,
189 				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
190 				 errmsg("relation \"%s\" cannot be the target of a modifying statement",
191 						relation->relname)));
192 
193 	/* Close old target; this could only happen for multi-action rules */
194 	if (pstate->p_target_relation != NULL)
195 		table_close(pstate->p_target_relation, NoLock);
196 
197 	/*
198 	 * Open target rel and grab suitable lock (which we will hold till end of
199 	 * transaction).
200 	 *
201 	 * free_parsestate() will eventually do the corresponding table_close(),
202 	 * but *not* release the lock.
203 	 */
204 	pstate->p_target_relation = parserOpenTable(pstate, relation,
205 												RowExclusiveLock);
206 
207 	/*
208 	 * Now build an RTE and a ParseNamespaceItem.
209 	 */
210 	nsitem = addRangeTableEntryForRelation(pstate, pstate->p_target_relation,
211 										   RowExclusiveLock,
212 										   relation->alias, inh, false);
213 
214 	/* remember the RTE/nsitem as being the query target */
215 	pstate->p_target_nsitem = nsitem;
216 
217 	/*
218 	 * Override addRangeTableEntry's default ACL_SELECT permissions check, and
219 	 * instead mark target table as requiring exactly the specified
220 	 * permissions.
221 	 *
222 	 * If we find an explicit reference to the rel later during parse
223 	 * analysis, we will add the ACL_SELECT bit back again; see
224 	 * markVarForSelectPriv and its callers.
225 	 */
226 	nsitem->p_rte->requiredPerms = requiredPerms;
227 
228 	/*
229 	 * If UPDATE/DELETE, add table to joinlist and namespace.
230 	 */
231 	if (alsoSource)
232 		addNSItemToQuery(pstate, nsitem, true, true, true);
233 
234 	return nsitem->p_rtindex;
235 }
236 
237 /*
238  * Extract all not-in-common columns from column lists of a source table
239  *
240  * src_nscolumns and src_colnames describe the source table.
241  *
242  * *src_colnos initially contains the column numbers of the already-merged
243  * columns.  We add to it the column number of each additional column.
244  * Also append to *res_colnames the name of each additional column,
245  * append to *res_colvars a Var for each additional column, and copy the
246  * columns' nscolumns data into res_nscolumns[] (which is caller-allocated
247  * space that had better be big enough).
248  *
249  * Returns the number of columns added.
250  */
251 static int
extractRemainingColumns(ParseNamespaceColumn * src_nscolumns,List * src_colnames,List ** src_colnos,List ** res_colnames,List ** res_colvars,ParseNamespaceColumn * res_nscolumns)252 extractRemainingColumns(ParseNamespaceColumn *src_nscolumns,
253 						List *src_colnames,
254 						List **src_colnos,
255 						List **res_colnames, List **res_colvars,
256 						ParseNamespaceColumn *res_nscolumns)
257 {
258 	int			colcount = 0;
259 	Bitmapset  *prevcols;
260 	int			attnum;
261 	ListCell   *lc;
262 
263 	/*
264 	 * While we could just test "list_member_int(*src_colnos, attnum)" to
265 	 * detect already-merged columns in the loop below, that would be O(N^2)
266 	 * for a wide input table.  Instead build a bitmapset of just the merged
267 	 * USING columns, which we won't add to within the main loop.
268 	 */
269 	prevcols = NULL;
270 	foreach(lc, *src_colnos)
271 	{
272 		prevcols = bms_add_member(prevcols, lfirst_int(lc));
273 	}
274 
275 	attnum = 0;
276 	foreach(lc, src_colnames)
277 	{
278 		char	   *colname = strVal(lfirst(lc));
279 
280 		attnum++;
281 		/* Non-dropped and not already merged? */
282 		if (colname[0] != '\0' && !bms_is_member(attnum, prevcols))
283 		{
284 			/* Yes, so emit it as next output column */
285 			*src_colnos = lappend_int(*src_colnos, attnum);
286 			*res_colnames = lappend(*res_colnames, lfirst(lc));
287 			*res_colvars = lappend(*res_colvars,
288 								   buildVarFromNSColumn(src_nscolumns + attnum - 1));
289 			/* Copy the input relation's nscolumn data for this column */
290 			res_nscolumns[colcount] = src_nscolumns[attnum - 1];
291 			colcount++;
292 		}
293 	}
294 	return colcount;
295 }
296 
297 /* transformJoinUsingClause()
298  *	  Build a complete ON clause from a partially-transformed USING list.
299  *	  We are given lists of nodes representing left and right match columns.
300  *	  Result is a transformed qualification expression.
301  */
302 static Node *
transformJoinUsingClause(ParseState * pstate,List * leftVars,List * rightVars)303 transformJoinUsingClause(ParseState *pstate,
304 						 List *leftVars, List *rightVars)
305 {
306 	Node	   *result;
307 	List	   *andargs = NIL;
308 	ListCell   *lvars,
309 			   *rvars;
310 
311 	/*
312 	 * We cheat a little bit here by building an untransformed operator tree
313 	 * whose leaves are the already-transformed Vars.  This requires collusion
314 	 * from transformExpr(), which normally could be expected to complain
315 	 * about already-transformed subnodes.  However, this does mean that we
316 	 * have to mark the columns as requiring SELECT privilege for ourselves;
317 	 * transformExpr() won't do it.
318 	 */
319 	forboth(lvars, leftVars, rvars, rightVars)
320 	{
321 		Var		   *lvar = (Var *) lfirst(lvars);
322 		Var		   *rvar = (Var *) lfirst(rvars);
323 		A_Expr	   *e;
324 
325 		/* Require read access to the join variables */
326 		markVarForSelectPriv(pstate, lvar);
327 		markVarForSelectPriv(pstate, rvar);
328 
329 		/* Now create the lvar = rvar join condition */
330 		e = makeSimpleA_Expr(AEXPR_OP, "=",
331 							 (Node *) copyObject(lvar), (Node *) copyObject(rvar),
332 							 -1);
333 
334 		/* Prepare to combine into an AND clause, if multiple join columns */
335 		andargs = lappend(andargs, e);
336 	}
337 
338 	/* Only need an AND if there's more than one join column */
339 	if (list_length(andargs) == 1)
340 		result = (Node *) linitial(andargs);
341 	else
342 		result = (Node *) makeBoolExpr(AND_EXPR, andargs, -1);
343 
344 	/*
345 	 * Since the references are already Vars, and are certainly from the input
346 	 * relations, we don't have to go through the same pushups that
347 	 * transformJoinOnClause() does.  Just invoke transformExpr() to fix up
348 	 * the operators, and we're done.
349 	 */
350 	result = transformExpr(pstate, result, EXPR_KIND_JOIN_USING);
351 
352 	result = coerce_to_boolean(pstate, result, "JOIN/USING");
353 
354 	return result;
355 }
356 
357 /* transformJoinOnClause()
358  *	  Transform the qual conditions for JOIN/ON.
359  *	  Result is a transformed qualification expression.
360  */
361 static Node *
transformJoinOnClause(ParseState * pstate,JoinExpr * j,List * namespace)362 transformJoinOnClause(ParseState *pstate, JoinExpr *j, List *namespace)
363 {
364 	Node	   *result;
365 	List	   *save_namespace;
366 
367 	/*
368 	 * The namespace that the join expression should see is just the two
369 	 * subtrees of the JOIN plus any outer references from upper pstate
370 	 * levels.  Temporarily set this pstate's namespace accordingly.  (We need
371 	 * not check for refname conflicts, because transformFromClauseItem()
372 	 * already did.)  All namespace items are marked visible regardless of
373 	 * LATERAL state.
374 	 */
375 	setNamespaceLateralState(namespace, false, true);
376 
377 	save_namespace = pstate->p_namespace;
378 	pstate->p_namespace = namespace;
379 
380 	result = transformWhereClause(pstate, j->quals,
381 								  EXPR_KIND_JOIN_ON, "JOIN/ON");
382 
383 	pstate->p_namespace = save_namespace;
384 
385 	return result;
386 }
387 
388 /*
389  * transformTableEntry --- transform a RangeVar (simple relation reference)
390  */
391 static ParseNamespaceItem *
transformTableEntry(ParseState * pstate,RangeVar * r)392 transformTableEntry(ParseState *pstate, RangeVar *r)
393 {
394 	/* addRangeTableEntry does all the work */
395 	return addRangeTableEntry(pstate, r, r->alias, r->inh, true);
396 }
397 
398 /*
399  * transformRangeSubselect --- transform a sub-SELECT appearing in FROM
400  */
401 static ParseNamespaceItem *
transformRangeSubselect(ParseState * pstate,RangeSubselect * r)402 transformRangeSubselect(ParseState *pstate, RangeSubselect *r)
403 {
404 	Query	   *query;
405 
406 	/*
407 	 * We require user to supply an alias for a subselect, per SQL92. To relax
408 	 * this, we'd have to be prepared to gin up a unique alias for an
409 	 * unlabeled subselect.  (This is just elog, not ereport, because the
410 	 * grammar should have enforced it already.  It'd probably be better to
411 	 * report the error here, but we don't have a good error location here.)
412 	 */
413 	if (r->alias == NULL)
414 		elog(ERROR, "subquery in FROM must have an alias");
415 
416 	/*
417 	 * Set p_expr_kind to show this parse level is recursing to a subselect.
418 	 * We can't be nested within any expression, so don't need save-restore
419 	 * logic here.
420 	 */
421 	Assert(pstate->p_expr_kind == EXPR_KIND_NONE);
422 	pstate->p_expr_kind = EXPR_KIND_FROM_SUBSELECT;
423 
424 	/*
425 	 * If the subselect is LATERAL, make lateral_only names of this level
426 	 * visible to it.  (LATERAL can't nest within a single pstate level, so we
427 	 * don't need save/restore logic here.)
428 	 */
429 	Assert(!pstate->p_lateral_active);
430 	pstate->p_lateral_active = r->lateral;
431 
432 	/*
433 	 * Analyze and transform the subquery.
434 	 */
435 	query = parse_sub_analyze(r->subquery, pstate, NULL,
436 							  isLockedRefname(pstate, r->alias->aliasname),
437 							  true);
438 
439 	/* Restore state */
440 	pstate->p_lateral_active = false;
441 	pstate->p_expr_kind = EXPR_KIND_NONE;
442 
443 	/*
444 	 * Check that we got a SELECT.  Anything else should be impossible given
445 	 * restrictions of the grammar, but check anyway.
446 	 */
447 	if (!IsA(query, Query) ||
448 		query->commandType != CMD_SELECT)
449 		elog(ERROR, "unexpected non-SELECT command in subquery in FROM");
450 
451 	/*
452 	 * OK, build an RTE and nsitem for the subquery.
453 	 */
454 	return addRangeTableEntryForSubquery(pstate,
455 										 query,
456 										 r->alias,
457 										 r->lateral,
458 										 true);
459 }
460 
461 
462 /*
463  * transformRangeFunction --- transform a function call appearing in FROM
464  */
465 static ParseNamespaceItem *
transformRangeFunction(ParseState * pstate,RangeFunction * r)466 transformRangeFunction(ParseState *pstate, RangeFunction *r)
467 {
468 	List	   *funcexprs = NIL;
469 	List	   *funcnames = NIL;
470 	List	   *coldeflists = NIL;
471 	bool		is_lateral;
472 	ListCell   *lc;
473 
474 	/*
475 	 * We make lateral_only names of this level visible, whether or not the
476 	 * RangeFunction is explicitly marked LATERAL.  This is needed for SQL
477 	 * spec compliance in the case of UNNEST(), and seems useful on
478 	 * convenience grounds for all functions in FROM.
479 	 *
480 	 * (LATERAL can't nest within a single pstate level, so we don't need
481 	 * save/restore logic here.)
482 	 */
483 	Assert(!pstate->p_lateral_active);
484 	pstate->p_lateral_active = true;
485 
486 	/*
487 	 * Transform the raw expressions.
488 	 *
489 	 * While transforming, also save function names for possible use as alias
490 	 * and column names.  We use the same transformation rules as for a SELECT
491 	 * output expression.  For a FuncCall node, the result will be the
492 	 * function name, but it is possible for the grammar to hand back other
493 	 * node types.
494 	 *
495 	 * We have to get this info now, because FigureColname only works on raw
496 	 * parsetrees.  Actually deciding what to do with the names is left up to
497 	 * addRangeTableEntryForFunction.
498 	 *
499 	 * Likewise, collect column definition lists if there were any.  But
500 	 * complain if we find one here and the RangeFunction has one too.
501 	 */
502 	foreach(lc, r->functions)
503 	{
504 		List	   *pair = (List *) lfirst(lc);
505 		Node	   *fexpr;
506 		List	   *coldeflist;
507 		Node	   *newfexpr;
508 		Node	   *last_srf;
509 
510 		/* Disassemble the function-call/column-def-list pairs */
511 		Assert(list_length(pair) == 2);
512 		fexpr = (Node *) linitial(pair);
513 		coldeflist = (List *) lsecond(pair);
514 
515 		/*
516 		 * If we find a function call unnest() with more than one argument and
517 		 * no special decoration, transform it into separate unnest() calls on
518 		 * each argument.  This is a kluge, for sure, but it's less nasty than
519 		 * other ways of implementing the SQL-standard UNNEST() syntax.
520 		 *
521 		 * If there is any decoration (including a coldeflist), we don't
522 		 * transform, which probably means a no-such-function error later.  We
523 		 * could alternatively throw an error right now, but that doesn't seem
524 		 * tremendously helpful.  If someone is using any such decoration,
525 		 * then they're not using the SQL-standard syntax, and they're more
526 		 * likely expecting an un-tweaked function call.
527 		 *
528 		 * Note: the transformation changes a non-schema-qualified unnest()
529 		 * function name into schema-qualified pg_catalog.unnest().  This
530 		 * choice is also a bit debatable, but it seems reasonable to force
531 		 * use of built-in unnest() when we make this transformation.
532 		 */
533 		if (IsA(fexpr, FuncCall))
534 		{
535 			FuncCall   *fc = (FuncCall *) fexpr;
536 
537 			if (list_length(fc->funcname) == 1 &&
538 				strcmp(strVal(linitial(fc->funcname)), "unnest") == 0 &&
539 				list_length(fc->args) > 1 &&
540 				fc->agg_order == NIL &&
541 				fc->agg_filter == NULL &&
542 				fc->over == NULL &&
543 				!fc->agg_star &&
544 				!fc->agg_distinct &&
545 				!fc->func_variadic &&
546 				coldeflist == NIL)
547 			{
548 				ListCell   *lc;
549 
550 				foreach(lc, fc->args)
551 				{
552 					Node	   *arg = (Node *) lfirst(lc);
553 					FuncCall   *newfc;
554 
555 					last_srf = pstate->p_last_srf;
556 
557 					newfc = makeFuncCall(SystemFuncName("unnest"),
558 										 list_make1(arg),
559 										 COERCE_EXPLICIT_CALL,
560 										 fc->location);
561 
562 					newfexpr = transformExpr(pstate, (Node *) newfc,
563 											 EXPR_KIND_FROM_FUNCTION);
564 
565 					/* nodeFunctionscan.c requires SRFs to be at top level */
566 					if (pstate->p_last_srf != last_srf &&
567 						pstate->p_last_srf != newfexpr)
568 						ereport(ERROR,
569 								(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
570 								 errmsg("set-returning functions must appear at top level of FROM"),
571 								 parser_errposition(pstate,
572 													exprLocation(pstate->p_last_srf))));
573 
574 					funcexprs = lappend(funcexprs, newfexpr);
575 
576 					funcnames = lappend(funcnames,
577 										FigureColname((Node *) newfc));
578 
579 					/* coldeflist is empty, so no error is possible */
580 
581 					coldeflists = lappend(coldeflists, coldeflist);
582 				}
583 				continue;		/* done with this function item */
584 			}
585 		}
586 
587 		/* normal case ... */
588 		last_srf = pstate->p_last_srf;
589 
590 		newfexpr = transformExpr(pstate, fexpr,
591 								 EXPR_KIND_FROM_FUNCTION);
592 
593 		/* nodeFunctionscan.c requires SRFs to be at top level */
594 		if (pstate->p_last_srf != last_srf &&
595 			pstate->p_last_srf != newfexpr)
596 			ereport(ERROR,
597 					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
598 					 errmsg("set-returning functions must appear at top level of FROM"),
599 					 parser_errposition(pstate,
600 										exprLocation(pstate->p_last_srf))));
601 
602 		funcexprs = lappend(funcexprs, newfexpr);
603 
604 		funcnames = lappend(funcnames,
605 							FigureColname(fexpr));
606 
607 		if (coldeflist && r->coldeflist)
608 			ereport(ERROR,
609 					(errcode(ERRCODE_SYNTAX_ERROR),
610 					 errmsg("multiple column definition lists are not allowed for the same function"),
611 					 parser_errposition(pstate,
612 										exprLocation((Node *) r->coldeflist))));
613 
614 		coldeflists = lappend(coldeflists, coldeflist);
615 	}
616 
617 	pstate->p_lateral_active = false;
618 
619 	/*
620 	 * We must assign collations now so that the RTE exposes correct collation
621 	 * info for Vars created from it.
622 	 */
623 	assign_list_collations(pstate, funcexprs);
624 
625 	/*
626 	 * Install the top-level coldeflist if there was one (we already checked
627 	 * that there was no conflicting per-function coldeflist).
628 	 *
629 	 * We only allow this when there's a single function (even after UNNEST
630 	 * expansion) and no WITH ORDINALITY.  The reason for the latter
631 	 * restriction is that it's not real clear whether the ordinality column
632 	 * should be in the coldeflist, and users are too likely to make mistakes
633 	 * in one direction or the other.  Putting the coldeflist inside ROWS
634 	 * FROM() is much clearer in this case.
635 	 */
636 	if (r->coldeflist)
637 	{
638 		if (list_length(funcexprs) != 1)
639 		{
640 			if (r->is_rowsfrom)
641 				ereport(ERROR,
642 						(errcode(ERRCODE_SYNTAX_ERROR),
643 						 errmsg("ROWS FROM() with multiple functions cannot have a column definition list"),
644 						 errhint("Put a separate column definition list for each function inside ROWS FROM()."),
645 						 parser_errposition(pstate,
646 											exprLocation((Node *) r->coldeflist))));
647 			else
648 				ereport(ERROR,
649 						(errcode(ERRCODE_SYNTAX_ERROR),
650 						 errmsg("UNNEST() with multiple arguments cannot have a column definition list"),
651 						 errhint("Use separate UNNEST() calls inside ROWS FROM(), and attach a column definition list to each one."),
652 						 parser_errposition(pstate,
653 											exprLocation((Node *) r->coldeflist))));
654 		}
655 		if (r->ordinality)
656 			ereport(ERROR,
657 					(errcode(ERRCODE_SYNTAX_ERROR),
658 					 errmsg("WITH ORDINALITY cannot be used with a column definition list"),
659 					 errhint("Put the column definition list inside ROWS FROM()."),
660 					 parser_errposition(pstate,
661 										exprLocation((Node *) r->coldeflist))));
662 
663 		coldeflists = list_make1(r->coldeflist);
664 	}
665 
666 	/*
667 	 * Mark the RTE as LATERAL if the user said LATERAL explicitly, or if
668 	 * there are any lateral cross-references in it.
669 	 */
670 	is_lateral = r->lateral || contain_vars_of_level((Node *) funcexprs, 0);
671 
672 	/*
673 	 * OK, build an RTE and nsitem for the function.
674 	 */
675 	return addRangeTableEntryForFunction(pstate,
676 										 funcnames, funcexprs, coldeflists,
677 										 r, is_lateral, true);
678 }
679 
680 /*
681  * transformRangeTableFunc -
682  *			Transform a raw RangeTableFunc into TableFunc.
683  *
684  * Transform the namespace clauses, the document-generating expression, the
685  * row-generating expression, the column-generating expressions, and the
686  * default value expressions.
687  */
688 static ParseNamespaceItem *
transformRangeTableFunc(ParseState * pstate,RangeTableFunc * rtf)689 transformRangeTableFunc(ParseState *pstate, RangeTableFunc *rtf)
690 {
691 	TableFunc  *tf = makeNode(TableFunc);
692 	const char *constructName;
693 	Oid			docType;
694 	bool		is_lateral;
695 	ListCell   *col;
696 	char	  **names;
697 	int			colno;
698 
699 	/* Currently only XMLTABLE is supported */
700 	constructName = "XMLTABLE";
701 	docType = XMLOID;
702 
703 	/*
704 	 * We make lateral_only names of this level visible, whether or not the
705 	 * RangeTableFunc is explicitly marked LATERAL.  This is needed for SQL
706 	 * spec compliance and seems useful on convenience grounds for all
707 	 * functions in FROM.
708 	 *
709 	 * (LATERAL can't nest within a single pstate level, so we don't need
710 	 * save/restore logic here.)
711 	 */
712 	Assert(!pstate->p_lateral_active);
713 	pstate->p_lateral_active = true;
714 
715 	/* Transform and apply typecast to the row-generating expression ... */
716 	Assert(rtf->rowexpr != NULL);
717 	tf->rowexpr = coerce_to_specific_type(pstate,
718 										  transformExpr(pstate, rtf->rowexpr, EXPR_KIND_FROM_FUNCTION),
719 										  TEXTOID,
720 										  constructName);
721 	assign_expr_collations(pstate, tf->rowexpr);
722 
723 	/* ... and to the document itself */
724 	Assert(rtf->docexpr != NULL);
725 	tf->docexpr = coerce_to_specific_type(pstate,
726 										  transformExpr(pstate, rtf->docexpr, EXPR_KIND_FROM_FUNCTION),
727 										  docType,
728 										  constructName);
729 	assign_expr_collations(pstate, tf->docexpr);
730 
731 	/* undef ordinality column number */
732 	tf->ordinalitycol = -1;
733 
734 	/* Process column specs */
735 	names = palloc(sizeof(char *) * list_length(rtf->columns));
736 
737 	colno = 0;
738 	foreach(col, rtf->columns)
739 	{
740 		RangeTableFuncCol *rawc = (RangeTableFuncCol *) lfirst(col);
741 		Oid			typid;
742 		int32		typmod;
743 		Node	   *colexpr;
744 		Node	   *coldefexpr;
745 		int			j;
746 
747 		tf->colnames = lappend(tf->colnames,
748 							   makeString(pstrdup(rawc->colname)));
749 
750 		/*
751 		 * Determine the type and typmod for the new column. FOR ORDINALITY
752 		 * columns are INTEGER per spec; the others are user-specified.
753 		 */
754 		if (rawc->for_ordinality)
755 		{
756 			if (tf->ordinalitycol != -1)
757 				ereport(ERROR,
758 						(errcode(ERRCODE_SYNTAX_ERROR),
759 						 errmsg("only one FOR ORDINALITY column is allowed"),
760 						 parser_errposition(pstate, rawc->location)));
761 
762 			typid = INT4OID;
763 			typmod = -1;
764 			tf->ordinalitycol = colno;
765 		}
766 		else
767 		{
768 			if (rawc->typeName->setof)
769 				ereport(ERROR,
770 						(errcode(ERRCODE_INVALID_TABLE_DEFINITION),
771 						 errmsg("column \"%s\" cannot be declared SETOF",
772 								rawc->colname),
773 						 parser_errposition(pstate, rawc->location)));
774 
775 			typenameTypeIdAndMod(pstate, rawc->typeName,
776 								 &typid, &typmod);
777 		}
778 
779 		tf->coltypes = lappend_oid(tf->coltypes, typid);
780 		tf->coltypmods = lappend_int(tf->coltypmods, typmod);
781 		tf->colcollations = lappend_oid(tf->colcollations,
782 										get_typcollation(typid));
783 
784 		/* Transform the PATH and DEFAULT expressions */
785 		if (rawc->colexpr)
786 		{
787 			colexpr = coerce_to_specific_type(pstate,
788 											  transformExpr(pstate, rawc->colexpr,
789 															EXPR_KIND_FROM_FUNCTION),
790 											  TEXTOID,
791 											  constructName);
792 			assign_expr_collations(pstate, colexpr);
793 		}
794 		else
795 			colexpr = NULL;
796 
797 		if (rawc->coldefexpr)
798 		{
799 			coldefexpr = coerce_to_specific_type_typmod(pstate,
800 														transformExpr(pstate, rawc->coldefexpr,
801 																	  EXPR_KIND_FROM_FUNCTION),
802 														typid, typmod,
803 														constructName);
804 			assign_expr_collations(pstate, coldefexpr);
805 		}
806 		else
807 			coldefexpr = NULL;
808 
809 		tf->colexprs = lappend(tf->colexprs, colexpr);
810 		tf->coldefexprs = lappend(tf->coldefexprs, coldefexpr);
811 
812 		if (rawc->is_not_null)
813 			tf->notnulls = bms_add_member(tf->notnulls, colno);
814 
815 		/* make sure column names are unique */
816 		for (j = 0; j < colno; j++)
817 			if (strcmp(names[j], rawc->colname) == 0)
818 				ereport(ERROR,
819 						(errcode(ERRCODE_SYNTAX_ERROR),
820 						 errmsg("column name \"%s\" is not unique",
821 								rawc->colname),
822 						 parser_errposition(pstate, rawc->location)));
823 		names[colno] = rawc->colname;
824 
825 		colno++;
826 	}
827 	pfree(names);
828 
829 	/* Namespaces, if any, also need to be transformed */
830 	if (rtf->namespaces != NIL)
831 	{
832 		ListCell   *ns;
833 		ListCell   *lc2;
834 		List	   *ns_uris = NIL;
835 		List	   *ns_names = NIL;
836 		bool		default_ns_seen = false;
837 
838 		foreach(ns, rtf->namespaces)
839 		{
840 			ResTarget  *r = (ResTarget *) lfirst(ns);
841 			Node	   *ns_uri;
842 
843 			Assert(IsA(r, ResTarget));
844 			ns_uri = transformExpr(pstate, r->val, EXPR_KIND_FROM_FUNCTION);
845 			ns_uri = coerce_to_specific_type(pstate, ns_uri,
846 											 TEXTOID, constructName);
847 			assign_expr_collations(pstate, ns_uri);
848 			ns_uris = lappend(ns_uris, ns_uri);
849 
850 			/* Verify consistency of name list: no dupes, only one DEFAULT */
851 			if (r->name != NULL)
852 			{
853 				foreach(lc2, ns_names)
854 				{
855 					Value	   *ns_node = (Value *) lfirst(lc2);
856 
857 					if (ns_node == NULL)
858 						continue;
859 					if (strcmp(strVal(ns_node), r->name) == 0)
860 						ereport(ERROR,
861 								(errcode(ERRCODE_SYNTAX_ERROR),
862 								 errmsg("namespace name \"%s\" is not unique",
863 										r->name),
864 								 parser_errposition(pstate, r->location)));
865 				}
866 			}
867 			else
868 			{
869 				if (default_ns_seen)
870 					ereport(ERROR,
871 							(errcode(ERRCODE_SYNTAX_ERROR),
872 							 errmsg("only one default namespace is allowed"),
873 							 parser_errposition(pstate, r->location)));
874 				default_ns_seen = true;
875 			}
876 
877 			/* We represent DEFAULT by a null pointer */
878 			ns_names = lappend(ns_names,
879 							   r->name ? makeString(r->name) : NULL);
880 		}
881 
882 		tf->ns_uris = ns_uris;
883 		tf->ns_names = ns_names;
884 	}
885 
886 	tf->location = rtf->location;
887 
888 	pstate->p_lateral_active = false;
889 
890 	/*
891 	 * Mark the RTE as LATERAL if the user said LATERAL explicitly, or if
892 	 * there are any lateral cross-references in it.
893 	 */
894 	is_lateral = rtf->lateral || contain_vars_of_level((Node *) tf, 0);
895 
896 	return addRangeTableEntryForTableFunc(pstate,
897 										  tf, rtf->alias, is_lateral, true);
898 }
899 
900 /*
901  * transformRangeTableSample --- transform a TABLESAMPLE clause
902  *
903  * Caller has already transformed rts->relation, we just have to validate
904  * the remaining fields and create a TableSampleClause node.
905  */
906 static TableSampleClause *
transformRangeTableSample(ParseState * pstate,RangeTableSample * rts)907 transformRangeTableSample(ParseState *pstate, RangeTableSample *rts)
908 {
909 	TableSampleClause *tablesample;
910 	Oid			handlerOid;
911 	Oid			funcargtypes[1];
912 	TsmRoutine *tsm;
913 	List	   *fargs;
914 	ListCell   *larg,
915 			   *ltyp;
916 
917 	/*
918 	 * To validate the sample method name, look up the handler function, which
919 	 * has the same name, one dummy INTERNAL argument, and a result type of
920 	 * tsm_handler.  (Note: tablesample method names are not schema-qualified
921 	 * in the SQL standard; but since they are just functions to us, we allow
922 	 * schema qualification to resolve any potential ambiguity.)
923 	 */
924 	funcargtypes[0] = INTERNALOID;
925 
926 	handlerOid = LookupFuncName(rts->method, 1, funcargtypes, true);
927 
928 	/* we want error to complain about no-such-method, not no-such-function */
929 	if (!OidIsValid(handlerOid))
930 		ereport(ERROR,
931 				(errcode(ERRCODE_UNDEFINED_OBJECT),
932 				 errmsg("tablesample method %s does not exist",
933 						NameListToString(rts->method)),
934 				 parser_errposition(pstate, rts->location)));
935 
936 	/* check that handler has correct return type */
937 	if (get_func_rettype(handlerOid) != TSM_HANDLEROID)
938 		ereport(ERROR,
939 				(errcode(ERRCODE_WRONG_OBJECT_TYPE),
940 				 errmsg("function %s must return type %s",
941 						NameListToString(rts->method), "tsm_handler"),
942 				 parser_errposition(pstate, rts->location)));
943 
944 	/* OK, run the handler to get TsmRoutine, for argument type info */
945 	tsm = GetTsmRoutine(handlerOid);
946 
947 	tablesample = makeNode(TableSampleClause);
948 	tablesample->tsmhandler = handlerOid;
949 
950 	/* check user provided the expected number of arguments */
951 	if (list_length(rts->args) != list_length(tsm->parameterTypes))
952 		ereport(ERROR,
953 				(errcode(ERRCODE_INVALID_TABLESAMPLE_ARGUMENT),
954 				 errmsg_plural("tablesample method %s requires %d argument, not %d",
955 							   "tablesample method %s requires %d arguments, not %d",
956 							   list_length(tsm->parameterTypes),
957 							   NameListToString(rts->method),
958 							   list_length(tsm->parameterTypes),
959 							   list_length(rts->args)),
960 				 parser_errposition(pstate, rts->location)));
961 
962 	/*
963 	 * Transform the arguments, typecasting them as needed.  Note we must also
964 	 * assign collations now, because assign_query_collations() doesn't
965 	 * examine any substructure of RTEs.
966 	 */
967 	fargs = NIL;
968 	forboth(larg, rts->args, ltyp, tsm->parameterTypes)
969 	{
970 		Node	   *arg = (Node *) lfirst(larg);
971 		Oid			argtype = lfirst_oid(ltyp);
972 
973 		arg = transformExpr(pstate, arg, EXPR_KIND_FROM_FUNCTION);
974 		arg = coerce_to_specific_type(pstate, arg, argtype, "TABLESAMPLE");
975 		assign_expr_collations(pstate, arg);
976 		fargs = lappend(fargs, arg);
977 	}
978 	tablesample->args = fargs;
979 
980 	/* Process REPEATABLE (seed) */
981 	if (rts->repeatable != NULL)
982 	{
983 		Node	   *arg;
984 
985 		if (!tsm->repeatable_across_queries)
986 			ereport(ERROR,
987 					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
988 					 errmsg("tablesample method %s does not support REPEATABLE",
989 							NameListToString(rts->method)),
990 					 parser_errposition(pstate, rts->location)));
991 
992 		arg = transformExpr(pstate, rts->repeatable, EXPR_KIND_FROM_FUNCTION);
993 		arg = coerce_to_specific_type(pstate, arg, FLOAT8OID, "REPEATABLE");
994 		assign_expr_collations(pstate, arg);
995 		tablesample->repeatable = (Expr *) arg;
996 	}
997 	else
998 		tablesample->repeatable = NULL;
999 
1000 	return tablesample;
1001 }
1002 
1003 /*
1004  * getNSItemForSpecialRelationTypes
1005  *
1006  * If given RangeVar refers to a CTE or an EphemeralNamedRelation,
1007  * build and return an appropriate ParseNamespaceItem, otherwise return NULL
1008  */
1009 static ParseNamespaceItem *
getNSItemForSpecialRelationTypes(ParseState * pstate,RangeVar * rv)1010 getNSItemForSpecialRelationTypes(ParseState *pstate, RangeVar *rv)
1011 {
1012 	ParseNamespaceItem *nsitem;
1013 	CommonTableExpr *cte;
1014 	Index		levelsup;
1015 
1016 	/*
1017 	 * if it is a qualified name, it can't be a CTE or tuplestore reference
1018 	 */
1019 	if (rv->schemaname)
1020 		return NULL;
1021 
1022 	cte = scanNameSpaceForCTE(pstate, rv->relname, &levelsup);
1023 	if (cte)
1024 		nsitem = addRangeTableEntryForCTE(pstate, cte, levelsup, rv, true);
1025 	else if (scanNameSpaceForENR(pstate, rv->relname))
1026 		nsitem = addRangeTableEntryForENR(pstate, rv, true);
1027 	else
1028 		nsitem = NULL;
1029 
1030 	return nsitem;
1031 }
1032 
1033 /*
1034  * transformFromClauseItem -
1035  *	  Transform a FROM-clause item, adding any required entries to the
1036  *	  range table list being built in the ParseState, and return the
1037  *	  transformed item ready to include in the joinlist.  Also build a
1038  *	  ParseNamespaceItem list describing the names exposed by this item.
1039  *	  This routine can recurse to handle SQL92 JOIN expressions.
1040  *
1041  * The function return value is the node to add to the jointree (a
1042  * RangeTblRef or JoinExpr).  Additional output parameters are:
1043  *
1044  * *top_nsitem: receives the ParseNamespaceItem directly corresponding to the
1045  * jointree item.  (This is only used during internal recursion, not by
1046  * outside callers.)
1047  *
1048  * *namespace: receives a List of ParseNamespaceItems for the RTEs exposed
1049  * as table/column names by this item.  (The lateral_only flags in these items
1050  * are indeterminate and should be explicitly set by the caller before use.)
1051  */
1052 static Node *
transformFromClauseItem(ParseState * pstate,Node * n,ParseNamespaceItem ** top_nsitem,List ** namespace)1053 transformFromClauseItem(ParseState *pstate, Node *n,
1054 						ParseNamespaceItem **top_nsitem,
1055 						List **namespace)
1056 {
1057 	if (IsA(n, RangeVar))
1058 	{
1059 		/* Plain relation reference, or perhaps a CTE reference */
1060 		RangeVar   *rv = (RangeVar *) n;
1061 		RangeTblRef *rtr;
1062 		ParseNamespaceItem *nsitem;
1063 
1064 		/* Check if it's a CTE or tuplestore reference */
1065 		nsitem = getNSItemForSpecialRelationTypes(pstate, rv);
1066 
1067 		/* if not found above, must be a table reference */
1068 		if (!nsitem)
1069 			nsitem = transformTableEntry(pstate, rv);
1070 
1071 		*top_nsitem = nsitem;
1072 		*namespace = list_make1(nsitem);
1073 		rtr = makeNode(RangeTblRef);
1074 		rtr->rtindex = nsitem->p_rtindex;
1075 		return (Node *) rtr;
1076 	}
1077 	else if (IsA(n, RangeSubselect))
1078 	{
1079 		/* sub-SELECT is like a plain relation */
1080 		RangeTblRef *rtr;
1081 		ParseNamespaceItem *nsitem;
1082 
1083 		nsitem = transformRangeSubselect(pstate, (RangeSubselect *) n);
1084 		*top_nsitem = nsitem;
1085 		*namespace = list_make1(nsitem);
1086 		rtr = makeNode(RangeTblRef);
1087 		rtr->rtindex = nsitem->p_rtindex;
1088 		return (Node *) rtr;
1089 	}
1090 	else if (IsA(n, RangeFunction))
1091 	{
1092 		/* function is like a plain relation */
1093 		RangeTblRef *rtr;
1094 		ParseNamespaceItem *nsitem;
1095 
1096 		nsitem = transformRangeFunction(pstate, (RangeFunction *) n);
1097 		*top_nsitem = nsitem;
1098 		*namespace = list_make1(nsitem);
1099 		rtr = makeNode(RangeTblRef);
1100 		rtr->rtindex = nsitem->p_rtindex;
1101 		return (Node *) rtr;
1102 	}
1103 	else if (IsA(n, RangeTableFunc))
1104 	{
1105 		/* table function is like a plain relation */
1106 		RangeTblRef *rtr;
1107 		ParseNamespaceItem *nsitem;
1108 
1109 		nsitem = transformRangeTableFunc(pstate, (RangeTableFunc *) n);
1110 		*top_nsitem = nsitem;
1111 		*namespace = list_make1(nsitem);
1112 		rtr = makeNode(RangeTblRef);
1113 		rtr->rtindex = nsitem->p_rtindex;
1114 		return (Node *) rtr;
1115 	}
1116 	else if (IsA(n, RangeTableSample))
1117 	{
1118 		/* TABLESAMPLE clause (wrapping some other valid FROM node) */
1119 		RangeTableSample *rts = (RangeTableSample *) n;
1120 		Node	   *rel;
1121 		RangeTblEntry *rte;
1122 
1123 		/* Recursively transform the contained relation */
1124 		rel = transformFromClauseItem(pstate, rts->relation,
1125 									  top_nsitem, namespace);
1126 		rte = (*top_nsitem)->p_rte;
1127 		/* We only support this on plain relations and matviews */
1128 		if (rte->rtekind != RTE_RELATION ||
1129 			(rte->relkind != RELKIND_RELATION &&
1130 			 rte->relkind != RELKIND_MATVIEW &&
1131 			 rte->relkind != RELKIND_PARTITIONED_TABLE))
1132 			ereport(ERROR,
1133 					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1134 					 errmsg("TABLESAMPLE clause can only be applied to tables and materialized views"),
1135 					 parser_errposition(pstate, exprLocation(rts->relation))));
1136 
1137 		/* Transform TABLESAMPLE details and attach to the RTE */
1138 		rte->tablesample = transformRangeTableSample(pstate, rts);
1139 		return rel;
1140 	}
1141 	else if (IsA(n, JoinExpr))
1142 	{
1143 		/* A newfangled join expression */
1144 		JoinExpr   *j = (JoinExpr *) n;
1145 		ParseNamespaceItem *nsitem;
1146 		ParseNamespaceItem *l_nsitem;
1147 		ParseNamespaceItem *r_nsitem;
1148 		List	   *l_namespace,
1149 				   *r_namespace,
1150 				   *my_namespace,
1151 				   *l_colnames,
1152 				   *r_colnames,
1153 				   *res_colnames,
1154 				   *l_colnos,
1155 				   *r_colnos,
1156 				   *res_colvars;
1157 		ParseNamespaceColumn *l_nscolumns,
1158 				   *r_nscolumns,
1159 				   *res_nscolumns;
1160 		int			res_colindex;
1161 		bool		lateral_ok;
1162 		int			sv_namespace_length;
1163 		int			k;
1164 
1165 		/*
1166 		 * Recursively process the left subtree, then the right.  We must do
1167 		 * it in this order for correct visibility of LATERAL references.
1168 		 */
1169 		j->larg = transformFromClauseItem(pstate, j->larg,
1170 										  &l_nsitem,
1171 										  &l_namespace);
1172 
1173 		/*
1174 		 * Make the left-side RTEs available for LATERAL access within the
1175 		 * right side, by temporarily adding them to the pstate's namespace
1176 		 * list.  Per SQL:2008, if the join type is not INNER or LEFT then the
1177 		 * left-side names must still be exposed, but it's an error to
1178 		 * reference them.  (Stupid design, but that's what it says.)  Hence,
1179 		 * we always push them into the namespace, but mark them as not
1180 		 * lateral_ok if the jointype is wrong.
1181 		 *
1182 		 * Notice that we don't require the merged namespace list to be
1183 		 * conflict-free.  See the comments for scanNameSpaceForRefname().
1184 		 */
1185 		lateral_ok = (j->jointype == JOIN_INNER || j->jointype == JOIN_LEFT);
1186 		setNamespaceLateralState(l_namespace, true, lateral_ok);
1187 
1188 		sv_namespace_length = list_length(pstate->p_namespace);
1189 		pstate->p_namespace = list_concat(pstate->p_namespace, l_namespace);
1190 
1191 		/* And now we can process the RHS */
1192 		j->rarg = transformFromClauseItem(pstate, j->rarg,
1193 										  &r_nsitem,
1194 										  &r_namespace);
1195 
1196 		/* Remove the left-side RTEs from the namespace list again */
1197 		pstate->p_namespace = list_truncate(pstate->p_namespace,
1198 											sv_namespace_length);
1199 
1200 		/*
1201 		 * Check for conflicting refnames in left and right subtrees. Must do
1202 		 * this because higher levels will assume I hand back a self-
1203 		 * consistent namespace list.
1204 		 */
1205 		checkNameSpaceConflicts(pstate, l_namespace, r_namespace);
1206 
1207 		/*
1208 		 * Generate combined namespace info for possible use below.
1209 		 */
1210 		my_namespace = list_concat(l_namespace, r_namespace);
1211 
1212 		/*
1213 		 * We'll work from the nscolumns data and eref alias column names for
1214 		 * each of the input nsitems.  Note that these include dropped
1215 		 * columns, which is helpful because we can keep track of physical
1216 		 * input column numbers more easily.
1217 		 */
1218 		l_nscolumns = l_nsitem->p_nscolumns;
1219 		l_colnames = l_nsitem->p_names->colnames;
1220 		r_nscolumns = r_nsitem->p_nscolumns;
1221 		r_colnames = r_nsitem->p_names->colnames;
1222 
1223 		/*
1224 		 * Natural join does not explicitly specify columns; must generate
1225 		 * columns to join. Need to run through the list of columns from each
1226 		 * table or join result and match up the column names. Use the first
1227 		 * table, and check every column in the second table for a match.
1228 		 * (We'll check that the matches were unique later on.) The result of
1229 		 * this step is a list of column names just like an explicitly-written
1230 		 * USING list.
1231 		 */
1232 		if (j->isNatural)
1233 		{
1234 			List	   *rlist = NIL;
1235 			ListCell   *lx,
1236 					   *rx;
1237 
1238 			Assert(j->usingClause == NIL);	/* shouldn't have USING() too */
1239 
1240 			foreach(lx, l_colnames)
1241 			{
1242 				char	   *l_colname = strVal(lfirst(lx));
1243 				Value	   *m_name = NULL;
1244 
1245 				if (l_colname[0] == '\0')
1246 					continue;	/* ignore dropped columns */
1247 
1248 				foreach(rx, r_colnames)
1249 				{
1250 					char	   *r_colname = strVal(lfirst(rx));
1251 
1252 					if (strcmp(l_colname, r_colname) == 0)
1253 					{
1254 						m_name = makeString(l_colname);
1255 						break;
1256 					}
1257 				}
1258 
1259 				/* matched a right column? then keep as join column... */
1260 				if (m_name != NULL)
1261 					rlist = lappend(rlist, m_name);
1262 			}
1263 
1264 			j->usingClause = rlist;
1265 		}
1266 
1267 		/*
1268 		 * If a USING clause alias was specified, save the USING columns as
1269 		 * its column list.
1270 		 */
1271 		if (j->join_using_alias)
1272 			j->join_using_alias->colnames = j->usingClause;
1273 
1274 		/*
1275 		 * Now transform the join qualifications, if any.
1276 		 */
1277 		l_colnos = NIL;
1278 		r_colnos = NIL;
1279 		res_colnames = NIL;
1280 		res_colvars = NIL;
1281 
1282 		/* this may be larger than needed, but it's not worth being exact */
1283 		res_nscolumns = (ParseNamespaceColumn *)
1284 			palloc0((list_length(l_colnames) + list_length(r_colnames)) *
1285 					sizeof(ParseNamespaceColumn));
1286 		res_colindex = 0;
1287 
1288 		if (j->usingClause)
1289 		{
1290 			/*
1291 			 * JOIN/USING (or NATURAL JOIN, as transformed above). Transform
1292 			 * the list into an explicit ON-condition, and generate a list of
1293 			 * merged result columns.
1294 			 */
1295 			List	   *ucols = j->usingClause;
1296 			List	   *l_usingvars = NIL;
1297 			List	   *r_usingvars = NIL;
1298 			ListCell   *ucol;
1299 
1300 			Assert(j->quals == NULL);	/* shouldn't have ON() too */
1301 
1302 			foreach(ucol, ucols)
1303 			{
1304 				char	   *u_colname = strVal(lfirst(ucol));
1305 				ListCell   *col;
1306 				int			ndx;
1307 				int			l_index = -1;
1308 				int			r_index = -1;
1309 				Var		   *l_colvar,
1310 						   *r_colvar;
1311 				Node	   *u_colvar;
1312 				ParseNamespaceColumn *res_nscolumn;
1313 
1314 				Assert(u_colname[0] != '\0');
1315 
1316 				/* Check for USING(foo,foo) */
1317 				foreach(col, res_colnames)
1318 				{
1319 					char	   *res_colname = strVal(lfirst(col));
1320 
1321 					if (strcmp(res_colname, u_colname) == 0)
1322 						ereport(ERROR,
1323 								(errcode(ERRCODE_DUPLICATE_COLUMN),
1324 								 errmsg("column name \"%s\" appears more than once in USING clause",
1325 										u_colname)));
1326 				}
1327 
1328 				/* Find it in left input */
1329 				ndx = 0;
1330 				foreach(col, l_colnames)
1331 				{
1332 					char	   *l_colname = strVal(lfirst(col));
1333 
1334 					if (strcmp(l_colname, u_colname) == 0)
1335 					{
1336 						if (l_index >= 0)
1337 							ereport(ERROR,
1338 									(errcode(ERRCODE_AMBIGUOUS_COLUMN),
1339 									 errmsg("common column name \"%s\" appears more than once in left table",
1340 											u_colname)));
1341 						l_index = ndx;
1342 					}
1343 					ndx++;
1344 				}
1345 				if (l_index < 0)
1346 					ereport(ERROR,
1347 							(errcode(ERRCODE_UNDEFINED_COLUMN),
1348 							 errmsg("column \"%s\" specified in USING clause does not exist in left table",
1349 									u_colname)));
1350 				l_colnos = lappend_int(l_colnos, l_index + 1);
1351 
1352 				/* Find it in right input */
1353 				ndx = 0;
1354 				foreach(col, r_colnames)
1355 				{
1356 					char	   *r_colname = strVal(lfirst(col));
1357 
1358 					if (strcmp(r_colname, u_colname) == 0)
1359 					{
1360 						if (r_index >= 0)
1361 							ereport(ERROR,
1362 									(errcode(ERRCODE_AMBIGUOUS_COLUMN),
1363 									 errmsg("common column name \"%s\" appears more than once in right table",
1364 											u_colname)));
1365 						r_index = ndx;
1366 					}
1367 					ndx++;
1368 				}
1369 				if (r_index < 0)
1370 					ereport(ERROR,
1371 							(errcode(ERRCODE_UNDEFINED_COLUMN),
1372 							 errmsg("column \"%s\" specified in USING clause does not exist in right table",
1373 									u_colname)));
1374 				r_colnos = lappend_int(r_colnos, r_index + 1);
1375 
1376 				l_colvar = buildVarFromNSColumn(l_nscolumns + l_index);
1377 				l_usingvars = lappend(l_usingvars, l_colvar);
1378 				r_colvar = buildVarFromNSColumn(r_nscolumns + r_index);
1379 				r_usingvars = lappend(r_usingvars, r_colvar);
1380 
1381 				res_colnames = lappend(res_colnames, lfirst(ucol));
1382 				u_colvar = buildMergedJoinVar(pstate,
1383 											  j->jointype,
1384 											  l_colvar,
1385 											  r_colvar);
1386 				res_colvars = lappend(res_colvars, u_colvar);
1387 				res_nscolumn = res_nscolumns + res_colindex;
1388 				res_colindex++;
1389 				if (u_colvar == (Node *) l_colvar)
1390 				{
1391 					/* Merged column is equivalent to left input */
1392 					*res_nscolumn = l_nscolumns[l_index];
1393 				}
1394 				else if (u_colvar == (Node *) r_colvar)
1395 				{
1396 					/* Merged column is equivalent to right input */
1397 					*res_nscolumn = r_nscolumns[r_index];
1398 				}
1399 				else
1400 				{
1401 					/*
1402 					 * Merged column is not semantically equivalent to either
1403 					 * input, so it needs to be referenced as the join output
1404 					 * column.  We don't know the join's varno yet, so we'll
1405 					 * replace these zeroes below.
1406 					 */
1407 					res_nscolumn->p_varno = 0;
1408 					res_nscolumn->p_varattno = res_colindex;
1409 					res_nscolumn->p_vartype = exprType(u_colvar);
1410 					res_nscolumn->p_vartypmod = exprTypmod(u_colvar);
1411 					res_nscolumn->p_varcollid = exprCollation(u_colvar);
1412 					res_nscolumn->p_varnosyn = 0;
1413 					res_nscolumn->p_varattnosyn = res_colindex;
1414 				}
1415 			}
1416 
1417 			j->quals = transformJoinUsingClause(pstate,
1418 												l_usingvars,
1419 												r_usingvars);
1420 		}
1421 		else if (j->quals)
1422 		{
1423 			/* User-written ON-condition; transform it */
1424 			j->quals = transformJoinOnClause(pstate, j, my_namespace);
1425 		}
1426 		else
1427 		{
1428 			/* CROSS JOIN: no quals */
1429 		}
1430 
1431 		/* Add remaining columns from each side to the output columns */
1432 		res_colindex +=
1433 			extractRemainingColumns(l_nscolumns, l_colnames, &l_colnos,
1434 									&res_colnames, &res_colvars,
1435 									res_nscolumns + res_colindex);
1436 		res_colindex +=
1437 			extractRemainingColumns(r_nscolumns, r_colnames, &r_colnos,
1438 									&res_colnames, &res_colvars,
1439 									res_nscolumns + res_colindex);
1440 
1441 		/*
1442 		 * Check alias (AS clause), if any.
1443 		 */
1444 		if (j->alias)
1445 		{
1446 			if (j->alias->colnames != NIL)
1447 			{
1448 				if (list_length(j->alias->colnames) > list_length(res_colnames))
1449 					ereport(ERROR,
1450 							(errcode(ERRCODE_SYNTAX_ERROR),
1451 							 errmsg("column alias list for \"%s\" has too many entries",
1452 									j->alias->aliasname)));
1453 			}
1454 		}
1455 
1456 		/*
1457 		 * Now build an RTE and nsitem for the result of the join.
1458 		 * res_nscolumns isn't totally done yet, but that's OK because
1459 		 * addRangeTableEntryForJoin doesn't examine it, only store a pointer.
1460 		 */
1461 		nsitem = addRangeTableEntryForJoin(pstate,
1462 										   res_colnames,
1463 										   res_nscolumns,
1464 										   j->jointype,
1465 										   list_length(j->usingClause),
1466 										   res_colvars,
1467 										   l_colnos,
1468 										   r_colnos,
1469 										   j->join_using_alias,
1470 										   j->alias,
1471 										   true);
1472 
1473 		j->rtindex = nsitem->p_rtindex;
1474 
1475 		/*
1476 		 * Now that we know the join RTE's rangetable index, we can fix up the
1477 		 * res_nscolumns data in places where it should contain that.
1478 		 */
1479 		Assert(res_colindex == list_length(nsitem->p_names->colnames));
1480 		for (k = 0; k < res_colindex; k++)
1481 		{
1482 			ParseNamespaceColumn *nscol = res_nscolumns + k;
1483 
1484 			/* fill in join RTI for merged columns */
1485 			if (nscol->p_varno == 0)
1486 				nscol->p_varno = j->rtindex;
1487 			if (nscol->p_varnosyn == 0)
1488 				nscol->p_varnosyn = j->rtindex;
1489 			/* if join has an alias, it syntactically hides all inputs */
1490 			if (j->alias)
1491 			{
1492 				nscol->p_varnosyn = j->rtindex;
1493 				nscol->p_varattnosyn = k + 1;
1494 			}
1495 		}
1496 
1497 		/* make a matching link to the JoinExpr for later use */
1498 		for (k = list_length(pstate->p_joinexprs) + 1; k < j->rtindex; k++)
1499 			pstate->p_joinexprs = lappend(pstate->p_joinexprs, NULL);
1500 		pstate->p_joinexprs = lappend(pstate->p_joinexprs, j);
1501 		Assert(list_length(pstate->p_joinexprs) == j->rtindex);
1502 
1503 		/*
1504 		 * If the join has a USING alias, build a ParseNamespaceItem for that
1505 		 * and add it to the list of nsitems in the join's input.
1506 		 */
1507 		if (j->join_using_alias)
1508 		{
1509 			ParseNamespaceItem *jnsitem;
1510 
1511 			jnsitem = (ParseNamespaceItem *) palloc(sizeof(ParseNamespaceItem));
1512 			jnsitem->p_names = j->join_using_alias;
1513 			jnsitem->p_rte = nsitem->p_rte;
1514 			jnsitem->p_rtindex = nsitem->p_rtindex;
1515 			/* no need to copy the first N columns, just use res_nscolumns */
1516 			jnsitem->p_nscolumns = res_nscolumns;
1517 			/* set default visibility flags; might get changed later */
1518 			jnsitem->p_rel_visible = true;
1519 			jnsitem->p_cols_visible = true;
1520 			jnsitem->p_lateral_only = false;
1521 			jnsitem->p_lateral_ok = true;
1522 			/* Per SQL, we must check for alias conflicts */
1523 			checkNameSpaceConflicts(pstate, list_make1(jnsitem), my_namespace);
1524 			my_namespace = lappend(my_namespace, jnsitem);
1525 		}
1526 
1527 		/*
1528 		 * Prepare returned namespace list.  If the JOIN has an alias then it
1529 		 * hides the contained RTEs completely; otherwise, the contained RTEs
1530 		 * are still visible as table names, but are not visible for
1531 		 * unqualified column-name access.
1532 		 *
1533 		 * Note: if there are nested alias-less JOINs, the lower-level ones
1534 		 * will remain in the list although they have neither p_rel_visible
1535 		 * nor p_cols_visible set.  We could delete such list items, but it's
1536 		 * unclear that it's worth expending cycles to do so.
1537 		 */
1538 		if (j->alias != NULL)
1539 			my_namespace = NIL;
1540 		else
1541 			setNamespaceColumnVisibility(my_namespace, false);
1542 
1543 		/*
1544 		 * The join RTE itself is always made visible for unqualified column
1545 		 * names.  It's visible as a relation name only if it has an alias.
1546 		 */
1547 		nsitem->p_rel_visible = (j->alias != NULL);
1548 		nsitem->p_cols_visible = true;
1549 		nsitem->p_lateral_only = false;
1550 		nsitem->p_lateral_ok = true;
1551 
1552 		*top_nsitem = nsitem;
1553 		*namespace = lappend(my_namespace, nsitem);
1554 
1555 		return (Node *) j;
1556 	}
1557 	else
1558 		elog(ERROR, "unrecognized node type: %d", (int) nodeTag(n));
1559 	return NULL;				/* can't get here, keep compiler quiet */
1560 }
1561 
1562 /*
1563  * buildVarFromNSColumn -
1564  *	  build a Var node using ParseNamespaceColumn data
1565  *
1566  * We assume varlevelsup should be 0, and no location is specified
1567  */
1568 static Var *
buildVarFromNSColumn(ParseNamespaceColumn * nscol)1569 buildVarFromNSColumn(ParseNamespaceColumn *nscol)
1570 {
1571 	Var		   *var;
1572 
1573 	Assert(nscol->p_varno > 0); /* i.e., not deleted column */
1574 	var = makeVar(nscol->p_varno,
1575 				  nscol->p_varattno,
1576 				  nscol->p_vartype,
1577 				  nscol->p_vartypmod,
1578 				  nscol->p_varcollid,
1579 				  0);
1580 	/* makeVar doesn't offer parameters for these, so set by hand: */
1581 	var->varnosyn = nscol->p_varnosyn;
1582 	var->varattnosyn = nscol->p_varattnosyn;
1583 	return var;
1584 }
1585 
1586 /*
1587  * buildMergedJoinVar -
1588  *	  generate a suitable replacement expression for a merged join column
1589  */
1590 static Node *
buildMergedJoinVar(ParseState * pstate,JoinType jointype,Var * l_colvar,Var * r_colvar)1591 buildMergedJoinVar(ParseState *pstate, JoinType jointype,
1592 				   Var *l_colvar, Var *r_colvar)
1593 {
1594 	Oid			outcoltype;
1595 	int32		outcoltypmod;
1596 	Node	   *l_node,
1597 			   *r_node,
1598 			   *res_node;
1599 
1600 	outcoltype = select_common_type(pstate,
1601 									list_make2(l_colvar, r_colvar),
1602 									"JOIN/USING",
1603 									NULL);
1604 	outcoltypmod = select_common_typmod(pstate,
1605 										list_make2(l_colvar, r_colvar),
1606 										outcoltype);
1607 
1608 	/*
1609 	 * Insert coercion functions if needed.  Note that a difference in typmod
1610 	 * can only happen if input has typmod but outcoltypmod is -1. In that
1611 	 * case we insert a RelabelType to clearly mark that result's typmod is
1612 	 * not same as input.  We never need coerce_type_typmod.
1613 	 */
1614 	if (l_colvar->vartype != outcoltype)
1615 		l_node = coerce_type(pstate, (Node *) l_colvar, l_colvar->vartype,
1616 							 outcoltype, outcoltypmod,
1617 							 COERCION_IMPLICIT, COERCE_IMPLICIT_CAST, -1);
1618 	else if (l_colvar->vartypmod != outcoltypmod)
1619 		l_node = (Node *) makeRelabelType((Expr *) l_colvar,
1620 										  outcoltype, outcoltypmod,
1621 										  InvalidOid,	/* fixed below */
1622 										  COERCE_IMPLICIT_CAST);
1623 	else
1624 		l_node = (Node *) l_colvar;
1625 
1626 	if (r_colvar->vartype != outcoltype)
1627 		r_node = coerce_type(pstate, (Node *) r_colvar, r_colvar->vartype,
1628 							 outcoltype, outcoltypmod,
1629 							 COERCION_IMPLICIT, COERCE_IMPLICIT_CAST, -1);
1630 	else if (r_colvar->vartypmod != outcoltypmod)
1631 		r_node = (Node *) makeRelabelType((Expr *) r_colvar,
1632 										  outcoltype, outcoltypmod,
1633 										  InvalidOid,	/* fixed below */
1634 										  COERCE_IMPLICIT_CAST);
1635 	else
1636 		r_node = (Node *) r_colvar;
1637 
1638 	/*
1639 	 * Choose what to emit
1640 	 */
1641 	switch (jointype)
1642 	{
1643 		case JOIN_INNER:
1644 
1645 			/*
1646 			 * We can use either var; prefer non-coerced one if available.
1647 			 */
1648 			if (IsA(l_node, Var))
1649 				res_node = l_node;
1650 			else if (IsA(r_node, Var))
1651 				res_node = r_node;
1652 			else
1653 				res_node = l_node;
1654 			break;
1655 		case JOIN_LEFT:
1656 			/* Always use left var */
1657 			res_node = l_node;
1658 			break;
1659 		case JOIN_RIGHT:
1660 			/* Always use right var */
1661 			res_node = r_node;
1662 			break;
1663 		case JOIN_FULL:
1664 			{
1665 				/*
1666 				 * Here we must build a COALESCE expression to ensure that the
1667 				 * join output is non-null if either input is.
1668 				 */
1669 				CoalesceExpr *c = makeNode(CoalesceExpr);
1670 
1671 				c->coalescetype = outcoltype;
1672 				/* coalescecollid will get set below */
1673 				c->args = list_make2(l_node, r_node);
1674 				c->location = -1;
1675 				res_node = (Node *) c;
1676 				break;
1677 			}
1678 		default:
1679 			elog(ERROR, "unrecognized join type: %d", (int) jointype);
1680 			res_node = NULL;	/* keep compiler quiet */
1681 			break;
1682 	}
1683 
1684 	/*
1685 	 * Apply assign_expr_collations to fix up the collation info in the
1686 	 * coercion and CoalesceExpr nodes, if we made any.  This must be done now
1687 	 * so that the join node's alias vars show correct collation info.
1688 	 */
1689 	assign_expr_collations(pstate, res_node);
1690 
1691 	return res_node;
1692 }
1693 
1694 /*
1695  * setNamespaceColumnVisibility -
1696  *	  Convenience subroutine to update cols_visible flags in a namespace list.
1697  */
1698 static void
setNamespaceColumnVisibility(List * namespace,bool cols_visible)1699 setNamespaceColumnVisibility(List *namespace, bool cols_visible)
1700 {
1701 	ListCell   *lc;
1702 
1703 	foreach(lc, namespace)
1704 	{
1705 		ParseNamespaceItem *nsitem = (ParseNamespaceItem *) lfirst(lc);
1706 
1707 		nsitem->p_cols_visible = cols_visible;
1708 	}
1709 }
1710 
1711 /*
1712  * setNamespaceLateralState -
1713  *	  Convenience subroutine to update LATERAL flags in a namespace list.
1714  */
1715 static void
setNamespaceLateralState(List * namespace,bool lateral_only,bool lateral_ok)1716 setNamespaceLateralState(List *namespace, bool lateral_only, bool lateral_ok)
1717 {
1718 	ListCell   *lc;
1719 
1720 	foreach(lc, namespace)
1721 	{
1722 		ParseNamespaceItem *nsitem = (ParseNamespaceItem *) lfirst(lc);
1723 
1724 		nsitem->p_lateral_only = lateral_only;
1725 		nsitem->p_lateral_ok = lateral_ok;
1726 	}
1727 }
1728 
1729 
1730 /*
1731  * transformWhereClause -
1732  *	  Transform the qualification and make sure it is of type boolean.
1733  *	  Used for WHERE and allied clauses.
1734  *
1735  * constructName does not affect the semantics, but is used in error messages
1736  */
1737 Node *
transformWhereClause(ParseState * pstate,Node * clause,ParseExprKind exprKind,const char * constructName)1738 transformWhereClause(ParseState *pstate, Node *clause,
1739 					 ParseExprKind exprKind, const char *constructName)
1740 {
1741 	Node	   *qual;
1742 
1743 	if (clause == NULL)
1744 		return NULL;
1745 
1746 	qual = transformExpr(pstate, clause, exprKind);
1747 
1748 	qual = coerce_to_boolean(pstate, qual, constructName);
1749 
1750 	return qual;
1751 }
1752 
1753 
1754 /*
1755  * transformLimitClause -
1756  *	  Transform the expression and make sure it is of type bigint.
1757  *	  Used for LIMIT and allied clauses.
1758  *
1759  * Note: as of Postgres 8.2, LIMIT expressions are expected to yield int8,
1760  * rather than int4 as before.
1761  *
1762  * constructName does not affect the semantics, but is used in error messages
1763  */
1764 Node *
transformLimitClause(ParseState * pstate,Node * clause,ParseExprKind exprKind,const char * constructName,LimitOption limitOption)1765 transformLimitClause(ParseState *pstate, Node *clause,
1766 					 ParseExprKind exprKind, const char *constructName,
1767 					 LimitOption limitOption)
1768 {
1769 	Node	   *qual;
1770 
1771 	if (clause == NULL)
1772 		return NULL;
1773 
1774 	qual = transformExpr(pstate, clause, exprKind);
1775 
1776 	qual = coerce_to_specific_type(pstate, qual, INT8OID, constructName);
1777 
1778 	/* LIMIT can't refer to any variables of the current query */
1779 	checkExprIsVarFree(pstate, qual, constructName);
1780 
1781 	/*
1782 	 * Don't allow NULLs in FETCH FIRST .. WITH TIES.  This test is ugly and
1783 	 * extremely simplistic, in that you can pass a NULL anyway by hiding it
1784 	 * inside an expression -- but this protects ruleutils against emitting an
1785 	 * unadorned NULL that's not accepted back by the grammar.
1786 	 */
1787 	if (exprKind == EXPR_KIND_LIMIT && limitOption == LIMIT_OPTION_WITH_TIES &&
1788 		IsA(clause, A_Const) && ((A_Const *) clause)->val.type == T_Null)
1789 		ereport(ERROR,
1790 				(errcode(ERRCODE_INVALID_ROW_COUNT_IN_LIMIT_CLAUSE),
1791 				 errmsg("row count cannot be null in FETCH FIRST ... WITH TIES clause")));
1792 
1793 	return qual;
1794 }
1795 
1796 /*
1797  * checkExprIsVarFree
1798  *		Check that given expr has no Vars of the current query level
1799  *		(aggregates and window functions should have been rejected already).
1800  *
1801  * This is used to check expressions that have to have a consistent value
1802  * across all rows of the query, such as a LIMIT.  Arguably it should reject
1803  * volatile functions, too, but we don't do that --- whatever value the
1804  * function gives on first execution is what you get.
1805  *
1806  * constructName does not affect the semantics, but is used in error messages
1807  */
1808 static void
checkExprIsVarFree(ParseState * pstate,Node * n,const char * constructName)1809 checkExprIsVarFree(ParseState *pstate, Node *n, const char *constructName)
1810 {
1811 	if (contain_vars_of_level(n, 0))
1812 	{
1813 		ereport(ERROR,
1814 				(errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
1815 		/* translator: %s is name of a SQL construct, eg LIMIT */
1816 				 errmsg("argument of %s must not contain variables",
1817 						constructName),
1818 				 parser_errposition(pstate,
1819 									locate_var_of_level(n, 0))));
1820 	}
1821 }
1822 
1823 
1824 /*
1825  * checkTargetlistEntrySQL92 -
1826  *	  Validate a targetlist entry found by findTargetlistEntrySQL92
1827  *
1828  * When we select a pre-existing tlist entry as a result of syntax such
1829  * as "GROUP BY 1", we have to make sure it is acceptable for use in the
1830  * indicated clause type; transformExpr() will have treated it as a regular
1831  * targetlist item.
1832  */
1833 static void
checkTargetlistEntrySQL92(ParseState * pstate,TargetEntry * tle,ParseExprKind exprKind)1834 checkTargetlistEntrySQL92(ParseState *pstate, TargetEntry *tle,
1835 						  ParseExprKind exprKind)
1836 {
1837 	switch (exprKind)
1838 	{
1839 		case EXPR_KIND_GROUP_BY:
1840 			/* reject aggregates and window functions */
1841 			if (pstate->p_hasAggs &&
1842 				contain_aggs_of_level((Node *) tle->expr, 0))
1843 				ereport(ERROR,
1844 						(errcode(ERRCODE_GROUPING_ERROR),
1845 				/* translator: %s is name of a SQL construct, eg GROUP BY */
1846 						 errmsg("aggregate functions are not allowed in %s",
1847 								ParseExprKindName(exprKind)),
1848 						 parser_errposition(pstate,
1849 											locate_agg_of_level((Node *) tle->expr, 0))));
1850 			if (pstate->p_hasWindowFuncs &&
1851 				contain_windowfuncs((Node *) tle->expr))
1852 				ereport(ERROR,
1853 						(errcode(ERRCODE_WINDOWING_ERROR),
1854 				/* translator: %s is name of a SQL construct, eg GROUP BY */
1855 						 errmsg("window functions are not allowed in %s",
1856 								ParseExprKindName(exprKind)),
1857 						 parser_errposition(pstate,
1858 											locate_windowfunc((Node *) tle->expr))));
1859 			break;
1860 		case EXPR_KIND_ORDER_BY:
1861 			/* no extra checks needed */
1862 			break;
1863 		case EXPR_KIND_DISTINCT_ON:
1864 			/* no extra checks needed */
1865 			break;
1866 		default:
1867 			elog(ERROR, "unexpected exprKind in checkTargetlistEntrySQL92");
1868 			break;
1869 	}
1870 }
1871 
1872 /*
1873  *	findTargetlistEntrySQL92 -
1874  *	  Returns the targetlist entry matching the given (untransformed) node.
1875  *	  If no matching entry exists, one is created and appended to the target
1876  *	  list as a "resjunk" node.
1877  *
1878  * This function supports the old SQL92 ORDER BY interpretation, where the
1879  * expression is an output column name or number.  If we fail to find a
1880  * match of that sort, we fall through to the SQL99 rules.  For historical
1881  * reasons, Postgres also allows this interpretation for GROUP BY, though
1882  * the standard never did.  However, for GROUP BY we prefer a SQL99 match.
1883  * This function is *not* used for WINDOW definitions.
1884  *
1885  * node		the ORDER BY, GROUP BY, or DISTINCT ON expression to be matched
1886  * tlist	the target list (passed by reference so we can append to it)
1887  * exprKind identifies clause type being processed
1888  */
1889 static TargetEntry *
findTargetlistEntrySQL92(ParseState * pstate,Node * node,List ** tlist,ParseExprKind exprKind)1890 findTargetlistEntrySQL92(ParseState *pstate, Node *node, List **tlist,
1891 						 ParseExprKind exprKind)
1892 {
1893 	ListCell   *tl;
1894 
1895 	/*----------
1896 	 * Handle two special cases as mandated by the SQL92 spec:
1897 	 *
1898 	 * 1. Bare ColumnName (no qualifier or subscripts)
1899 	 *	  For a bare identifier, we search for a matching column name
1900 	 *	  in the existing target list.  Multiple matches are an error
1901 	 *	  unless they refer to identical values; for example,
1902 	 *	  we allow	SELECT a, a FROM table ORDER BY a
1903 	 *	  but not	SELECT a AS b, b FROM table ORDER BY b
1904 	 *	  If no match is found, we fall through and treat the identifier
1905 	 *	  as an expression.
1906 	 *	  For GROUP BY, it is incorrect to match the grouping item against
1907 	 *	  targetlist entries: according to SQL92, an identifier in GROUP BY
1908 	 *	  is a reference to a column name exposed by FROM, not to a target
1909 	 *	  list column.  However, many implementations (including pre-7.0
1910 	 *	  PostgreSQL) accept this anyway.  So for GROUP BY, we look first
1911 	 *	  to see if the identifier matches any FROM column name, and only
1912 	 *	  try for a targetlist name if it doesn't.  This ensures that we
1913 	 *	  adhere to the spec in the case where the name could be both.
1914 	 *	  DISTINCT ON isn't in the standard, so we can do what we like there;
1915 	 *	  we choose to make it work like ORDER BY, on the rather flimsy
1916 	 *	  grounds that ordinary DISTINCT works on targetlist entries.
1917 	 *
1918 	 * 2. IntegerConstant
1919 	 *	  This means to use the n'th item in the existing target list.
1920 	 *	  Note that it would make no sense to order/group/distinct by an
1921 	 *	  actual constant, so this does not create a conflict with SQL99.
1922 	 *	  GROUP BY column-number is not allowed by SQL92, but since
1923 	 *	  the standard has no other behavior defined for this syntax,
1924 	 *	  we may as well accept this common extension.
1925 	 *
1926 	 * Note that pre-existing resjunk targets must not be used in either case,
1927 	 * since the user didn't write them in his SELECT list.
1928 	 *
1929 	 * If neither special case applies, fall through to treat the item as
1930 	 * an expression per SQL99.
1931 	 *----------
1932 	 */
1933 	if (IsA(node, ColumnRef) &&
1934 		list_length(((ColumnRef *) node)->fields) == 1 &&
1935 		IsA(linitial(((ColumnRef *) node)->fields), String))
1936 	{
1937 		char	   *name = strVal(linitial(((ColumnRef *) node)->fields));
1938 		int			location = ((ColumnRef *) node)->location;
1939 
1940 		if (exprKind == EXPR_KIND_GROUP_BY)
1941 		{
1942 			/*
1943 			 * In GROUP BY, we must prefer a match against a FROM-clause
1944 			 * column to one against the targetlist.  Look to see if there is
1945 			 * a matching column.  If so, fall through to use SQL99 rules.
1946 			 * NOTE: if name could refer ambiguously to more than one column
1947 			 * name exposed by FROM, colNameToVar will ereport(ERROR). That's
1948 			 * just what we want here.
1949 			 *
1950 			 * Small tweak for 7.4.3: ignore matches in upper query levels.
1951 			 * This effectively changes the search order for bare names to (1)
1952 			 * local FROM variables, (2) local targetlist aliases, (3) outer
1953 			 * FROM variables, whereas before it was (1) (3) (2). SQL92 and
1954 			 * SQL99 do not allow GROUPing BY an outer reference, so this
1955 			 * breaks no cases that are legal per spec, and it seems a more
1956 			 * self-consistent behavior.
1957 			 */
1958 			if (colNameToVar(pstate, name, true, location) != NULL)
1959 				name = NULL;
1960 		}
1961 
1962 		if (name != NULL)
1963 		{
1964 			TargetEntry *target_result = NULL;
1965 
1966 			foreach(tl, *tlist)
1967 			{
1968 				TargetEntry *tle = (TargetEntry *) lfirst(tl);
1969 
1970 				if (!tle->resjunk &&
1971 					strcmp(tle->resname, name) == 0)
1972 				{
1973 					if (target_result != NULL)
1974 					{
1975 						if (!equal(target_result->expr, tle->expr))
1976 							ereport(ERROR,
1977 									(errcode(ERRCODE_AMBIGUOUS_COLUMN),
1978 
1979 							/*------
1980 							  translator: first %s is name of a SQL construct, eg ORDER BY */
1981 									 errmsg("%s \"%s\" is ambiguous",
1982 											ParseExprKindName(exprKind),
1983 											name),
1984 									 parser_errposition(pstate, location)));
1985 					}
1986 					else
1987 						target_result = tle;
1988 					/* Stay in loop to check for ambiguity */
1989 				}
1990 			}
1991 			if (target_result != NULL)
1992 			{
1993 				/* return the first match, after suitable validation */
1994 				checkTargetlistEntrySQL92(pstate, target_result, exprKind);
1995 				return target_result;
1996 			}
1997 		}
1998 	}
1999 	if (IsA(node, A_Const))
2000 	{
2001 		Value	   *val = &((A_Const *) node)->val;
2002 		int			location = ((A_Const *) node)->location;
2003 		int			targetlist_pos = 0;
2004 		int			target_pos;
2005 
2006 		if (!IsA(val, Integer))
2007 			ereport(ERROR,
2008 					(errcode(ERRCODE_SYNTAX_ERROR),
2009 			/* translator: %s is name of a SQL construct, eg ORDER BY */
2010 					 errmsg("non-integer constant in %s",
2011 							ParseExprKindName(exprKind)),
2012 					 parser_errposition(pstate, location)));
2013 
2014 		target_pos = intVal(val);
2015 		foreach(tl, *tlist)
2016 		{
2017 			TargetEntry *tle = (TargetEntry *) lfirst(tl);
2018 
2019 			if (!tle->resjunk)
2020 			{
2021 				if (++targetlist_pos == target_pos)
2022 				{
2023 					/* return the unique match, after suitable validation */
2024 					checkTargetlistEntrySQL92(pstate, tle, exprKind);
2025 					return tle;
2026 				}
2027 			}
2028 		}
2029 		ereport(ERROR,
2030 				(errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
2031 		/* translator: %s is name of a SQL construct, eg ORDER BY */
2032 				 errmsg("%s position %d is not in select list",
2033 						ParseExprKindName(exprKind), target_pos),
2034 				 parser_errposition(pstate, location)));
2035 	}
2036 
2037 	/*
2038 	 * Otherwise, we have an expression, so process it per SQL99 rules.
2039 	 */
2040 	return findTargetlistEntrySQL99(pstate, node, tlist, exprKind);
2041 }
2042 
2043 /*
2044  *	findTargetlistEntrySQL99 -
2045  *	  Returns the targetlist entry matching the given (untransformed) node.
2046  *	  If no matching entry exists, one is created and appended to the target
2047  *	  list as a "resjunk" node.
2048  *
2049  * This function supports the SQL99 interpretation, wherein the expression
2050  * is just an ordinary expression referencing input column names.
2051  *
2052  * node		the ORDER BY, GROUP BY, etc expression to be matched
2053  * tlist	the target list (passed by reference so we can append to it)
2054  * exprKind identifies clause type being processed
2055  */
2056 static TargetEntry *
findTargetlistEntrySQL99(ParseState * pstate,Node * node,List ** tlist,ParseExprKind exprKind)2057 findTargetlistEntrySQL99(ParseState *pstate, Node *node, List **tlist,
2058 						 ParseExprKind exprKind)
2059 {
2060 	TargetEntry *target_result;
2061 	ListCell   *tl;
2062 	Node	   *expr;
2063 
2064 	/*
2065 	 * Convert the untransformed node to a transformed expression, and search
2066 	 * for a match in the tlist.  NOTE: it doesn't really matter whether there
2067 	 * is more than one match.  Also, we are willing to match an existing
2068 	 * resjunk target here, though the SQL92 cases above must ignore resjunk
2069 	 * targets.
2070 	 */
2071 	expr = transformExpr(pstate, node, exprKind);
2072 
2073 	foreach(tl, *tlist)
2074 	{
2075 		TargetEntry *tle = (TargetEntry *) lfirst(tl);
2076 		Node	   *texpr;
2077 
2078 		/*
2079 		 * Ignore any implicit cast on the existing tlist expression.
2080 		 *
2081 		 * This essentially allows the ORDER/GROUP/etc item to adopt the same
2082 		 * datatype previously selected for a textually-equivalent tlist item.
2083 		 * There can't be any implicit cast at top level in an ordinary SELECT
2084 		 * tlist at this stage, but the case does arise with ORDER BY in an
2085 		 * aggregate function.
2086 		 */
2087 		texpr = strip_implicit_coercions((Node *) tle->expr);
2088 
2089 		if (equal(expr, texpr))
2090 			return tle;
2091 	}
2092 
2093 	/*
2094 	 * If no matches, construct a new target entry which is appended to the
2095 	 * end of the target list.  This target is given resjunk = true so that it
2096 	 * will not be projected into the final tuple.
2097 	 */
2098 	target_result = transformTargetEntry(pstate, node, expr, exprKind,
2099 										 NULL, true);
2100 
2101 	*tlist = lappend(*tlist, target_result);
2102 
2103 	return target_result;
2104 }
2105 
2106 /*-------------------------------------------------------------------------
2107  * Flatten out parenthesized sublists in grouping lists, and some cases
2108  * of nested grouping sets.
2109  *
2110  * Inside a grouping set (ROLLUP, CUBE, or GROUPING SETS), we expect the
2111  * content to be nested no more than 2 deep: i.e. ROLLUP((a,b),(c,d)) is
2112  * ok, but ROLLUP((a,(b,c)),d) is flattened to ((a,b,c),d), which we then
2113  * (later) normalize to ((a,b,c),(d)).
2114  *
2115  * CUBE or ROLLUP can be nested inside GROUPING SETS (but not the reverse),
2116  * and we leave that alone if we find it. But if we see GROUPING SETS inside
2117  * GROUPING SETS, we can flatten and normalize as follows:
2118  *	 GROUPING SETS (a, (b,c), GROUPING SETS ((c,d),(e)), (f,g))
2119  * becomes
2120  *	 GROUPING SETS ((a), (b,c), (c,d), (e), (f,g))
2121  *
2122  * This is per the spec's syntax transformations, but these are the only such
2123  * transformations we do in parse analysis, so that queries retain the
2124  * originally specified grouping set syntax for CUBE and ROLLUP as much as
2125  * possible when deparsed. (Full expansion of the result into a list of
2126  * grouping sets is left to the planner.)
2127  *
2128  * When we're done, the resulting list should contain only these possible
2129  * elements:
2130  *	 - an expression
2131  *	 - a CUBE or ROLLUP with a list of expressions nested 2 deep
2132  *	 - a GROUPING SET containing any of:
2133  *		- expression lists
2134  *		- empty grouping sets
2135  *		- CUBE or ROLLUP nodes with lists nested 2 deep
2136  * The return is a new list, but doesn't deep-copy the old nodes except for
2137  * GroupingSet nodes.
2138  *
2139  * As a side effect, flag whether the list has any GroupingSet nodes.
2140  *-------------------------------------------------------------------------
2141  */
2142 static Node *
flatten_grouping_sets(Node * expr,bool toplevel,bool * hasGroupingSets)2143 flatten_grouping_sets(Node *expr, bool toplevel, bool *hasGroupingSets)
2144 {
2145 	/* just in case of pathological input */
2146 	check_stack_depth();
2147 
2148 	if (expr == (Node *) NIL)
2149 		return (Node *) NIL;
2150 
2151 	switch (expr->type)
2152 	{
2153 		case T_RowExpr:
2154 			{
2155 				RowExpr    *r = (RowExpr *) expr;
2156 
2157 				if (r->row_format == COERCE_IMPLICIT_CAST)
2158 					return flatten_grouping_sets((Node *) r->args,
2159 												 false, NULL);
2160 			}
2161 			break;
2162 		case T_GroupingSet:
2163 			{
2164 				GroupingSet *gset = (GroupingSet *) expr;
2165 				ListCell   *l2;
2166 				List	   *result_set = NIL;
2167 
2168 				if (hasGroupingSets)
2169 					*hasGroupingSets = true;
2170 
2171 				/*
2172 				 * at the top level, we skip over all empty grouping sets; the
2173 				 * caller can supply the canonical GROUP BY () if nothing is
2174 				 * left.
2175 				 */
2176 
2177 				if (toplevel && gset->kind == GROUPING_SET_EMPTY)
2178 					return (Node *) NIL;
2179 
2180 				foreach(l2, gset->content)
2181 				{
2182 					Node	   *n1 = lfirst(l2);
2183 					Node	   *n2 = flatten_grouping_sets(n1, false, NULL);
2184 
2185 					if (IsA(n1, GroupingSet) &&
2186 						((GroupingSet *) n1)->kind == GROUPING_SET_SETS)
2187 						result_set = list_concat(result_set, (List *) n2);
2188 					else
2189 						result_set = lappend(result_set, n2);
2190 				}
2191 
2192 				/*
2193 				 * At top level, keep the grouping set node; but if we're in a
2194 				 * nested grouping set, then we need to concat the flattened
2195 				 * result into the outer list if it's simply nested.
2196 				 */
2197 
2198 				if (toplevel || (gset->kind != GROUPING_SET_SETS))
2199 				{
2200 					return (Node *) makeGroupingSet(gset->kind, result_set, gset->location);
2201 				}
2202 				else
2203 					return (Node *) result_set;
2204 			}
2205 		case T_List:
2206 			{
2207 				List	   *result = NIL;
2208 				ListCell   *l;
2209 
2210 				foreach(l, (List *) expr)
2211 				{
2212 					Node	   *n = flatten_grouping_sets(lfirst(l), toplevel, hasGroupingSets);
2213 
2214 					if (n != (Node *) NIL)
2215 					{
2216 						if (IsA(n, List))
2217 							result = list_concat(result, (List *) n);
2218 						else
2219 							result = lappend(result, n);
2220 					}
2221 				}
2222 
2223 				return (Node *) result;
2224 			}
2225 		default:
2226 			break;
2227 	}
2228 
2229 	return expr;
2230 }
2231 
2232 /*
2233  * Transform a single expression within a GROUP BY clause or grouping set.
2234  *
2235  * The expression is added to the targetlist if not already present, and to the
2236  * flatresult list (which will become the groupClause) if not already present
2237  * there.  The sortClause is consulted for operator and sort order hints.
2238  *
2239  * Returns the ressortgroupref of the expression.
2240  *
2241  * flatresult	reference to flat list of SortGroupClause nodes
2242  * seen_local	bitmapset of sortgrouprefs already seen at the local level
2243  * pstate		ParseState
2244  * gexpr		node to transform
2245  * targetlist	reference to TargetEntry list
2246  * sortClause	ORDER BY clause (SortGroupClause nodes)
2247  * exprKind		expression kind
2248  * useSQL99		SQL99 rather than SQL92 syntax
2249  * toplevel		false if within any grouping set
2250  */
2251 static Index
transformGroupClauseExpr(List ** flatresult,Bitmapset * seen_local,ParseState * pstate,Node * gexpr,List ** targetlist,List * sortClause,ParseExprKind exprKind,bool useSQL99,bool toplevel)2252 transformGroupClauseExpr(List **flatresult, Bitmapset *seen_local,
2253 						 ParseState *pstate, Node *gexpr,
2254 						 List **targetlist, List *sortClause,
2255 						 ParseExprKind exprKind, bool useSQL99, bool toplevel)
2256 {
2257 	TargetEntry *tle;
2258 	bool		found = false;
2259 
2260 	if (useSQL99)
2261 		tle = findTargetlistEntrySQL99(pstate, gexpr,
2262 									   targetlist, exprKind);
2263 	else
2264 		tle = findTargetlistEntrySQL92(pstate, gexpr,
2265 									   targetlist, exprKind);
2266 
2267 	if (tle->ressortgroupref > 0)
2268 	{
2269 		ListCell   *sl;
2270 
2271 		/*
2272 		 * Eliminate duplicates (GROUP BY x, x) but only at local level.
2273 		 * (Duplicates in grouping sets can affect the number of returned
2274 		 * rows, so can't be dropped indiscriminately.)
2275 		 *
2276 		 * Since we don't care about anything except the sortgroupref, we can
2277 		 * use a bitmapset rather than scanning lists.
2278 		 */
2279 		if (bms_is_member(tle->ressortgroupref, seen_local))
2280 			return 0;
2281 
2282 		/*
2283 		 * If we're already in the flat clause list, we don't need to consider
2284 		 * adding ourselves again.
2285 		 */
2286 		found = targetIsInSortList(tle, InvalidOid, *flatresult);
2287 		if (found)
2288 			return tle->ressortgroupref;
2289 
2290 		/*
2291 		 * If the GROUP BY tlist entry also appears in ORDER BY, copy operator
2292 		 * info from the (first) matching ORDER BY item.  This means that if
2293 		 * you write something like "GROUP BY foo ORDER BY foo USING <<<", the
2294 		 * GROUP BY operation silently takes on the equality semantics implied
2295 		 * by the ORDER BY.  There are two reasons to do this: it improves the
2296 		 * odds that we can implement both GROUP BY and ORDER BY with a single
2297 		 * sort step, and it allows the user to choose the equality semantics
2298 		 * used by GROUP BY, should she be working with a datatype that has
2299 		 * more than one equality operator.
2300 		 *
2301 		 * If we're in a grouping set, though, we force our requested ordering
2302 		 * to be NULLS LAST, because if we have any hope of using a sorted agg
2303 		 * for the job, we're going to be tacking on generated NULL values
2304 		 * after the corresponding groups. If the user demands nulls first,
2305 		 * another sort step is going to be inevitable, but that's the
2306 		 * planner's problem.
2307 		 */
2308 
2309 		foreach(sl, sortClause)
2310 		{
2311 			SortGroupClause *sc = (SortGroupClause *) lfirst(sl);
2312 
2313 			if (sc->tleSortGroupRef == tle->ressortgroupref)
2314 			{
2315 				SortGroupClause *grpc = copyObject(sc);
2316 
2317 				if (!toplevel)
2318 					grpc->nulls_first = false;
2319 				*flatresult = lappend(*flatresult, grpc);
2320 				found = true;
2321 				break;
2322 			}
2323 		}
2324 	}
2325 
2326 	/*
2327 	 * If no match in ORDER BY, just add it to the result using default
2328 	 * sort/group semantics.
2329 	 */
2330 	if (!found)
2331 		*flatresult = addTargetToGroupList(pstate, tle,
2332 										   *flatresult, *targetlist,
2333 										   exprLocation(gexpr));
2334 
2335 	/*
2336 	 * _something_ must have assigned us a sortgroupref by now...
2337 	 */
2338 
2339 	return tle->ressortgroupref;
2340 }
2341 
2342 /*
2343  * Transform a list of expressions within a GROUP BY clause or grouping set.
2344  *
2345  * The list of expressions belongs to a single clause within which duplicates
2346  * can be safely eliminated.
2347  *
2348  * Returns an integer list of ressortgroupref values.
2349  *
2350  * flatresult	reference to flat list of SortGroupClause nodes
2351  * pstate		ParseState
2352  * list			nodes to transform
2353  * targetlist	reference to TargetEntry list
2354  * sortClause	ORDER BY clause (SortGroupClause nodes)
2355  * exprKind		expression kind
2356  * useSQL99		SQL99 rather than SQL92 syntax
2357  * toplevel		false if within any grouping set
2358  */
2359 static List *
transformGroupClauseList(List ** flatresult,ParseState * pstate,List * list,List ** targetlist,List * sortClause,ParseExprKind exprKind,bool useSQL99,bool toplevel)2360 transformGroupClauseList(List **flatresult,
2361 						 ParseState *pstate, List *list,
2362 						 List **targetlist, List *sortClause,
2363 						 ParseExprKind exprKind, bool useSQL99, bool toplevel)
2364 {
2365 	Bitmapset  *seen_local = NULL;
2366 	List	   *result = NIL;
2367 	ListCell   *gl;
2368 
2369 	foreach(gl, list)
2370 	{
2371 		Node	   *gexpr = (Node *) lfirst(gl);
2372 
2373 		Index		ref = transformGroupClauseExpr(flatresult,
2374 												   seen_local,
2375 												   pstate,
2376 												   gexpr,
2377 												   targetlist,
2378 												   sortClause,
2379 												   exprKind,
2380 												   useSQL99,
2381 												   toplevel);
2382 
2383 		if (ref > 0)
2384 		{
2385 			seen_local = bms_add_member(seen_local, ref);
2386 			result = lappend_int(result, ref);
2387 		}
2388 	}
2389 
2390 	return result;
2391 }
2392 
2393 /*
2394  * Transform a grouping set and (recursively) its content.
2395  *
2396  * The grouping set might be a GROUPING SETS node with other grouping sets
2397  * inside it, but SETS within SETS have already been flattened out before
2398  * reaching here.
2399  *
2400  * Returns the transformed node, which now contains SIMPLE nodes with lists
2401  * of ressortgrouprefs rather than expressions.
2402  *
2403  * flatresult	reference to flat list of SortGroupClause nodes
2404  * pstate		ParseState
2405  * gset			grouping set to transform
2406  * targetlist	reference to TargetEntry list
2407  * sortClause	ORDER BY clause (SortGroupClause nodes)
2408  * exprKind		expression kind
2409  * useSQL99		SQL99 rather than SQL92 syntax
2410  * toplevel		false if within any grouping set
2411  */
2412 static Node *
transformGroupingSet(List ** flatresult,ParseState * pstate,GroupingSet * gset,List ** targetlist,List * sortClause,ParseExprKind exprKind,bool useSQL99,bool toplevel)2413 transformGroupingSet(List **flatresult,
2414 					 ParseState *pstate, GroupingSet *gset,
2415 					 List **targetlist, List *sortClause,
2416 					 ParseExprKind exprKind, bool useSQL99, bool toplevel)
2417 {
2418 	ListCell   *gl;
2419 	List	   *content = NIL;
2420 
2421 	Assert(toplevel || gset->kind != GROUPING_SET_SETS);
2422 
2423 	foreach(gl, gset->content)
2424 	{
2425 		Node	   *n = lfirst(gl);
2426 
2427 		if (IsA(n, List))
2428 		{
2429 			List	   *l = transformGroupClauseList(flatresult,
2430 													 pstate, (List *) n,
2431 													 targetlist, sortClause,
2432 													 exprKind, useSQL99, false);
2433 
2434 			content = lappend(content, makeGroupingSet(GROUPING_SET_SIMPLE,
2435 													   l,
2436 													   exprLocation(n)));
2437 		}
2438 		else if (IsA(n, GroupingSet))
2439 		{
2440 			GroupingSet *gset2 = (GroupingSet *) lfirst(gl);
2441 
2442 			content = lappend(content, transformGroupingSet(flatresult,
2443 															pstate, gset2,
2444 															targetlist, sortClause,
2445 															exprKind, useSQL99, false));
2446 		}
2447 		else
2448 		{
2449 			Index		ref = transformGroupClauseExpr(flatresult,
2450 													   NULL,
2451 													   pstate,
2452 													   n,
2453 													   targetlist,
2454 													   sortClause,
2455 													   exprKind,
2456 													   useSQL99,
2457 													   false);
2458 
2459 			content = lappend(content, makeGroupingSet(GROUPING_SET_SIMPLE,
2460 													   list_make1_int(ref),
2461 													   exprLocation(n)));
2462 		}
2463 	}
2464 
2465 	/* Arbitrarily cap the size of CUBE, which has exponential growth */
2466 	if (gset->kind == GROUPING_SET_CUBE)
2467 	{
2468 		if (list_length(content) > 12)
2469 			ereport(ERROR,
2470 					(errcode(ERRCODE_TOO_MANY_COLUMNS),
2471 					 errmsg("CUBE is limited to 12 elements"),
2472 					 parser_errposition(pstate, gset->location)));
2473 	}
2474 
2475 	return (Node *) makeGroupingSet(gset->kind, content, gset->location);
2476 }
2477 
2478 
2479 /*
2480  * transformGroupClause -
2481  *	  transform a GROUP BY clause
2482  *
2483  * GROUP BY items will be added to the targetlist (as resjunk columns)
2484  * if not already present, so the targetlist must be passed by reference.
2485  *
2486  * This is also used for window PARTITION BY clauses (which act almost the
2487  * same, but are always interpreted per SQL99 rules).
2488  *
2489  * Grouping sets make this a lot more complex than it was. Our goal here is
2490  * twofold: we make a flat list of SortGroupClause nodes referencing each
2491  * distinct expression used for grouping, with those expressions added to the
2492  * targetlist if needed. At the same time, we build the groupingSets tree,
2493  * which stores only ressortgrouprefs as integer lists inside GroupingSet nodes
2494  * (possibly nested, but limited in depth: a GROUPING_SET_SETS node can contain
2495  * nested SIMPLE, CUBE or ROLLUP nodes, but not more sets - we flatten that
2496  * out; while CUBE and ROLLUP can contain only SIMPLE nodes).
2497  *
2498  * We skip much of the hard work if there are no grouping sets.
2499  *
2500  * One subtlety is that the groupClause list can end up empty while the
2501  * groupingSets list is not; this happens if there are only empty grouping
2502  * sets, or an explicit GROUP BY (). This has the same effect as specifying
2503  * aggregates or a HAVING clause with no GROUP BY; the output is one row per
2504  * grouping set even if the input is empty.
2505  *
2506  * Returns the transformed (flat) groupClause.
2507  *
2508  * pstate		ParseState
2509  * grouplist	clause to transform
2510  * groupingSets reference to list to contain the grouping set tree
2511  * targetlist	reference to TargetEntry list
2512  * sortClause	ORDER BY clause (SortGroupClause nodes)
2513  * exprKind		expression kind
2514  * useSQL99		SQL99 rather than SQL92 syntax
2515  */
2516 List *
transformGroupClause(ParseState * pstate,List * grouplist,List ** groupingSets,List ** targetlist,List * sortClause,ParseExprKind exprKind,bool useSQL99)2517 transformGroupClause(ParseState *pstate, List *grouplist, List **groupingSets,
2518 					 List **targetlist, List *sortClause,
2519 					 ParseExprKind exprKind, bool useSQL99)
2520 {
2521 	List	   *result = NIL;
2522 	List	   *flat_grouplist;
2523 	List	   *gsets = NIL;
2524 	ListCell   *gl;
2525 	bool		hasGroupingSets = false;
2526 	Bitmapset  *seen_local = NULL;
2527 
2528 	/*
2529 	 * Recursively flatten implicit RowExprs. (Technically this is only needed
2530 	 * for GROUP BY, per the syntax rules for grouping sets, but we do it
2531 	 * anyway.)
2532 	 */
2533 	flat_grouplist = (List *) flatten_grouping_sets((Node *) grouplist,
2534 													true,
2535 													&hasGroupingSets);
2536 
2537 	/*
2538 	 * If the list is now empty, but hasGroupingSets is true, it's because we
2539 	 * elided redundant empty grouping sets. Restore a single empty grouping
2540 	 * set to leave a canonical form: GROUP BY ()
2541 	 */
2542 
2543 	if (flat_grouplist == NIL && hasGroupingSets)
2544 	{
2545 		flat_grouplist = list_make1(makeGroupingSet(GROUPING_SET_EMPTY,
2546 													NIL,
2547 													exprLocation((Node *) grouplist)));
2548 	}
2549 
2550 	foreach(gl, flat_grouplist)
2551 	{
2552 		Node	   *gexpr = (Node *) lfirst(gl);
2553 
2554 		if (IsA(gexpr, GroupingSet))
2555 		{
2556 			GroupingSet *gset = (GroupingSet *) gexpr;
2557 
2558 			switch (gset->kind)
2559 			{
2560 				case GROUPING_SET_EMPTY:
2561 					gsets = lappend(gsets, gset);
2562 					break;
2563 				case GROUPING_SET_SIMPLE:
2564 					/* can't happen */
2565 					Assert(false);
2566 					break;
2567 				case GROUPING_SET_SETS:
2568 				case GROUPING_SET_CUBE:
2569 				case GROUPING_SET_ROLLUP:
2570 					gsets = lappend(gsets,
2571 									transformGroupingSet(&result,
2572 														 pstate, gset,
2573 														 targetlist, sortClause,
2574 														 exprKind, useSQL99, true));
2575 					break;
2576 			}
2577 		}
2578 		else
2579 		{
2580 			Index		ref = transformGroupClauseExpr(&result, seen_local,
2581 													   pstate, gexpr,
2582 													   targetlist, sortClause,
2583 													   exprKind, useSQL99, true);
2584 
2585 			if (ref > 0)
2586 			{
2587 				seen_local = bms_add_member(seen_local, ref);
2588 				if (hasGroupingSets)
2589 					gsets = lappend(gsets,
2590 									makeGroupingSet(GROUPING_SET_SIMPLE,
2591 													list_make1_int(ref),
2592 													exprLocation(gexpr)));
2593 			}
2594 		}
2595 	}
2596 
2597 	/* parser should prevent this */
2598 	Assert(gsets == NIL || groupingSets != NULL);
2599 
2600 	if (groupingSets)
2601 		*groupingSets = gsets;
2602 
2603 	return result;
2604 }
2605 
2606 /*
2607  * transformSortClause -
2608  *	  transform an ORDER BY clause
2609  *
2610  * ORDER BY items will be added to the targetlist (as resjunk columns)
2611  * if not already present, so the targetlist must be passed by reference.
2612  *
2613  * This is also used for window and aggregate ORDER BY clauses (which act
2614  * almost the same, but are always interpreted per SQL99 rules).
2615  */
2616 List *
transformSortClause(ParseState * pstate,List * orderlist,List ** targetlist,ParseExprKind exprKind,bool useSQL99)2617 transformSortClause(ParseState *pstate,
2618 					List *orderlist,
2619 					List **targetlist,
2620 					ParseExprKind exprKind,
2621 					bool useSQL99)
2622 {
2623 	List	   *sortlist = NIL;
2624 	ListCell   *olitem;
2625 
2626 	foreach(olitem, orderlist)
2627 	{
2628 		SortBy	   *sortby = (SortBy *) lfirst(olitem);
2629 		TargetEntry *tle;
2630 
2631 		if (useSQL99)
2632 			tle = findTargetlistEntrySQL99(pstate, sortby->node,
2633 										   targetlist, exprKind);
2634 		else
2635 			tle = findTargetlistEntrySQL92(pstate, sortby->node,
2636 										   targetlist, exprKind);
2637 
2638 		sortlist = addTargetToSortList(pstate, tle,
2639 									   sortlist, *targetlist, sortby);
2640 	}
2641 
2642 	return sortlist;
2643 }
2644 
2645 /*
2646  * transformWindowDefinitions -
2647  *		transform window definitions (WindowDef to WindowClause)
2648  */
2649 List *
transformWindowDefinitions(ParseState * pstate,List * windowdefs,List ** targetlist)2650 transformWindowDefinitions(ParseState *pstate,
2651 						   List *windowdefs,
2652 						   List **targetlist)
2653 {
2654 	List	   *result = NIL;
2655 	Index		winref = 0;
2656 	ListCell   *lc;
2657 
2658 	foreach(lc, windowdefs)
2659 	{
2660 		WindowDef  *windef = (WindowDef *) lfirst(lc);
2661 		WindowClause *refwc = NULL;
2662 		List	   *partitionClause;
2663 		List	   *orderClause;
2664 		Oid			rangeopfamily = InvalidOid;
2665 		Oid			rangeopcintype = InvalidOid;
2666 		WindowClause *wc;
2667 
2668 		winref++;
2669 
2670 		/*
2671 		 * Check for duplicate window names.
2672 		 */
2673 		if (windef->name &&
2674 			findWindowClause(result, windef->name) != NULL)
2675 			ereport(ERROR,
2676 					(errcode(ERRCODE_WINDOWING_ERROR),
2677 					 errmsg("window \"%s\" is already defined", windef->name),
2678 					 parser_errposition(pstate, windef->location)));
2679 
2680 		/*
2681 		 * If it references a previous window, look that up.
2682 		 */
2683 		if (windef->refname)
2684 		{
2685 			refwc = findWindowClause(result, windef->refname);
2686 			if (refwc == NULL)
2687 				ereport(ERROR,
2688 						(errcode(ERRCODE_UNDEFINED_OBJECT),
2689 						 errmsg("window \"%s\" does not exist",
2690 								windef->refname),
2691 						 parser_errposition(pstate, windef->location)));
2692 		}
2693 
2694 		/*
2695 		 * Transform PARTITION and ORDER specs, if any.  These are treated
2696 		 * almost exactly like top-level GROUP BY and ORDER BY clauses,
2697 		 * including the special handling of nondefault operator semantics.
2698 		 */
2699 		orderClause = transformSortClause(pstate,
2700 										  windef->orderClause,
2701 										  targetlist,
2702 										  EXPR_KIND_WINDOW_ORDER,
2703 										  true /* force SQL99 rules */ );
2704 		partitionClause = transformGroupClause(pstate,
2705 											   windef->partitionClause,
2706 											   NULL,
2707 											   targetlist,
2708 											   orderClause,
2709 											   EXPR_KIND_WINDOW_PARTITION,
2710 											   true /* force SQL99 rules */ );
2711 
2712 		/*
2713 		 * And prepare the new WindowClause.
2714 		 */
2715 		wc = makeNode(WindowClause);
2716 		wc->name = windef->name;
2717 		wc->refname = windef->refname;
2718 
2719 		/*
2720 		 * Per spec, a windowdef that references a previous one copies the
2721 		 * previous partition clause (and mustn't specify its own).  It can
2722 		 * specify its own ordering clause, but only if the previous one had
2723 		 * none.  It always specifies its own frame clause, and the previous
2724 		 * one must not have a frame clause.  Yeah, it's bizarre that each of
2725 		 * these cases works differently, but SQL:2008 says so; see 7.11
2726 		 * <window clause> syntax rule 10 and general rule 1.  The frame
2727 		 * clause rule is especially bizarre because it makes "OVER foo"
2728 		 * different from "OVER (foo)", and requires the latter to throw an
2729 		 * error if foo has a nondefault frame clause.  Well, ours not to
2730 		 * reason why, but we do go out of our way to throw a useful error
2731 		 * message for such cases.
2732 		 */
2733 		if (refwc)
2734 		{
2735 			if (partitionClause)
2736 				ereport(ERROR,
2737 						(errcode(ERRCODE_WINDOWING_ERROR),
2738 						 errmsg("cannot override PARTITION BY clause of window \"%s\"",
2739 								windef->refname),
2740 						 parser_errposition(pstate, windef->location)));
2741 			wc->partitionClause = copyObject(refwc->partitionClause);
2742 		}
2743 		else
2744 			wc->partitionClause = partitionClause;
2745 		if (refwc)
2746 		{
2747 			if (orderClause && refwc->orderClause)
2748 				ereport(ERROR,
2749 						(errcode(ERRCODE_WINDOWING_ERROR),
2750 						 errmsg("cannot override ORDER BY clause of window \"%s\"",
2751 								windef->refname),
2752 						 parser_errposition(pstate, windef->location)));
2753 			if (orderClause)
2754 			{
2755 				wc->orderClause = orderClause;
2756 				wc->copiedOrder = false;
2757 			}
2758 			else
2759 			{
2760 				wc->orderClause = copyObject(refwc->orderClause);
2761 				wc->copiedOrder = true;
2762 			}
2763 		}
2764 		else
2765 		{
2766 			wc->orderClause = orderClause;
2767 			wc->copiedOrder = false;
2768 		}
2769 		if (refwc && refwc->frameOptions != FRAMEOPTION_DEFAULTS)
2770 		{
2771 			/*
2772 			 * Use this message if this is a WINDOW clause, or if it's an OVER
2773 			 * clause that includes ORDER BY or framing clauses.  (We already
2774 			 * rejected PARTITION BY above, so no need to check that.)
2775 			 */
2776 			if (windef->name ||
2777 				orderClause || windef->frameOptions != FRAMEOPTION_DEFAULTS)
2778 				ereport(ERROR,
2779 						(errcode(ERRCODE_WINDOWING_ERROR),
2780 						 errmsg("cannot copy window \"%s\" because it has a frame clause",
2781 								windef->refname),
2782 						 parser_errposition(pstate, windef->location)));
2783 			/* Else this clause is just OVER (foo), so say this: */
2784 			ereport(ERROR,
2785 					(errcode(ERRCODE_WINDOWING_ERROR),
2786 					 errmsg("cannot copy window \"%s\" because it has a frame clause",
2787 							windef->refname),
2788 					 errhint("Omit the parentheses in this OVER clause."),
2789 					 parser_errposition(pstate, windef->location)));
2790 		}
2791 		wc->frameOptions = windef->frameOptions;
2792 
2793 		/*
2794 		 * RANGE offset PRECEDING/FOLLOWING requires exactly one ORDER BY
2795 		 * column; check that and get its sort opfamily info.
2796 		 */
2797 		if ((wc->frameOptions & FRAMEOPTION_RANGE) &&
2798 			(wc->frameOptions & (FRAMEOPTION_START_OFFSET |
2799 								 FRAMEOPTION_END_OFFSET)))
2800 		{
2801 			SortGroupClause *sortcl;
2802 			Node	   *sortkey;
2803 			int16		rangestrategy;
2804 
2805 			if (list_length(wc->orderClause) != 1)
2806 				ereport(ERROR,
2807 						(errcode(ERRCODE_WINDOWING_ERROR),
2808 						 errmsg("RANGE with offset PRECEDING/FOLLOWING requires exactly one ORDER BY column"),
2809 						 parser_errposition(pstate, windef->location)));
2810 			sortcl = castNode(SortGroupClause, linitial(wc->orderClause));
2811 			sortkey = get_sortgroupclause_expr(sortcl, *targetlist);
2812 			/* Find the sort operator in pg_amop */
2813 			if (!get_ordering_op_properties(sortcl->sortop,
2814 											&rangeopfamily,
2815 											&rangeopcintype,
2816 											&rangestrategy))
2817 				elog(ERROR, "operator %u is not a valid ordering operator",
2818 					 sortcl->sortop);
2819 			/* Record properties of sort ordering */
2820 			wc->inRangeColl = exprCollation(sortkey);
2821 			wc->inRangeAsc = (rangestrategy == BTLessStrategyNumber);
2822 			wc->inRangeNullsFirst = sortcl->nulls_first;
2823 		}
2824 
2825 		/* Per spec, GROUPS mode requires an ORDER BY clause */
2826 		if (wc->frameOptions & FRAMEOPTION_GROUPS)
2827 		{
2828 			if (wc->orderClause == NIL)
2829 				ereport(ERROR,
2830 						(errcode(ERRCODE_WINDOWING_ERROR),
2831 						 errmsg("GROUPS mode requires an ORDER BY clause"),
2832 						 parser_errposition(pstate, windef->location)));
2833 		}
2834 
2835 		/* Process frame offset expressions */
2836 		wc->startOffset = transformFrameOffset(pstate, wc->frameOptions,
2837 											   rangeopfamily, rangeopcintype,
2838 											   &wc->startInRangeFunc,
2839 											   windef->startOffset);
2840 		wc->endOffset = transformFrameOffset(pstate, wc->frameOptions,
2841 											 rangeopfamily, rangeopcintype,
2842 											 &wc->endInRangeFunc,
2843 											 windef->endOffset);
2844 		wc->winref = winref;
2845 
2846 		result = lappend(result, wc);
2847 	}
2848 
2849 	return result;
2850 }
2851 
2852 /*
2853  * transformDistinctClause -
2854  *	  transform a DISTINCT clause
2855  *
2856  * Since we may need to add items to the query's targetlist, that list
2857  * is passed by reference.
2858  *
2859  * As with GROUP BY, we absorb the sorting semantics of ORDER BY as much as
2860  * possible into the distinctClause.  This avoids a possible need to re-sort,
2861  * and allows the user to choose the equality semantics used by DISTINCT,
2862  * should she be working with a datatype that has more than one equality
2863  * operator.
2864  *
2865  * is_agg is true if we are transforming an aggregate(DISTINCT ...)
2866  * function call.  This does not affect any behavior, only the phrasing
2867  * of error messages.
2868  */
2869 List *
transformDistinctClause(ParseState * pstate,List ** targetlist,List * sortClause,bool is_agg)2870 transformDistinctClause(ParseState *pstate,
2871 						List **targetlist, List *sortClause, bool is_agg)
2872 {
2873 	List	   *result = NIL;
2874 	ListCell   *slitem;
2875 	ListCell   *tlitem;
2876 
2877 	/*
2878 	 * The distinctClause should consist of all ORDER BY items followed by all
2879 	 * other non-resjunk targetlist items.  There must not be any resjunk
2880 	 * ORDER BY items --- that would imply that we are sorting by a value that
2881 	 * isn't necessarily unique within a DISTINCT group, so the results
2882 	 * wouldn't be well-defined.  This construction ensures we follow the rule
2883 	 * that sortClause and distinctClause match; in fact the sortClause will
2884 	 * always be a prefix of distinctClause.
2885 	 *
2886 	 * Note a corner case: the same TLE could be in the ORDER BY list multiple
2887 	 * times with different sortops.  We have to include it in the
2888 	 * distinctClause the same way to preserve the prefix property. The net
2889 	 * effect will be that the TLE value will be made unique according to both
2890 	 * sortops.
2891 	 */
2892 	foreach(slitem, sortClause)
2893 	{
2894 		SortGroupClause *scl = (SortGroupClause *) lfirst(slitem);
2895 		TargetEntry *tle = get_sortgroupclause_tle(scl, *targetlist);
2896 
2897 		if (tle->resjunk)
2898 			ereport(ERROR,
2899 					(errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
2900 					 is_agg ?
2901 					 errmsg("in an aggregate with DISTINCT, ORDER BY expressions must appear in argument list") :
2902 					 errmsg("for SELECT DISTINCT, ORDER BY expressions must appear in select list"),
2903 					 parser_errposition(pstate,
2904 										exprLocation((Node *) tle->expr))));
2905 		result = lappend(result, copyObject(scl));
2906 	}
2907 
2908 	/*
2909 	 * Now add any remaining non-resjunk tlist items, using default sort/group
2910 	 * semantics for their data types.
2911 	 */
2912 	foreach(tlitem, *targetlist)
2913 	{
2914 		TargetEntry *tle = (TargetEntry *) lfirst(tlitem);
2915 
2916 		if (tle->resjunk)
2917 			continue;			/* ignore junk */
2918 		result = addTargetToGroupList(pstate, tle,
2919 									  result, *targetlist,
2920 									  exprLocation((Node *) tle->expr));
2921 	}
2922 
2923 	/*
2924 	 * Complain if we found nothing to make DISTINCT.  Returning an empty list
2925 	 * would cause the parsed Query to look like it didn't have DISTINCT, with
2926 	 * results that would probably surprise the user.  Note: this case is
2927 	 * presently impossible for aggregates because of grammar restrictions,
2928 	 * but we check anyway.
2929 	 */
2930 	if (result == NIL)
2931 		ereport(ERROR,
2932 				(errcode(ERRCODE_SYNTAX_ERROR),
2933 				 is_agg ?
2934 				 errmsg("an aggregate with DISTINCT must have at least one argument") :
2935 				 errmsg("SELECT DISTINCT must have at least one column")));
2936 
2937 	return result;
2938 }
2939 
2940 /*
2941  * transformDistinctOnClause -
2942  *	  transform a DISTINCT ON clause
2943  *
2944  * Since we may need to add items to the query's targetlist, that list
2945  * is passed by reference.
2946  *
2947  * As with GROUP BY, we absorb the sorting semantics of ORDER BY as much as
2948  * possible into the distinctClause.  This avoids a possible need to re-sort,
2949  * and allows the user to choose the equality semantics used by DISTINCT,
2950  * should she be working with a datatype that has more than one equality
2951  * operator.
2952  */
2953 List *
transformDistinctOnClause(ParseState * pstate,List * distinctlist,List ** targetlist,List * sortClause)2954 transformDistinctOnClause(ParseState *pstate, List *distinctlist,
2955 						  List **targetlist, List *sortClause)
2956 {
2957 	List	   *result = NIL;
2958 	List	   *sortgrouprefs = NIL;
2959 	bool		skipped_sortitem;
2960 	ListCell   *lc;
2961 	ListCell   *lc2;
2962 
2963 	/*
2964 	 * Add all the DISTINCT ON expressions to the tlist (if not already
2965 	 * present, they are added as resjunk items).  Assign sortgroupref numbers
2966 	 * to them, and make a list of these numbers.  (NB: we rely below on the
2967 	 * sortgrouprefs list being one-for-one with the original distinctlist.
2968 	 * Also notice that we could have duplicate DISTINCT ON expressions and
2969 	 * hence duplicate entries in sortgrouprefs.)
2970 	 */
2971 	foreach(lc, distinctlist)
2972 	{
2973 		Node	   *dexpr = (Node *) lfirst(lc);
2974 		int			sortgroupref;
2975 		TargetEntry *tle;
2976 
2977 		tle = findTargetlistEntrySQL92(pstate, dexpr, targetlist,
2978 									   EXPR_KIND_DISTINCT_ON);
2979 		sortgroupref = assignSortGroupRef(tle, *targetlist);
2980 		sortgrouprefs = lappend_int(sortgrouprefs, sortgroupref);
2981 	}
2982 
2983 	/*
2984 	 * If the user writes both DISTINCT ON and ORDER BY, adopt the sorting
2985 	 * semantics from ORDER BY items that match DISTINCT ON items, and also
2986 	 * adopt their column sort order.  We insist that the distinctClause and
2987 	 * sortClause match, so throw error if we find the need to add any more
2988 	 * distinctClause items after we've skipped an ORDER BY item that wasn't
2989 	 * in DISTINCT ON.
2990 	 */
2991 	skipped_sortitem = false;
2992 	foreach(lc, sortClause)
2993 	{
2994 		SortGroupClause *scl = (SortGroupClause *) lfirst(lc);
2995 
2996 		if (list_member_int(sortgrouprefs, scl->tleSortGroupRef))
2997 		{
2998 			if (skipped_sortitem)
2999 				ereport(ERROR,
3000 						(errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3001 						 errmsg("SELECT DISTINCT ON expressions must match initial ORDER BY expressions"),
3002 						 parser_errposition(pstate,
3003 											get_matching_location(scl->tleSortGroupRef,
3004 																  sortgrouprefs,
3005 																  distinctlist))));
3006 			else
3007 				result = lappend(result, copyObject(scl));
3008 		}
3009 		else
3010 			skipped_sortitem = true;
3011 	}
3012 
3013 	/*
3014 	 * Now add any remaining DISTINCT ON items, using default sort/group
3015 	 * semantics for their data types.  (Note: this is pretty questionable; if
3016 	 * the ORDER BY list doesn't include all the DISTINCT ON items and more
3017 	 * besides, you certainly aren't using DISTINCT ON in the intended way,
3018 	 * and you probably aren't going to get consistent results.  It might be
3019 	 * better to throw an error or warning here.  But historically we've
3020 	 * allowed it, so keep doing so.)
3021 	 */
3022 	forboth(lc, distinctlist, lc2, sortgrouprefs)
3023 	{
3024 		Node	   *dexpr = (Node *) lfirst(lc);
3025 		int			sortgroupref = lfirst_int(lc2);
3026 		TargetEntry *tle = get_sortgroupref_tle(sortgroupref, *targetlist);
3027 
3028 		if (targetIsInSortList(tle, InvalidOid, result))
3029 			continue;			/* already in list (with some semantics) */
3030 		if (skipped_sortitem)
3031 			ereport(ERROR,
3032 					(errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3033 					 errmsg("SELECT DISTINCT ON expressions must match initial ORDER BY expressions"),
3034 					 parser_errposition(pstate, exprLocation(dexpr))));
3035 		result = addTargetToGroupList(pstate, tle,
3036 									  result, *targetlist,
3037 									  exprLocation(dexpr));
3038 	}
3039 
3040 	/*
3041 	 * An empty result list is impossible here because of grammar
3042 	 * restrictions.
3043 	 */
3044 	Assert(result != NIL);
3045 
3046 	return result;
3047 }
3048 
3049 /*
3050  * get_matching_location
3051  *		Get the exprLocation of the exprs member corresponding to the
3052  *		(first) member of sortgrouprefs that equals sortgroupref.
3053  *
3054  * This is used so that we can point at a troublesome DISTINCT ON entry.
3055  * (Note that we need to use the original untransformed DISTINCT ON list
3056  * item, as whatever TLE it corresponds to will very possibly have a
3057  * parse location pointing to some matching entry in the SELECT list
3058  * or ORDER BY list.)
3059  */
3060 static int
get_matching_location(int sortgroupref,List * sortgrouprefs,List * exprs)3061 get_matching_location(int sortgroupref, List *sortgrouprefs, List *exprs)
3062 {
3063 	ListCell   *lcs;
3064 	ListCell   *lce;
3065 
3066 	forboth(lcs, sortgrouprefs, lce, exprs)
3067 	{
3068 		if (lfirst_int(lcs) == sortgroupref)
3069 			return exprLocation((Node *) lfirst(lce));
3070 	}
3071 	/* if no match, caller blew it */
3072 	elog(ERROR, "get_matching_location: no matching sortgroupref");
3073 	return -1;					/* keep compiler quiet */
3074 }
3075 
3076 /*
3077  * resolve_unique_index_expr
3078  *		Infer a unique index from a list of indexElems, for ON
3079  *		CONFLICT clause
3080  *
3081  * Perform parse analysis of expressions and columns appearing within ON
3082  * CONFLICT clause.  During planning, the returned list of expressions is used
3083  * to infer which unique index to use.
3084  */
3085 static List *
resolve_unique_index_expr(ParseState * pstate,InferClause * infer,Relation heapRel)3086 resolve_unique_index_expr(ParseState *pstate, InferClause *infer,
3087 						  Relation heapRel)
3088 {
3089 	List	   *result = NIL;
3090 	ListCell   *l;
3091 
3092 	foreach(l, infer->indexElems)
3093 	{
3094 		IndexElem  *ielem = (IndexElem *) lfirst(l);
3095 		InferenceElem *pInfer = makeNode(InferenceElem);
3096 		Node	   *parse;
3097 
3098 		/*
3099 		 * Raw grammar re-uses CREATE INDEX infrastructure for unique index
3100 		 * inference clause, and so will accept opclasses by name and so on.
3101 		 *
3102 		 * Make no attempt to match ASC or DESC ordering or NULLS FIRST/NULLS
3103 		 * LAST ordering, since those are not significant for inference
3104 		 * purposes (any unique index matching the inference specification in
3105 		 * other regards is accepted indifferently).  Actively reject this as
3106 		 * wrong-headed.
3107 		 */
3108 		if (ielem->ordering != SORTBY_DEFAULT)
3109 			ereport(ERROR,
3110 					(errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3111 					 errmsg("ASC/DESC is not allowed in ON CONFLICT clause"),
3112 					 parser_errposition(pstate,
3113 										exprLocation((Node *) infer))));
3114 		if (ielem->nulls_ordering != SORTBY_NULLS_DEFAULT)
3115 			ereport(ERROR,
3116 					(errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3117 					 errmsg("NULLS FIRST/LAST is not allowed in ON CONFLICT clause"),
3118 					 parser_errposition(pstate,
3119 										exprLocation((Node *) infer))));
3120 
3121 		if (!ielem->expr)
3122 		{
3123 			/* Simple index attribute */
3124 			ColumnRef  *n;
3125 
3126 			/*
3127 			 * Grammar won't have built raw expression for us in event of
3128 			 * plain column reference.  Create one directly, and perform
3129 			 * expression transformation.  Planner expects this, and performs
3130 			 * its own normalization for the purposes of matching against
3131 			 * pg_index.
3132 			 */
3133 			n = makeNode(ColumnRef);
3134 			n->fields = list_make1(makeString(ielem->name));
3135 			/* Location is approximately that of inference specification */
3136 			n->location = infer->location;
3137 			parse = (Node *) n;
3138 		}
3139 		else
3140 		{
3141 			/* Do parse transformation of the raw expression */
3142 			parse = (Node *) ielem->expr;
3143 		}
3144 
3145 		/*
3146 		 * transformExpr() will reject subqueries, aggregates, window
3147 		 * functions, and SRFs, based on being passed
3148 		 * EXPR_KIND_INDEX_EXPRESSION.  So we needn't worry about those
3149 		 * further ... not that they would match any available index
3150 		 * expression anyway.
3151 		 */
3152 		pInfer->expr = transformExpr(pstate, parse, EXPR_KIND_INDEX_EXPRESSION);
3153 
3154 		/* Perform lookup of collation and operator class as required */
3155 		if (!ielem->collation)
3156 			pInfer->infercollid = InvalidOid;
3157 		else
3158 			pInfer->infercollid = LookupCollation(pstate, ielem->collation,
3159 												  exprLocation(pInfer->expr));
3160 
3161 		if (!ielem->opclass)
3162 			pInfer->inferopclass = InvalidOid;
3163 		else
3164 			pInfer->inferopclass = get_opclass_oid(BTREE_AM_OID,
3165 												   ielem->opclass, false);
3166 
3167 		result = lappend(result, pInfer);
3168 	}
3169 
3170 	return result;
3171 }
3172 
3173 /*
3174  * transformOnConflictArbiter -
3175  *		transform arbiter expressions in an ON CONFLICT clause.
3176  *
3177  * Transformed expressions used to infer one unique index relation to serve as
3178  * an ON CONFLICT arbiter.  Partial unique indexes may be inferred using WHERE
3179  * clause from inference specification clause.
3180  */
3181 void
transformOnConflictArbiter(ParseState * pstate,OnConflictClause * onConflictClause,List ** arbiterExpr,Node ** arbiterWhere,Oid * constraint)3182 transformOnConflictArbiter(ParseState *pstate,
3183 						   OnConflictClause *onConflictClause,
3184 						   List **arbiterExpr, Node **arbiterWhere,
3185 						   Oid *constraint)
3186 {
3187 	InferClause *infer = onConflictClause->infer;
3188 
3189 	*arbiterExpr = NIL;
3190 	*arbiterWhere = NULL;
3191 	*constraint = InvalidOid;
3192 
3193 	if (onConflictClause->action == ONCONFLICT_UPDATE && !infer)
3194 		ereport(ERROR,
3195 				(errcode(ERRCODE_SYNTAX_ERROR),
3196 				 errmsg("ON CONFLICT DO UPDATE requires inference specification or constraint name"),
3197 				 errhint("For example, ON CONFLICT (column_name)."),
3198 				 parser_errposition(pstate,
3199 									exprLocation((Node *) onConflictClause))));
3200 
3201 	/*
3202 	 * To simplify certain aspects of its design, speculative insertion into
3203 	 * system catalogs is disallowed
3204 	 */
3205 	if (IsCatalogRelation(pstate->p_target_relation))
3206 		ereport(ERROR,
3207 				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3208 				 errmsg("ON CONFLICT is not supported with system catalog tables"),
3209 				 parser_errposition(pstate,
3210 									exprLocation((Node *) onConflictClause))));
3211 
3212 	/* Same applies to table used by logical decoding as catalog table */
3213 	if (RelationIsUsedAsCatalogTable(pstate->p_target_relation))
3214 		ereport(ERROR,
3215 				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3216 				 errmsg("ON CONFLICT is not supported on table \"%s\" used as a catalog table",
3217 						RelationGetRelationName(pstate->p_target_relation)),
3218 				 parser_errposition(pstate,
3219 									exprLocation((Node *) onConflictClause))));
3220 
3221 	/* ON CONFLICT DO NOTHING does not require an inference clause */
3222 	if (infer)
3223 	{
3224 		if (infer->indexElems)
3225 			*arbiterExpr = resolve_unique_index_expr(pstate, infer,
3226 													 pstate->p_target_relation);
3227 
3228 		/*
3229 		 * Handling inference WHERE clause (for partial unique index
3230 		 * inference)
3231 		 */
3232 		if (infer->whereClause)
3233 			*arbiterWhere = transformExpr(pstate, infer->whereClause,
3234 										  EXPR_KIND_INDEX_PREDICATE);
3235 
3236 		/*
3237 		 * If the arbiter is specified by constraint name, get the constraint
3238 		 * OID and mark the constrained columns as requiring SELECT privilege,
3239 		 * in the same way as would have happened if the arbiter had been
3240 		 * specified by explicit reference to the constraint's index columns.
3241 		 */
3242 		if (infer->conname)
3243 		{
3244 			Oid			relid = RelationGetRelid(pstate->p_target_relation);
3245 			RangeTblEntry *rte = pstate->p_target_nsitem->p_rte;
3246 			Bitmapset  *conattnos;
3247 
3248 			conattnos = get_relation_constraint_attnos(relid, infer->conname,
3249 													   false, constraint);
3250 
3251 			/* Make sure the rel as a whole is marked for SELECT access */
3252 			rte->requiredPerms |= ACL_SELECT;
3253 			/* Mark the constrained columns as requiring SELECT access */
3254 			rte->selectedCols = bms_add_members(rte->selectedCols, conattnos);
3255 		}
3256 	}
3257 
3258 	/*
3259 	 * It's convenient to form a list of expressions based on the
3260 	 * representation used by CREATE INDEX, since the same restrictions are
3261 	 * appropriate (e.g. on subqueries).  However, from here on, a dedicated
3262 	 * primnode representation is used for inference elements, and so
3263 	 * assign_query_collations() can be trusted to do the right thing with the
3264 	 * post parse analysis query tree inference clause representation.
3265 	 */
3266 }
3267 
3268 /*
3269  * addTargetToSortList
3270  *		If the given targetlist entry isn't already in the SortGroupClause
3271  *		list, add it to the end of the list, using the given sort ordering
3272  *		info.
3273  *
3274  * Returns the updated SortGroupClause list.
3275  */
3276 List *
addTargetToSortList(ParseState * pstate,TargetEntry * tle,List * sortlist,List * targetlist,SortBy * sortby)3277 addTargetToSortList(ParseState *pstate, TargetEntry *tle,
3278 					List *sortlist, List *targetlist, SortBy *sortby)
3279 {
3280 	Oid			restype = exprType((Node *) tle->expr);
3281 	Oid			sortop;
3282 	Oid			eqop;
3283 	bool		hashable;
3284 	bool		reverse;
3285 	int			location;
3286 	ParseCallbackState pcbstate;
3287 
3288 	/* if tlist item is an UNKNOWN literal, change it to TEXT */
3289 	if (restype == UNKNOWNOID)
3290 	{
3291 		tle->expr = (Expr *) coerce_type(pstate, (Node *) tle->expr,
3292 										 restype, TEXTOID, -1,
3293 										 COERCION_IMPLICIT,
3294 										 COERCE_IMPLICIT_CAST,
3295 										 -1);
3296 		restype = TEXTOID;
3297 	}
3298 
3299 	/*
3300 	 * Rather than clutter the API of get_sort_group_operators and the other
3301 	 * functions we're about to use, make use of error context callback to
3302 	 * mark any error reports with a parse position.  We point to the operator
3303 	 * location if present, else to the expression being sorted.  (NB: use the
3304 	 * original untransformed expression here; the TLE entry might well point
3305 	 * at a duplicate expression in the regular SELECT list.)
3306 	 */
3307 	location = sortby->location;
3308 	if (location < 0)
3309 		location = exprLocation(sortby->node);
3310 	setup_parser_errposition_callback(&pcbstate, pstate, location);
3311 
3312 	/* determine the sortop, eqop, and directionality */
3313 	switch (sortby->sortby_dir)
3314 	{
3315 		case SORTBY_DEFAULT:
3316 		case SORTBY_ASC:
3317 			get_sort_group_operators(restype,
3318 									 true, true, false,
3319 									 &sortop, &eqop, NULL,
3320 									 &hashable);
3321 			reverse = false;
3322 			break;
3323 		case SORTBY_DESC:
3324 			get_sort_group_operators(restype,
3325 									 false, true, true,
3326 									 NULL, &eqop, &sortop,
3327 									 &hashable);
3328 			reverse = true;
3329 			break;
3330 		case SORTBY_USING:
3331 			Assert(sortby->useOp != NIL);
3332 			sortop = compatible_oper_opid(sortby->useOp,
3333 										  restype,
3334 										  restype,
3335 										  false);
3336 
3337 			/*
3338 			 * Verify it's a valid ordering operator, fetch the corresponding
3339 			 * equality operator, and determine whether to consider it like
3340 			 * ASC or DESC.
3341 			 */
3342 			eqop = get_equality_op_for_ordering_op(sortop, &reverse);
3343 			if (!OidIsValid(eqop))
3344 				ereport(ERROR,
3345 						(errcode(ERRCODE_WRONG_OBJECT_TYPE),
3346 						 errmsg("operator %s is not a valid ordering operator",
3347 								strVal(llast(sortby->useOp))),
3348 						 errhint("Ordering operators must be \"<\" or \">\" members of btree operator families.")));
3349 
3350 			/*
3351 			 * Also see if the equality operator is hashable.
3352 			 */
3353 			hashable = op_hashjoinable(eqop, restype);
3354 			break;
3355 		default:
3356 			elog(ERROR, "unrecognized sortby_dir: %d", sortby->sortby_dir);
3357 			sortop = InvalidOid;	/* keep compiler quiet */
3358 			eqop = InvalidOid;
3359 			hashable = false;
3360 			reverse = false;
3361 			break;
3362 	}
3363 
3364 	cancel_parser_errposition_callback(&pcbstate);
3365 
3366 	/* avoid making duplicate sortlist entries */
3367 	if (!targetIsInSortList(tle, sortop, sortlist))
3368 	{
3369 		SortGroupClause *sortcl = makeNode(SortGroupClause);
3370 
3371 		sortcl->tleSortGroupRef = assignSortGroupRef(tle, targetlist);
3372 
3373 		sortcl->eqop = eqop;
3374 		sortcl->sortop = sortop;
3375 		sortcl->hashable = hashable;
3376 
3377 		switch (sortby->sortby_nulls)
3378 		{
3379 			case SORTBY_NULLS_DEFAULT:
3380 				/* NULLS FIRST is default for DESC; other way for ASC */
3381 				sortcl->nulls_first = reverse;
3382 				break;
3383 			case SORTBY_NULLS_FIRST:
3384 				sortcl->nulls_first = true;
3385 				break;
3386 			case SORTBY_NULLS_LAST:
3387 				sortcl->nulls_first = false;
3388 				break;
3389 			default:
3390 				elog(ERROR, "unrecognized sortby_nulls: %d",
3391 					 sortby->sortby_nulls);
3392 				break;
3393 		}
3394 
3395 		sortlist = lappend(sortlist, sortcl);
3396 	}
3397 
3398 	return sortlist;
3399 }
3400 
3401 /*
3402  * addTargetToGroupList
3403  *		If the given targetlist entry isn't already in the SortGroupClause
3404  *		list, add it to the end of the list, using default sort/group
3405  *		semantics.
3406  *
3407  * This is very similar to addTargetToSortList, except that we allow the
3408  * case where only a grouping (equality) operator can be found, and that
3409  * the TLE is considered "already in the list" if it appears there with any
3410  * sorting semantics.
3411  *
3412  * location is the parse location to be fingered in event of trouble.  Note
3413  * that we can't rely on exprLocation(tle->expr), because that might point
3414  * to a SELECT item that matches the GROUP BY item; it'd be pretty confusing
3415  * to report such a location.
3416  *
3417  * Returns the updated SortGroupClause list.
3418  */
3419 static List *
addTargetToGroupList(ParseState * pstate,TargetEntry * tle,List * grouplist,List * targetlist,int location)3420 addTargetToGroupList(ParseState *pstate, TargetEntry *tle,
3421 					 List *grouplist, List *targetlist, int location)
3422 {
3423 	Oid			restype = exprType((Node *) tle->expr);
3424 
3425 	/* if tlist item is an UNKNOWN literal, change it to TEXT */
3426 	if (restype == UNKNOWNOID)
3427 	{
3428 		tle->expr = (Expr *) coerce_type(pstate, (Node *) tle->expr,
3429 										 restype, TEXTOID, -1,
3430 										 COERCION_IMPLICIT,
3431 										 COERCE_IMPLICIT_CAST,
3432 										 -1);
3433 		restype = TEXTOID;
3434 	}
3435 
3436 	/* avoid making duplicate grouplist entries */
3437 	if (!targetIsInSortList(tle, InvalidOid, grouplist))
3438 	{
3439 		SortGroupClause *grpcl = makeNode(SortGroupClause);
3440 		Oid			sortop;
3441 		Oid			eqop;
3442 		bool		hashable;
3443 		ParseCallbackState pcbstate;
3444 
3445 		setup_parser_errposition_callback(&pcbstate, pstate, location);
3446 
3447 		/* determine the eqop and optional sortop */
3448 		get_sort_group_operators(restype,
3449 								 false, true, false,
3450 								 &sortop, &eqop, NULL,
3451 								 &hashable);
3452 
3453 		cancel_parser_errposition_callback(&pcbstate);
3454 
3455 		grpcl->tleSortGroupRef = assignSortGroupRef(tle, targetlist);
3456 		grpcl->eqop = eqop;
3457 		grpcl->sortop = sortop;
3458 		grpcl->nulls_first = false; /* OK with or without sortop */
3459 		grpcl->hashable = hashable;
3460 
3461 		grouplist = lappend(grouplist, grpcl);
3462 	}
3463 
3464 	return grouplist;
3465 }
3466 
3467 /*
3468  * assignSortGroupRef
3469  *	  Assign the targetentry an unused ressortgroupref, if it doesn't
3470  *	  already have one.  Return the assigned or pre-existing refnumber.
3471  *
3472  * 'tlist' is the targetlist containing (or to contain) the given targetentry.
3473  */
3474 Index
assignSortGroupRef(TargetEntry * tle,List * tlist)3475 assignSortGroupRef(TargetEntry *tle, List *tlist)
3476 {
3477 	Index		maxRef;
3478 	ListCell   *l;
3479 
3480 	if (tle->ressortgroupref)	/* already has one? */
3481 		return tle->ressortgroupref;
3482 
3483 	/* easiest way to pick an unused refnumber: max used + 1 */
3484 	maxRef = 0;
3485 	foreach(l, tlist)
3486 	{
3487 		Index		ref = ((TargetEntry *) lfirst(l))->ressortgroupref;
3488 
3489 		if (ref > maxRef)
3490 			maxRef = ref;
3491 	}
3492 	tle->ressortgroupref = maxRef + 1;
3493 	return tle->ressortgroupref;
3494 }
3495 
3496 /*
3497  * targetIsInSortList
3498  *		Is the given target item already in the sortlist?
3499  *		If sortop is not InvalidOid, also test for a match to the sortop.
3500  *
3501  * It is not an oversight that this function ignores the nulls_first flag.
3502  * We check sortop when determining if an ORDER BY item is redundant with
3503  * earlier ORDER BY items, because it's conceivable that "ORDER BY
3504  * foo USING <, foo USING <<<" is not redundant, if <<< distinguishes
3505  * values that < considers equal.  We need not check nulls_first
3506  * however, because a lower-order column with the same sortop but
3507  * opposite nulls direction is redundant.  Also, we can consider
3508  * ORDER BY foo ASC, foo DESC redundant, so check for a commutator match.
3509  *
3510  * Works for both ordering and grouping lists (sortop would normally be
3511  * InvalidOid when considering grouping).  Note that the main reason we need
3512  * this routine (and not just a quick test for nonzeroness of ressortgroupref)
3513  * is that a TLE might be in only one of the lists.
3514  */
3515 bool
targetIsInSortList(TargetEntry * tle,Oid sortop,List * sortList)3516 targetIsInSortList(TargetEntry *tle, Oid sortop, List *sortList)
3517 {
3518 	Index		ref = tle->ressortgroupref;
3519 	ListCell   *l;
3520 
3521 	/* no need to scan list if tle has no marker */
3522 	if (ref == 0)
3523 		return false;
3524 
3525 	foreach(l, sortList)
3526 	{
3527 		SortGroupClause *scl = (SortGroupClause *) lfirst(l);
3528 
3529 		if (scl->tleSortGroupRef == ref &&
3530 			(sortop == InvalidOid ||
3531 			 sortop == scl->sortop ||
3532 			 sortop == get_commutator(scl->sortop)))
3533 			return true;
3534 	}
3535 	return false;
3536 }
3537 
3538 /*
3539  * findWindowClause
3540  *		Find the named WindowClause in the list, or return NULL if not there
3541  */
3542 static WindowClause *
findWindowClause(List * wclist,const char * name)3543 findWindowClause(List *wclist, const char *name)
3544 {
3545 	ListCell   *l;
3546 
3547 	foreach(l, wclist)
3548 	{
3549 		WindowClause *wc = (WindowClause *) lfirst(l);
3550 
3551 		if (wc->name && strcmp(wc->name, name) == 0)
3552 			return wc;
3553 	}
3554 
3555 	return NULL;
3556 }
3557 
3558 /*
3559  * transformFrameOffset
3560  *		Process a window frame offset expression
3561  *
3562  * In RANGE mode, rangeopfamily is the sort opfamily for the input ORDER BY
3563  * column, and rangeopcintype is the input data type the sort operator is
3564  * registered with.  We expect the in_range function to be registered with
3565  * that same type.  (In binary-compatible cases, it might be different from
3566  * the input column's actual type, so we can't use that for the lookups.)
3567  * We'll return the OID of the in_range function to *inRangeFunc.
3568  */
3569 static Node *
transformFrameOffset(ParseState * pstate,int frameOptions,Oid rangeopfamily,Oid rangeopcintype,Oid * inRangeFunc,Node * clause)3570 transformFrameOffset(ParseState *pstate, int frameOptions,
3571 					 Oid rangeopfamily, Oid rangeopcintype, Oid *inRangeFunc,
3572 					 Node *clause)
3573 {
3574 	const char *constructName = NULL;
3575 	Node	   *node;
3576 
3577 	*inRangeFunc = InvalidOid;	/* default result */
3578 
3579 	/* Quick exit if no offset expression */
3580 	if (clause == NULL)
3581 		return NULL;
3582 
3583 	if (frameOptions & FRAMEOPTION_ROWS)
3584 	{
3585 		/* Transform the raw expression tree */
3586 		node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_ROWS);
3587 
3588 		/*
3589 		 * Like LIMIT clause, simply coerce to int8
3590 		 */
3591 		constructName = "ROWS";
3592 		node = coerce_to_specific_type(pstate, node, INT8OID, constructName);
3593 	}
3594 	else if (frameOptions & FRAMEOPTION_RANGE)
3595 	{
3596 		/*
3597 		 * We must look up the in_range support function that's to be used,
3598 		 * possibly choosing one of several, and coerce the "offset" value to
3599 		 * the appropriate input type.
3600 		 */
3601 		Oid			nodeType;
3602 		Oid			preferredType;
3603 		int			nfuncs = 0;
3604 		int			nmatches = 0;
3605 		Oid			selectedType = InvalidOid;
3606 		Oid			selectedFunc = InvalidOid;
3607 		CatCList   *proclist;
3608 		int			i;
3609 
3610 		/* Transform the raw expression tree */
3611 		node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_RANGE);
3612 		nodeType = exprType(node);
3613 
3614 		/*
3615 		 * If there are multiple candidates, we'll prefer the one that exactly
3616 		 * matches nodeType; or if nodeType is as yet unknown, prefer the one
3617 		 * that exactly matches the sort column type.  (The second rule is
3618 		 * like what we do for "known_type operator unknown".)
3619 		 */
3620 		preferredType = (nodeType != UNKNOWNOID) ? nodeType : rangeopcintype;
3621 
3622 		/* Find the in_range support functions applicable to this case */
3623 		proclist = SearchSysCacheList2(AMPROCNUM,
3624 									   ObjectIdGetDatum(rangeopfamily),
3625 									   ObjectIdGetDatum(rangeopcintype));
3626 		for (i = 0; i < proclist->n_members; i++)
3627 		{
3628 			HeapTuple	proctup = &proclist->members[i]->tuple;
3629 			Form_pg_amproc procform = (Form_pg_amproc) GETSTRUCT(proctup);
3630 
3631 			/* The search will find all support proc types; ignore others */
3632 			if (procform->amprocnum != BTINRANGE_PROC)
3633 				continue;
3634 			nfuncs++;
3635 
3636 			/* Ignore function if given value can't be coerced to that type */
3637 			if (!can_coerce_type(1, &nodeType, &procform->amprocrighttype,
3638 								 COERCION_IMPLICIT))
3639 				continue;
3640 			nmatches++;
3641 
3642 			/* Remember preferred match, or any match if didn't find that */
3643 			if (selectedType != preferredType)
3644 			{
3645 				selectedType = procform->amprocrighttype;
3646 				selectedFunc = procform->amproc;
3647 			}
3648 		}
3649 		ReleaseCatCacheList(proclist);
3650 
3651 		/*
3652 		 * Throw error if needed.  It seems worth taking the trouble to
3653 		 * distinguish "no support at all" from "you didn't match any
3654 		 * available offset type".
3655 		 */
3656 		if (nfuncs == 0)
3657 			ereport(ERROR,
3658 					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3659 					 errmsg("RANGE with offset PRECEDING/FOLLOWING is not supported for column type %s",
3660 							format_type_be(rangeopcintype)),
3661 					 parser_errposition(pstate, exprLocation(node))));
3662 		if (nmatches == 0)
3663 			ereport(ERROR,
3664 					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3665 					 errmsg("RANGE with offset PRECEDING/FOLLOWING is not supported for column type %s and offset type %s",
3666 							format_type_be(rangeopcintype),
3667 							format_type_be(nodeType)),
3668 					 errhint("Cast the offset value to an appropriate type."),
3669 					 parser_errposition(pstate, exprLocation(node))));
3670 		if (nmatches != 1 && selectedType != preferredType)
3671 			ereport(ERROR,
3672 					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3673 					 errmsg("RANGE with offset PRECEDING/FOLLOWING has multiple interpretations for column type %s and offset type %s",
3674 							format_type_be(rangeopcintype),
3675 							format_type_be(nodeType)),
3676 					 errhint("Cast the offset value to the exact intended type."),
3677 					 parser_errposition(pstate, exprLocation(node))));
3678 
3679 		/* OK, coerce the offset to the right type */
3680 		constructName = "RANGE";
3681 		node = coerce_to_specific_type(pstate, node,
3682 									   selectedType, constructName);
3683 		*inRangeFunc = selectedFunc;
3684 	}
3685 	else if (frameOptions & FRAMEOPTION_GROUPS)
3686 	{
3687 		/* Transform the raw expression tree */
3688 		node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_GROUPS);
3689 
3690 		/*
3691 		 * Like LIMIT clause, simply coerce to int8
3692 		 */
3693 		constructName = "GROUPS";
3694 		node = coerce_to_specific_type(pstate, node, INT8OID, constructName);
3695 	}
3696 	else
3697 	{
3698 		Assert(false);
3699 		node = NULL;
3700 	}
3701 
3702 	/* Disallow variables in frame offsets */
3703 	checkExprIsVarFree(pstate, node, constructName);
3704 
3705 	return node;
3706 }
3707