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