Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * parse_clause.c
4 : * handle clauses in parser
5 : *
6 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/parser/parse_clause.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 :
16 : #include "postgres.h"
17 :
18 : #include "access/htup_details.h"
19 : #include "access/nbtree.h"
20 : #include "access/relation.h"
21 : #include "access/table.h"
22 : #include "access/tsmapi.h"
23 : #include "catalog/catalog.h"
24 : #include "catalog/pg_am.h"
25 : #include "catalog/pg_amproc.h"
26 : #include "catalog/pg_constraint.h"
27 : #include "catalog/pg_type.h"
28 : #include "commands/defrem.h"
29 : #include "miscadmin.h"
30 : #include "nodes/makefuncs.h"
31 : #include "nodes/nodeFuncs.h"
32 : #include "optimizer/optimizer.h"
33 : #include "parser/analyze.h"
34 : #include "parser/parse_clause.h"
35 : #include "parser/parse_coerce.h"
36 : #include "parser/parse_collate.h"
37 : #include "parser/parse_expr.h"
38 : #include "parser/parse_func.h"
39 : #include "parser/parse_graphtable.h"
40 : #include "parser/parse_oper.h"
41 : #include "parser/parse_relation.h"
42 : #include "parser/parse_target.h"
43 : #include "parser/parse_type.h"
44 : #include "parser/parser.h"
45 : #include "rewrite/rewriteManip.h"
46 : #include "utils/builtins.h"
47 : #include "utils/catcache.h"
48 : #include "utils/lsyscache.h"
49 : #include "utils/rel.h"
50 : #include "utils/syscache.h"
51 :
52 :
53 : static int extractRemainingColumns(ParseState *pstate,
54 : ParseNamespaceColumn *src_nscolumns,
55 : List *src_colnames,
56 : List **src_colnos,
57 : List **res_colnames, List **res_colvars,
58 : ParseNamespaceColumn *res_nscolumns);
59 : static Node *transformJoinUsingClause(ParseState *pstate,
60 : List *leftVars, List *rightVars);
61 : static Node *transformJoinOnClause(ParseState *pstate, JoinExpr *j,
62 : List *namespace);
63 : static ParseNamespaceItem *transformTableEntry(ParseState *pstate, RangeVar *r);
64 : static ParseNamespaceItem *transformRangeSubselect(ParseState *pstate,
65 : RangeSubselect *r);
66 : static ParseNamespaceItem *transformRangeFunction(ParseState *pstate,
67 : RangeFunction *r);
68 : static ParseNamespaceItem *transformRangeTableFunc(ParseState *pstate,
69 : RangeTableFunc *rtf);
70 : static ParseNamespaceItem *transformRangeGraphTable(ParseState *pstate,
71 : RangeGraphTable *rgt);
72 : static TableSampleClause *transformRangeTableSample(ParseState *pstate,
73 : RangeTableSample *rts);
74 : static ParseNamespaceItem *getNSItemForSpecialRelationTypes(ParseState *pstate,
75 : RangeVar *rv);
76 : static Node *transformFromClauseItem(ParseState *pstate, Node *n,
77 : ParseNamespaceItem **top_nsitem,
78 : List **namespace);
79 : static Var *buildVarFromNSColumn(ParseState *pstate,
80 : ParseNamespaceColumn *nscol);
81 : static Node *buildMergedJoinVar(ParseState *pstate, JoinType jointype,
82 : Var *l_colvar, Var *r_colvar);
83 : static void markRelsAsNulledBy(ParseState *pstate, Node *n, int jindex);
84 : static void setNamespaceColumnVisibility(List *namespace, bool cols_visible);
85 : static void setNamespaceLateralState(List *namespace,
86 : bool lateral_only, bool lateral_ok);
87 : static void checkExprIsVarFree(ParseState *pstate, Node *n,
88 : const char *constructName);
89 : static TargetEntry *findTargetlistEntrySQL92(ParseState *pstate, Node *node,
90 : List **tlist, ParseExprKind exprKind);
91 : static TargetEntry *findTargetlistEntrySQL99(ParseState *pstate, Node *node,
92 : List **tlist, ParseExprKind exprKind);
93 : static int get_matching_location(int sortgroupref,
94 : List *sortgrouprefs, List *exprs);
95 : static List *resolve_unique_index_expr(ParseState *pstate, InferClause *infer,
96 : Relation heapRel);
97 : static List *addTargetToGroupList(ParseState *pstate, TargetEntry *tle,
98 : List *grouplist, List *targetlist, int location);
99 : static WindowClause *findWindowClause(List *wclist, const char *name);
100 : static Node *transformFrameOffset(ParseState *pstate, int frameOptions,
101 : Oid rangeopfamily, Oid rangeopcintype, Oid *inRangeFunc,
102 : Node *clause);
103 :
104 :
105 : /*
106 : * transformFromClause -
107 : * Process the FROM clause and add items to the query's range table,
108 : * joinlist, and namespace.
109 : *
110 : * Note: we assume that the pstate's p_rtable, p_joinlist, and p_namespace
111 : * lists were initialized to NIL when the pstate was created.
112 : * We will add onto any entries already present --- this is needed for rule
113 : * processing, as well as for UPDATE and DELETE.
114 : */
115 : void
116 314855 : transformFromClause(ParseState *pstate, List *frmList)
117 : {
118 : ListCell *fl;
119 :
120 : /*
121 : * The grammar will have produced a list of RangeVars, RangeSubselects,
122 : * RangeFunctions, and/or JoinExprs. Transform each one (possibly adding
123 : * entries to the rtable), check for duplicate refnames, and then add it
124 : * to the joinlist and namespace.
125 : *
126 : * Note we must process the items left-to-right for proper handling of
127 : * LATERAL references.
128 : */
129 564089 : foreach(fl, frmList)
130 : {
131 249720 : Node *n = lfirst(fl);
132 : ParseNamespaceItem *nsitem;
133 : List *namespace;
134 :
135 249720 : n = transformFromClauseItem(pstate, n,
136 : &nsitem,
137 : &namespace);
138 :
139 249238 : checkNameSpaceConflicts(pstate, pstate->p_namespace, namespace);
140 :
141 : /* Mark the new namespace items as visible only to LATERAL */
142 249234 : setNamespaceLateralState(namespace, true, true);
143 :
144 249234 : pstate->p_joinlist = lappend(pstate->p_joinlist, n);
145 249234 : pstate->p_namespace = list_concat(pstate->p_namespace, namespace);
146 : }
147 :
148 : /*
149 : * We're done parsing the FROM list, so make all namespace items
150 : * unconditionally visible. Note that this will also reset lateral_only
151 : * for any namespace items that were already present when we were called;
152 : * but those should have been that way already.
153 : */
154 314369 : setNamespaceLateralState(pstate->p_namespace, false, true);
155 314369 : }
156 :
157 : /*
158 : * setTargetTable
159 : * Add the target relation of INSERT/UPDATE/DELETE/MERGE to the range table,
160 : * and make the special links to it in the ParseState.
161 : *
162 : * We also open the target relation and acquire a write lock on it.
163 : * This must be done before processing the FROM list, in case the target
164 : * is also mentioned as a source relation --- we want to be sure to grab
165 : * the write lock before any read lock.
166 : *
167 : * If alsoSource is true, add the target to the query's joinlist and
168 : * namespace. For INSERT, we don't want the target to be joined to;
169 : * it's a destination of tuples, not a source. MERGE is actually
170 : * both, but we'll add it separately to joinlist and namespace, so
171 : * doing nothing (like INSERT) is correct here. For UPDATE/DELETE,
172 : * we do need to scan or join the target. (NOTE: we do not bother
173 : * to check for namespace conflict; we assume that the namespace was
174 : * initially empty in these cases.)
175 : *
176 : * Finally, we mark the relation as requiring the permissions specified
177 : * by requiredPerms.
178 : *
179 : * Returns the rangetable index of the target relation.
180 : */
181 : int
182 56913 : setTargetTable(ParseState *pstate, RangeVar *relation,
183 : bool inh, bool alsoSource, AclMode requiredPerms)
184 : {
185 : ParseNamespaceItem *nsitem;
186 :
187 : /*
188 : * ENRs hide tables of the same name, so we need to check for them first.
189 : * In contrast, CTEs don't hide tables (for this purpose).
190 : */
191 107863 : if (relation->schemaname == NULL &&
192 50950 : scanNameSpaceForENR(pstate, relation->relname))
193 4 : ereport(ERROR,
194 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
195 : errmsg("relation \"%s\" cannot be the target of a modifying statement",
196 : relation->relname)));
197 :
198 : /* Close old target; this could only happen for multi-action rules */
199 56909 : if (pstate->p_target_relation != NULL)
200 0 : table_close(pstate->p_target_relation, NoLock);
201 :
202 : /*
203 : * Open target rel and grab suitable lock (which we will hold till end of
204 : * transaction).
205 : *
206 : * free_parsestate() will eventually do the corresponding table_close(),
207 : * but *not* release the lock.
208 : */
209 56909 : pstate->p_target_relation = parserOpenTable(pstate, relation,
210 : RowExclusiveLock);
211 :
212 : /*
213 : * Now build an RTE and a ParseNamespaceItem.
214 : */
215 56892 : nsitem = addRangeTableEntryForRelation(pstate, pstate->p_target_relation,
216 : RowExclusiveLock,
217 : relation->alias, inh, false);
218 :
219 : /* remember the RTE/nsitem as being the query target */
220 56892 : pstate->p_target_nsitem = nsitem;
221 :
222 : /*
223 : * Override addRangeTableEntry's default ACL_SELECT permissions check, and
224 : * instead mark target table as requiring exactly the specified
225 : * permissions.
226 : *
227 : * If we find an explicit reference to the rel later during parse
228 : * analysis, we will add the ACL_SELECT bit back again; see
229 : * markVarForSelectPriv and its callers.
230 : */
231 56892 : nsitem->p_perminfo->requiredPerms = requiredPerms;
232 :
233 : /*
234 : * If UPDATE/DELETE, add table to joinlist and namespace.
235 : */
236 56892 : if (alsoSource)
237 11453 : addNSItemToQuery(pstate, nsitem, true, true, true);
238 :
239 56892 : return nsitem->p_rtindex;
240 : }
241 :
242 : /*
243 : * Extract all not-in-common columns from column lists of a source table
244 : *
245 : * src_nscolumns and src_colnames describe the source table.
246 : *
247 : * *src_colnos initially contains the column numbers of the already-merged
248 : * columns. We add to it the column number of each additional column.
249 : * Also append to *res_colnames the name of each additional column,
250 : * append to *res_colvars a Var for each additional column, and copy the
251 : * columns' nscolumns data into res_nscolumns[] (which is caller-allocated
252 : * space that had better be big enough).
253 : *
254 : * Returns the number of columns added.
255 : */
256 : static int
257 110694 : extractRemainingColumns(ParseState *pstate,
258 : ParseNamespaceColumn *src_nscolumns,
259 : List *src_colnames,
260 : List **src_colnos,
261 : List **res_colnames, List **res_colvars,
262 : ParseNamespaceColumn *res_nscolumns)
263 : {
264 110694 : int colcount = 0;
265 : Bitmapset *prevcols;
266 : int attnum;
267 : ListCell *lc;
268 :
269 : /*
270 : * While we could just test "list_member_int(*src_colnos, attnum)" to
271 : * detect already-merged columns in the loop below, that would be O(N^2)
272 : * for a wide input table. Instead build a bitmapset of just the merged
273 : * USING columns, which we won't add to within the main loop.
274 : */
275 110694 : prevcols = NULL;
276 113024 : foreach(lc, *src_colnos)
277 : {
278 2330 : prevcols = bms_add_member(prevcols, lfirst_int(lc));
279 : }
280 :
281 110694 : attnum = 0;
282 2240081 : foreach(lc, src_colnames)
283 : {
284 2129387 : char *colname = strVal(lfirst(lc));
285 :
286 2129387 : attnum++;
287 : /* Non-dropped and not already merged? */
288 2129387 : if (colname[0] != '\0' && !bms_is_member(attnum, prevcols))
289 : {
290 : /* Yes, so emit it as next output column */
291 2126738 : *src_colnos = lappend_int(*src_colnos, attnum);
292 2126738 : *res_colnames = lappend(*res_colnames, lfirst(lc));
293 2126738 : *res_colvars = lappend(*res_colvars,
294 2126738 : buildVarFromNSColumn(pstate,
295 2126738 : src_nscolumns + attnum - 1));
296 : /* Copy the input relation's nscolumn data for this column */
297 2126738 : res_nscolumns[colcount] = src_nscolumns[attnum - 1];
298 2126738 : colcount++;
299 : }
300 : }
301 110694 : return colcount;
302 : }
303 :
304 : /* transformJoinUsingClause()
305 : * Build a complete ON clause from a partially-transformed USING list.
306 : * We are given lists of nodes representing left and right match columns.
307 : * Result is a transformed qualification expression.
308 : */
309 : static Node *
310 1025 : transformJoinUsingClause(ParseState *pstate,
311 : List *leftVars, List *rightVars)
312 : {
313 : Node *result;
314 1025 : List *andargs = NIL;
315 : ListCell *lvars,
316 : *rvars;
317 :
318 : /*
319 : * We cheat a little bit here by building an untransformed operator tree
320 : * whose leaves are the already-transformed Vars. This requires collusion
321 : * from transformExpr(), which normally could be expected to complain
322 : * about already-transformed subnodes. However, this does mean that we
323 : * have to mark the columns as requiring SELECT privilege for ourselves;
324 : * transformExpr() won't do it.
325 : */
326 2190 : forboth(lvars, leftVars, rvars, rightVars)
327 : {
328 1165 : Var *lvar = (Var *) lfirst(lvars);
329 1165 : Var *rvar = (Var *) lfirst(rvars);
330 : A_Expr *e;
331 :
332 : /* Require read access to the join variables */
333 1165 : markVarForSelectPriv(pstate, lvar);
334 1165 : markVarForSelectPriv(pstate, rvar);
335 :
336 : /* Now create the lvar = rvar join condition */
337 1165 : e = makeSimpleA_Expr(AEXPR_OP, "=",
338 1165 : (Node *) copyObject(lvar), (Node *) copyObject(rvar),
339 : -1);
340 :
341 : /* Prepare to combine into an AND clause, if multiple join columns */
342 1165 : andargs = lappend(andargs, e);
343 : }
344 :
345 : /* Only need an AND if there's more than one join column */
346 1025 : if (list_length(andargs) == 1)
347 899 : result = (Node *) linitial(andargs);
348 : else
349 126 : result = (Node *) makeBoolExpr(AND_EXPR, andargs, -1);
350 :
351 : /*
352 : * Since the references are already Vars, and are certainly from the input
353 : * relations, we don't have to go through the same pushups that
354 : * transformJoinOnClause() does. Just invoke transformExpr() to fix up
355 : * the operators, and we're done.
356 : */
357 1025 : result = transformExpr(pstate, result, EXPR_KIND_JOIN_USING);
358 :
359 1025 : result = coerce_to_boolean(pstate, result, "JOIN/USING");
360 :
361 1025 : return result;
362 : }
363 :
364 : /* transformJoinOnClause()
365 : * Transform the qual conditions for JOIN/ON.
366 : * Result is a transformed qualification expression.
367 : */
368 : static Node *
369 53988 : transformJoinOnClause(ParseState *pstate, JoinExpr *j, List *namespace)
370 : {
371 : Node *result;
372 : List *save_namespace;
373 :
374 : /*
375 : * The namespace that the join expression should see is just the two
376 : * subtrees of the JOIN plus any outer references from upper pstate
377 : * levels. Temporarily set this pstate's namespace accordingly. (We need
378 : * not check for refname conflicts, because transformFromClauseItem()
379 : * already did.) All namespace items are marked visible regardless of
380 : * LATERAL state.
381 : */
382 53988 : setNamespaceLateralState(namespace, false, true);
383 :
384 53988 : save_namespace = pstate->p_namespace;
385 53988 : pstate->p_namespace = namespace;
386 :
387 53988 : result = transformWhereClause(pstate, j->quals,
388 : EXPR_KIND_JOIN_ON, "JOIN/ON");
389 :
390 53976 : pstate->p_namespace = save_namespace;
391 :
392 53976 : return result;
393 : }
394 :
395 : /*
396 : * transformTableEntry --- transform a RangeVar (simple relation reference)
397 : */
398 : static ParseNamespaceItem *
399 256751 : transformTableEntry(ParseState *pstate, RangeVar *r)
400 : {
401 : /* addRangeTableEntry does all the work */
402 256751 : return addRangeTableEntry(pstate, r, r->alias, r->inh, true);
403 : }
404 :
405 : /*
406 : * transformRangeSubselect --- transform a sub-SELECT appearing in FROM
407 : */
408 : static ParseNamespaceItem *
409 13393 : transformRangeSubselect(ParseState *pstate, RangeSubselect *r)
410 : {
411 : Query *query;
412 :
413 : /*
414 : * Set p_expr_kind to show this parse level is recursing to a subselect.
415 : * We can't be nested within any expression, so don't need save-restore
416 : * logic here.
417 : */
418 : Assert(pstate->p_expr_kind == EXPR_KIND_NONE);
419 13393 : pstate->p_expr_kind = EXPR_KIND_FROM_SUBSELECT;
420 :
421 : /*
422 : * If the subselect is LATERAL, make lateral_only names of this level
423 : * visible to it. (LATERAL can't nest within a single pstate level, so we
424 : * don't need save/restore logic here.)
425 : */
426 : Assert(!pstate->p_lateral_active);
427 13393 : pstate->p_lateral_active = r->lateral;
428 :
429 : /*
430 : * Analyze and transform the subquery. Note that if the subquery doesn't
431 : * have an alias, it can't be explicitly selected for locking, but locking
432 : * might still be required (if there is an all-tables locking clause).
433 : */
434 13393 : query = parse_sub_analyze(r->subquery, pstate, NULL,
435 13393 : isLockedRefname(pstate,
436 13393 : r->alias == NULL ? NULL :
437 13228 : r->alias->aliasname),
438 : true);
439 :
440 : /* Restore state */
441 13321 : pstate->p_lateral_active = false;
442 13321 : pstate->p_expr_kind = EXPR_KIND_NONE;
443 :
444 : /*
445 : * Check that we got a SELECT. Anything else should be impossible given
446 : * restrictions of the grammar, but check anyway.
447 : */
448 13321 : if (!IsA(query, Query) ||
449 13321 : query->commandType != CMD_SELECT)
450 0 : elog(ERROR, "unexpected non-SELECT command in subquery in FROM");
451 :
452 : /*
453 : * OK, build an RTE and nsitem for the subquery.
454 : */
455 26638 : return addRangeTableEntryForSubquery(pstate,
456 : query,
457 : r->alias,
458 13321 : r->lateral,
459 : true);
460 : }
461 :
462 :
463 : /*
464 : * transformRangeFunction --- transform a function call appearing in FROM
465 : */
466 : static ParseNamespaceItem *
467 29245 : transformRangeFunction(ParseState *pstate, RangeFunction *r)
468 : {
469 29245 : List *funcexprs = NIL;
470 29245 : List *funcnames = NIL;
471 29245 : List *coldeflists = NIL;
472 : bool is_lateral;
473 : ListCell *lc;
474 :
475 : /*
476 : * We make lateral_only names of this level visible, whether or not the
477 : * RangeFunction is explicitly marked LATERAL. This is needed for SQL
478 : * spec compliance in the case of UNNEST(), and seems useful on
479 : * convenience grounds for all functions in FROM.
480 : *
481 : * (LATERAL can't nest within a single pstate level, so we don't need
482 : * save/restore logic here.)
483 : */
484 : Assert(!pstate->p_lateral_active);
485 29245 : pstate->p_lateral_active = true;
486 :
487 : /*
488 : * Transform the raw expressions.
489 : *
490 : * While transforming, also save function names for possible use as alias
491 : * and column names. We use the same transformation rules as for a SELECT
492 : * output expression. For a FuncCall node, the result will be the
493 : * function name, but it is possible for the grammar to hand back other
494 : * node types.
495 : *
496 : * We have to get this info now, because FigureColname only works on raw
497 : * parsetrees. Actually deciding what to do with the names is left up to
498 : * addRangeTableEntryForFunction.
499 : *
500 : * Likewise, collect column definition lists if there were any. But
501 : * complain if we find one here and the RangeFunction has one too.
502 : */
503 58502 : foreach(lc, r->functions)
504 : {
505 29369 : List *pair = (List *) lfirst(lc);
506 : Node *fexpr;
507 : List *coldeflist;
508 : Node *newfexpr;
509 : Node *last_srf;
510 :
511 : /* Disassemble the function-call/column-def-list pairs */
512 : Assert(list_length(pair) == 2);
513 29369 : fexpr = (Node *) linitial(pair);
514 29369 : coldeflist = (List *) lsecond(pair);
515 :
516 : /*
517 : * If we find a function call unnest() with more than one argument and
518 : * no special decoration, transform it into separate unnest() calls on
519 : * each argument. This is a kluge, for sure, but it's less nasty than
520 : * other ways of implementing the SQL-standard UNNEST() syntax.
521 : *
522 : * If there is any decoration (including a coldeflist), we don't
523 : * transform, which probably means a no-such-function error later. We
524 : * could alternatively throw an error right now, but that doesn't seem
525 : * tremendously helpful. If someone is using any such decoration,
526 : * then they're not using the SQL-standard syntax, and they're more
527 : * likely expecting an un-tweaked function call.
528 : *
529 : * Note: the transformation changes a non-schema-qualified unnest()
530 : * function name into schema-qualified pg_catalog.unnest(). This
531 : * choice is also a bit debatable, but it seems reasonable to force
532 : * use of built-in unnest() when we make this transformation.
533 : */
534 29369 : if (IsA(fexpr, FuncCall))
535 : {
536 29273 : FuncCall *fc = (FuncCall *) fexpr;
537 :
538 29273 : if (list_length(fc->funcname) == 1 &&
539 20789 : strcmp(strVal(linitial(fc->funcname)), "unnest") == 0 &&
540 2039 : list_length(fc->args) > 1 &&
541 43 : fc->agg_order == NIL &&
542 43 : fc->agg_filter == NULL &&
543 43 : fc->over == NULL &&
544 43 : !fc->agg_star &&
545 43 : !fc->agg_distinct &&
546 43 : !fc->func_variadic &&
547 : coldeflist == NIL)
548 43 : {
549 : ListCell *lc2;
550 :
551 155 : foreach(lc2, fc->args)
552 : {
553 112 : Node *arg = (Node *) lfirst(lc2);
554 : FuncCall *newfc;
555 :
556 112 : last_srf = pstate->p_last_srf;
557 :
558 112 : newfc = makeFuncCall(SystemFuncName("unnest"),
559 112 : list_make1(arg),
560 : COERCE_EXPLICIT_CALL,
561 : fc->location);
562 :
563 112 : newfexpr = transformExpr(pstate, (Node *) newfc,
564 : EXPR_KIND_FROM_FUNCTION);
565 :
566 : /* nodeFunctionscan.c requires SRFs to be at top level */
567 112 : if (pstate->p_last_srf != last_srf &&
568 112 : pstate->p_last_srf != newfexpr)
569 0 : ereport(ERROR,
570 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
571 : errmsg("set-returning functions must appear at top level of FROM"),
572 : parser_errposition(pstate,
573 : exprLocation(pstate->p_last_srf))));
574 :
575 112 : funcexprs = lappend(funcexprs, newfexpr);
576 :
577 112 : funcnames = lappend(funcnames,
578 112 : FigureColname((Node *) newfc));
579 :
580 : /* coldeflist is empty, so no error is possible */
581 :
582 112 : coldeflists = lappend(coldeflists, coldeflist);
583 : }
584 43 : continue; /* done with this function item */
585 : }
586 : }
587 :
588 : /* normal case ... */
589 29326 : last_srf = pstate->p_last_srf;
590 :
591 29326 : newfexpr = transformExpr(pstate, fexpr,
592 : EXPR_KIND_FROM_FUNCTION);
593 :
594 : /* nodeFunctionscan.c requires SRFs to be at top level */
595 29218 : if (pstate->p_last_srf != last_srf &&
596 23507 : pstate->p_last_srf != newfexpr)
597 4 : ereport(ERROR,
598 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
599 : errmsg("set-returning functions must appear at top level of FROM"),
600 : parser_errposition(pstate,
601 : exprLocation(pstate->p_last_srf))));
602 :
603 29214 : funcexprs = lappend(funcexprs, newfexpr);
604 :
605 29214 : funcnames = lappend(funcnames,
606 29214 : FigureColname(fexpr));
607 :
608 29214 : if (coldeflist && r->coldeflist)
609 0 : ereport(ERROR,
610 : (errcode(ERRCODE_SYNTAX_ERROR),
611 : errmsg("multiple column definition lists are not allowed for the same function"),
612 : parser_errposition(pstate,
613 : exprLocation((Node *) r->coldeflist))));
614 :
615 29214 : coldeflists = lappend(coldeflists, coldeflist);
616 : }
617 :
618 29133 : pstate->p_lateral_active = false;
619 :
620 : /*
621 : * We must assign collations now so that the RTE exposes correct collation
622 : * info for Vars created from it.
623 : */
624 29133 : assign_list_collations(pstate, funcexprs);
625 :
626 : /*
627 : * Install the top-level coldeflist if there was one (we already checked
628 : * that there was no conflicting per-function coldeflist).
629 : *
630 : * We only allow this when there's a single function (even after UNNEST
631 : * expansion) and no WITH ORDINALITY. The reason for the latter
632 : * restriction is that it's not real clear whether the ordinality column
633 : * should be in the coldeflist, and users are too likely to make mistakes
634 : * in one direction or the other. Putting the coldeflist inside ROWS
635 : * FROM() is much clearer in this case.
636 : */
637 29133 : if (r->coldeflist)
638 : {
639 461 : if (list_length(funcexprs) != 1)
640 : {
641 0 : if (r->is_rowsfrom)
642 0 : ereport(ERROR,
643 : (errcode(ERRCODE_SYNTAX_ERROR),
644 : errmsg("ROWS FROM() with multiple functions cannot have a column definition list"),
645 : errhint("Put a separate column definition list for each function inside ROWS FROM()."),
646 : parser_errposition(pstate,
647 : exprLocation((Node *) r->coldeflist))));
648 : else
649 0 : ereport(ERROR,
650 : (errcode(ERRCODE_SYNTAX_ERROR),
651 : errmsg("UNNEST() with multiple arguments cannot have a column definition list"),
652 : errhint("Use separate UNNEST() calls inside ROWS FROM(), and attach a column definition list to each one."),
653 : parser_errposition(pstate,
654 : exprLocation((Node *) r->coldeflist))));
655 : }
656 461 : if (r->ordinality)
657 0 : ereport(ERROR,
658 : (errcode(ERRCODE_SYNTAX_ERROR),
659 : errmsg("WITH ORDINALITY cannot be used with a column definition list"),
660 : errhint("Put the column definition list inside ROWS FROM()."),
661 : parser_errposition(pstate,
662 : exprLocation((Node *) r->coldeflist))));
663 :
664 461 : coldeflists = list_make1(r->coldeflist);
665 : }
666 :
667 : /*
668 : * Mark the RTE as LATERAL if the user said LATERAL explicitly, or if
669 : * there are any lateral cross-references in it.
670 : */
671 29133 : is_lateral = r->lateral || contain_vars_of_level((Node *) funcexprs, 0);
672 :
673 : /*
674 : * OK, build an RTE and nsitem for the function.
675 : */
676 29133 : return addRangeTableEntryForFunction(pstate,
677 : funcnames, funcexprs, coldeflists,
678 : r, is_lateral, true);
679 : }
680 :
681 : /*
682 : * transformRangeTableFunc -
683 : * Transform a raw RangeTableFunc into TableFunc.
684 : *
685 : * Transform the namespace clauses, the document-generating expression, the
686 : * row-generating expression, the column-generating expressions, and the
687 : * default value expressions.
688 : */
689 : static ParseNamespaceItem *
690 146 : transformRangeTableFunc(ParseState *pstate, RangeTableFunc *rtf)
691 : {
692 146 : TableFunc *tf = makeNode(TableFunc);
693 : const char *constructName;
694 : Oid docType;
695 : bool is_lateral;
696 : ListCell *col;
697 : char **names;
698 : int colno;
699 :
700 : /*
701 : * Currently we only support XMLTABLE here. See transformJsonTable() for
702 : * JSON_TABLE support.
703 : */
704 146 : tf->functype = TFT_XMLTABLE;
705 146 : constructName = "XMLTABLE";
706 146 : docType = XMLOID;
707 :
708 : /*
709 : * We make lateral_only names of this level visible, whether or not the
710 : * RangeTableFunc is explicitly marked LATERAL. This is needed for SQL
711 : * spec compliance and seems useful on convenience grounds for all
712 : * functions in FROM.
713 : *
714 : * (LATERAL can't nest within a single pstate level, so we don't need
715 : * save/restore logic here.)
716 : */
717 : Assert(!pstate->p_lateral_active);
718 146 : pstate->p_lateral_active = true;
719 :
720 : /* Transform and apply typecast to the row-generating expression ... */
721 : Assert(rtf->rowexpr != NULL);
722 146 : tf->rowexpr = coerce_to_specific_type(pstate,
723 : transformExpr(pstate, rtf->rowexpr, EXPR_KIND_FROM_FUNCTION),
724 : TEXTOID,
725 : constructName);
726 146 : assign_expr_collations(pstate, tf->rowexpr);
727 :
728 : /* ... and to the document itself */
729 : Assert(rtf->docexpr != NULL);
730 146 : tf->docexpr = coerce_to_specific_type(pstate,
731 : transformExpr(pstate, rtf->docexpr, EXPR_KIND_FROM_FUNCTION),
732 : docType,
733 : constructName);
734 146 : assign_expr_collations(pstate, tf->docexpr);
735 :
736 : /* undef ordinality column number */
737 146 : tf->ordinalitycol = -1;
738 :
739 : /* Process column specs */
740 146 : names = palloc_array(char *, list_length(rtf->columns));
741 :
742 146 : colno = 0;
743 643 : foreach(col, rtf->columns)
744 : {
745 497 : RangeTableFuncCol *rawc = (RangeTableFuncCol *) lfirst(col);
746 : Oid typid;
747 : int32 typmod;
748 : Node *colexpr;
749 : Node *coldefexpr;
750 : int j;
751 :
752 497 : tf->colnames = lappend(tf->colnames,
753 497 : makeString(pstrdup(rawc->colname)));
754 :
755 : /*
756 : * Determine the type and typmod for the new column. FOR ORDINALITY
757 : * columns are INTEGER per spec; the others are user-specified.
758 : */
759 497 : if (rawc->for_ordinality)
760 : {
761 41 : if (tf->ordinalitycol != -1)
762 0 : ereport(ERROR,
763 : (errcode(ERRCODE_SYNTAX_ERROR),
764 : errmsg("only one FOR ORDINALITY column is allowed"),
765 : parser_errposition(pstate, rawc->location)));
766 :
767 41 : typid = INT4OID;
768 41 : typmod = -1;
769 41 : tf->ordinalitycol = colno;
770 : }
771 : else
772 : {
773 456 : if (rawc->typeName->setof)
774 0 : ereport(ERROR,
775 : (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
776 : errmsg("column \"%s\" cannot be declared SETOF",
777 : rawc->colname),
778 : parser_errposition(pstate, rawc->location)));
779 :
780 456 : typenameTypeIdAndMod(pstate, rawc->typeName,
781 : &typid, &typmod);
782 : }
783 :
784 497 : tf->coltypes = lappend_oid(tf->coltypes, typid);
785 497 : tf->coltypmods = lappend_int(tf->coltypmods, typmod);
786 497 : tf->colcollations = lappend_oid(tf->colcollations,
787 : get_typcollation(typid));
788 :
789 : /* Transform the PATH and DEFAULT expressions */
790 497 : if (rawc->colexpr)
791 : {
792 324 : colexpr = coerce_to_specific_type(pstate,
793 : transformExpr(pstate, rawc->colexpr,
794 : EXPR_KIND_FROM_FUNCTION),
795 : TEXTOID,
796 : constructName);
797 324 : assign_expr_collations(pstate, colexpr);
798 : }
799 : else
800 173 : colexpr = NULL;
801 :
802 497 : if (rawc->coldefexpr)
803 : {
804 37 : coldefexpr = coerce_to_specific_type_typmod(pstate,
805 : transformExpr(pstate, rawc->coldefexpr,
806 : EXPR_KIND_FROM_FUNCTION),
807 : typid, typmod,
808 : constructName);
809 37 : assign_expr_collations(pstate, coldefexpr);
810 : }
811 : else
812 460 : coldefexpr = NULL;
813 :
814 497 : tf->colexprs = lappend(tf->colexprs, colexpr);
815 497 : tf->coldefexprs = lappend(tf->coldefexprs, coldefexpr);
816 :
817 497 : if (rawc->is_not_null)
818 37 : tf->notnulls = bms_add_member(tf->notnulls, colno);
819 :
820 : /* make sure column names are unique */
821 1677 : for (j = 0; j < colno; j++)
822 1180 : if (strcmp(names[j], rawc->colname) == 0)
823 0 : ereport(ERROR,
824 : (errcode(ERRCODE_SYNTAX_ERROR),
825 : errmsg("column name \"%s\" is not unique",
826 : rawc->colname),
827 : parser_errposition(pstate, rawc->location)));
828 497 : names[colno] = rawc->colname;
829 :
830 497 : colno++;
831 : }
832 146 : pfree(names);
833 :
834 : /* Namespaces, if any, also need to be transformed */
835 146 : if (rtf->namespaces != NIL)
836 : {
837 : ListCell *ns;
838 : ListCell *lc2;
839 13 : List *ns_uris = NIL;
840 13 : List *ns_names = NIL;
841 13 : bool default_ns_seen = false;
842 :
843 26 : foreach(ns, rtf->namespaces)
844 : {
845 13 : ResTarget *r = (ResTarget *) lfirst(ns);
846 : Node *ns_uri;
847 :
848 : Assert(IsA(r, ResTarget));
849 13 : ns_uri = transformExpr(pstate, r->val, EXPR_KIND_FROM_FUNCTION);
850 13 : ns_uri = coerce_to_specific_type(pstate, ns_uri,
851 : TEXTOID, constructName);
852 13 : assign_expr_collations(pstate, ns_uri);
853 13 : ns_uris = lappend(ns_uris, ns_uri);
854 :
855 : /* Verify consistency of name list: no dupes, only one DEFAULT */
856 13 : if (r->name != NULL)
857 : {
858 9 : foreach(lc2, ns_names)
859 : {
860 0 : String *ns_node = lfirst_node(String, lc2);
861 :
862 0 : if (ns_node == NULL)
863 0 : continue;
864 0 : if (strcmp(strVal(ns_node), r->name) == 0)
865 0 : ereport(ERROR,
866 : (errcode(ERRCODE_SYNTAX_ERROR),
867 : errmsg("namespace name \"%s\" is not unique",
868 : r->name),
869 : parser_errposition(pstate, r->location)));
870 : }
871 : }
872 : else
873 : {
874 4 : if (default_ns_seen)
875 0 : ereport(ERROR,
876 : (errcode(ERRCODE_SYNTAX_ERROR),
877 : errmsg("only one default namespace is allowed"),
878 : parser_errposition(pstate, r->location)));
879 4 : default_ns_seen = true;
880 : }
881 :
882 : /* We represent DEFAULT by a null pointer */
883 13 : ns_names = lappend(ns_names,
884 13 : r->name ? makeString(r->name) : NULL);
885 : }
886 :
887 13 : tf->ns_uris = ns_uris;
888 13 : tf->ns_names = ns_names;
889 : }
890 :
891 146 : tf->location = rtf->location;
892 :
893 146 : pstate->p_lateral_active = false;
894 :
895 : /*
896 : * Mark the RTE as LATERAL if the user said LATERAL explicitly, or if
897 : * there are any lateral cross-references in it.
898 : */
899 146 : is_lateral = rtf->lateral || contain_vars_of_level((Node *) tf, 0);
900 :
901 146 : return addRangeTableEntryForTableFunc(pstate,
902 : tf, rtf->alias, is_lateral, true);
903 : }
904 :
905 : /*
906 : * Similar to parserOpenTable() but for property graphs.
907 : */
908 : static Relation
909 536 : parserOpenPropGraph(ParseState *pstate, const RangeVar *relation, LOCKMODE lockmode)
910 : {
911 : Relation rel;
912 : ParseCallbackState pcbstate;
913 :
914 536 : setup_parser_errposition_callback(&pcbstate, pstate, relation->location);
915 :
916 536 : rel = relation_openrv(relation, lockmode);
917 :
918 : /*
919 : * In parserOpenTable(), the relkind check is done inside table_openrv*.
920 : * We do it here since we don't have anything like propgraph_open.
921 : */
922 532 : if (rel->rd_rel->relkind != RELKIND_PROPGRAPH)
923 4 : ereport(ERROR,
924 : errcode(ERRCODE_WRONG_OBJECT_TYPE),
925 : errmsg("\"%s\" is not a property graph",
926 : RelationGetRelationName(rel)));
927 :
928 528 : cancel_parser_errposition_callback(&pcbstate);
929 528 : return rel;
930 : }
931 :
932 : /*
933 : * transformRangeGraphTable -- transform a GRAPH_TABLE clause
934 : */
935 : static ParseNamespaceItem *
936 536 : transformRangeGraphTable(ParseState *pstate, RangeGraphTable *rgt)
937 : {
938 : Relation rel;
939 : Oid graphid;
940 536 : GraphTableParseState *gpstate = palloc0_object(GraphTableParseState);
941 : Node *gp;
942 536 : List *columns = NIL;
943 536 : List *colnames = NIL;
944 : ListCell *lc;
945 536 : int resno = 0;
946 : bool saved_hasSublinks;
947 :
948 536 : rel = parserOpenPropGraph(pstate, rgt->graph_name, AccessShareLock);
949 :
950 528 : graphid = RelationGetRelid(rel);
951 :
952 528 : gpstate->graphid = graphid;
953 :
954 : /*
955 : * The syntax does not allow nested GRAPH_TABLE and this function
956 : * prohibits subquery within GRAPH_TABLE. There should be only one
957 : * GRAPH_TABLE being transformed at a time.
958 : */
959 : Assert(!pstate->p_graph_table_pstate);
960 528 : pstate->p_graph_table_pstate = gpstate;
961 :
962 : Assert(!pstate->p_lateral_active);
963 528 : pstate->p_lateral_active = true;
964 :
965 528 : saved_hasSublinks = pstate->p_hasSubLinks;
966 528 : pstate->p_hasSubLinks = false;
967 :
968 528 : gp = transformGraphPattern(pstate, rgt->graph_pattern);
969 :
970 : /*
971 : * Construct a targetlist representing the COLUMNS specified in the
972 : * GRAPH_TABLE. This uses previously constructed list of element pattern
973 : * variables in the GraphTableParseState.
974 : */
975 1808 : foreach(lc, rgt->columns)
976 : {
977 1312 : ResTarget *rt = lfirst_node(ResTarget, lc);
978 : Node *colexpr;
979 : TargetEntry *te;
980 : char *colname;
981 :
982 1312 : colexpr = transformExpr(pstate, rt->val, EXPR_KIND_SELECT_TARGET);
983 :
984 1300 : if (rt->name)
985 589 : colname = rt->name;
986 : else
987 : {
988 711 : if (IsA(colexpr, GraphPropertyRef))
989 711 : colname = get_propgraph_property_name(castNode(GraphPropertyRef, colexpr)->propid);
990 : else
991 : {
992 0 : ereport(ERROR,
993 : errcode(ERRCODE_SYNTAX_ERROR),
994 : errmsg("complex graph table column must specify an explicit column name"),
995 : parser_errposition(pstate, rt->location));
996 : colname = NULL;
997 : }
998 : }
999 :
1000 1300 : colnames = lappend(colnames, makeString(colname));
1001 :
1002 1300 : te = makeTargetEntry((Expr *) colexpr, ++resno, colname, false);
1003 1300 : columns = lappend(columns, te);
1004 : }
1005 :
1006 496 : table_close(rel, NoLock);
1007 :
1008 496 : pstate->p_graph_table_pstate = NULL;
1009 496 : pstate->p_lateral_active = false;
1010 :
1011 : /*
1012 : * If we support subqueries within GRAPH_TABLE, those need to be
1013 : * propagated to the queries resulting from rewriting graph table RTE. We
1014 : * don't do that right now, hence prohibit it for now.
1015 : */
1016 496 : if (pstate->p_hasSubLinks)
1017 8 : ereport(ERROR,
1018 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1019 : errmsg("subqueries within GRAPH_TABLE reference are not supported")));
1020 488 : pstate->p_hasSubLinks = saved_hasSublinks;
1021 :
1022 488 : return addRangeTableEntryForGraphTable(pstate, graphid, castNode(GraphPattern, gp), columns, colnames, rgt->alias, false, true);
1023 : }
1024 :
1025 : /*
1026 : * transformRangeTableSample --- transform a TABLESAMPLE clause
1027 : *
1028 : * Caller has already transformed rts->relation, we just have to validate
1029 : * the remaining fields and create a TableSampleClause node.
1030 : */
1031 : static TableSampleClause *
1032 158 : transformRangeTableSample(ParseState *pstate, RangeTableSample *rts)
1033 : {
1034 : TableSampleClause *tablesample;
1035 : Oid handlerOid;
1036 : Oid funcargtypes[1];
1037 : TsmRoutine *tsm;
1038 : List *fargs;
1039 : ListCell *larg,
1040 : *ltyp;
1041 :
1042 : /*
1043 : * To validate the sample method name, look up the handler function, which
1044 : * has the same name, one dummy INTERNAL argument, and a result type of
1045 : * tsm_handler. (Note: tablesample method names are not schema-qualified
1046 : * in the SQL standard; but since they are just functions to us, we allow
1047 : * schema qualification to resolve any potential ambiguity.)
1048 : */
1049 158 : funcargtypes[0] = INTERNALOID;
1050 :
1051 158 : handlerOid = LookupFuncName(rts->method, 1, funcargtypes, true);
1052 :
1053 : /* we want error to complain about no-such-method, not no-such-function */
1054 158 : if (!OidIsValid(handlerOid))
1055 4 : ereport(ERROR,
1056 : (errcode(ERRCODE_UNDEFINED_OBJECT),
1057 : errmsg("tablesample method %s does not exist",
1058 : NameListToString(rts->method)),
1059 : parser_errposition(pstate, rts->location)));
1060 :
1061 : /* check that handler has correct return type */
1062 154 : if (get_func_rettype(handlerOid) != TSM_HANDLEROID)
1063 0 : ereport(ERROR,
1064 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1065 : errmsg("function %s must return type %s",
1066 : NameListToString(rts->method), "tsm_handler"),
1067 : parser_errposition(pstate, rts->location)));
1068 :
1069 : /* OK, run the handler to get TsmRoutine, for argument type info */
1070 154 : tsm = GetTsmRoutine(handlerOid);
1071 :
1072 154 : tablesample = makeNode(TableSampleClause);
1073 154 : tablesample->tsmhandler = handlerOid;
1074 :
1075 : /* check user provided the expected number of arguments */
1076 154 : if (list_length(rts->args) != list_length(tsm->parameterTypes))
1077 0 : ereport(ERROR,
1078 : (errcode(ERRCODE_INVALID_TABLESAMPLE_ARGUMENT),
1079 : errmsg_plural("tablesample method %s requires %d argument, not %d",
1080 : "tablesample method %s requires %d arguments, not %d",
1081 : list_length(tsm->parameterTypes),
1082 : NameListToString(rts->method),
1083 : list_length(tsm->parameterTypes),
1084 : list_length(rts->args)),
1085 : parser_errposition(pstate, rts->location)));
1086 :
1087 : /*
1088 : * Transform the arguments, typecasting them as needed. Note we must also
1089 : * assign collations now, because assign_query_collations() doesn't
1090 : * examine any substructure of RTEs.
1091 : */
1092 154 : fargs = NIL;
1093 308 : forboth(larg, rts->args, ltyp, tsm->parameterTypes)
1094 : {
1095 154 : Node *arg = (Node *) lfirst(larg);
1096 154 : Oid argtype = lfirst_oid(ltyp);
1097 :
1098 154 : arg = transformExpr(pstate, arg, EXPR_KIND_FROM_FUNCTION);
1099 154 : arg = coerce_to_specific_type(pstate, arg, argtype, "TABLESAMPLE");
1100 154 : assign_expr_collations(pstate, arg);
1101 154 : fargs = lappend(fargs, arg);
1102 : }
1103 154 : tablesample->args = fargs;
1104 :
1105 : /* Process REPEATABLE (seed) */
1106 154 : if (rts->repeatable != NULL)
1107 : {
1108 : Node *arg;
1109 :
1110 67 : if (!tsm->repeatable_across_queries)
1111 2 : ereport(ERROR,
1112 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1113 : errmsg("tablesample method %s does not support REPEATABLE",
1114 : NameListToString(rts->method)),
1115 : parser_errposition(pstate, rts->location)));
1116 :
1117 65 : arg = transformExpr(pstate, rts->repeatable, EXPR_KIND_FROM_FUNCTION);
1118 65 : arg = coerce_to_specific_type(pstate, arg, FLOAT8OID, "REPEATABLE");
1119 65 : assign_expr_collations(pstate, arg);
1120 65 : tablesample->repeatable = (Expr *) arg;
1121 : }
1122 : else
1123 87 : tablesample->repeatable = NULL;
1124 :
1125 152 : return tablesample;
1126 : }
1127 :
1128 : /*
1129 : * getNSItemForSpecialRelationTypes
1130 : *
1131 : * If given RangeVar refers to a CTE or an EphemeralNamedRelation,
1132 : * build and return an appropriate ParseNamespaceItem, otherwise return NULL
1133 : */
1134 : static ParseNamespaceItem *
1135 261435 : getNSItemForSpecialRelationTypes(ParseState *pstate, RangeVar *rv)
1136 : {
1137 : ParseNamespaceItem *nsitem;
1138 : CommonTableExpr *cte;
1139 : Index levelsup;
1140 :
1141 : /*
1142 : * if it is a qualified name, it can't be a CTE or tuplestore reference
1143 : */
1144 261435 : if (rv->schemaname)
1145 130444 : return NULL;
1146 :
1147 130991 : cte = scanNameSpaceForCTE(pstate, rv->relname, &levelsup);
1148 130991 : if (cte)
1149 4362 : nsitem = addRangeTableEntryForCTE(pstate, cte, levelsup, rv, true);
1150 126629 : else if (scanNameSpaceForENR(pstate, rv->relname))
1151 322 : nsitem = addRangeTableEntryForENR(pstate, rv, true);
1152 : else
1153 126307 : nsitem = NULL;
1154 :
1155 130983 : return nsitem;
1156 : }
1157 :
1158 : /*
1159 : * transformFromClauseItem -
1160 : * Transform a FROM-clause item, adding any required entries to the
1161 : * range table list being built in the ParseState, and return the
1162 : * transformed item ready to include in the joinlist. Also build a
1163 : * ParseNamespaceItem list describing the names exposed by this item.
1164 : * This routine can recurse to handle SQL92 JOIN expressions.
1165 : *
1166 : * The function return value is the node to add to the jointree (a
1167 : * RangeTblRef or JoinExpr). Additional output parameters are:
1168 : *
1169 : * *top_nsitem: receives the ParseNamespaceItem directly corresponding to the
1170 : * jointree item. (This is only used during internal recursion, not by
1171 : * outside callers.)
1172 : *
1173 : * *namespace: receives a List of ParseNamespaceItems for the RTEs exposed
1174 : * as table/column names by this item. (The lateral_only flags in these items
1175 : * are indeterminate and should be explicitly set by the caller before use.)
1176 : */
1177 : static Node *
1178 360652 : transformFromClauseItem(ParseState *pstate, Node *n,
1179 : ParseNamespaceItem **top_nsitem,
1180 : List **namespace)
1181 : {
1182 : /* Guard against stack overflow due to overly deep subtree */
1183 360652 : check_stack_depth();
1184 :
1185 360652 : if (IsA(n, RangeVar))
1186 : {
1187 : /* Plain relation reference, or perhaps a CTE reference */
1188 261435 : RangeVar *rv = (RangeVar *) n;
1189 : RangeTblRef *rtr;
1190 : ParseNamespaceItem *nsitem;
1191 :
1192 : /* Check if it's a CTE or tuplestore reference */
1193 261435 : nsitem = getNSItemForSpecialRelationTypes(pstate, rv);
1194 :
1195 : /* if not found above, must be a table reference */
1196 261427 : if (!nsitem)
1197 256751 : nsitem = transformTableEntry(pstate, rv);
1198 :
1199 261317 : *top_nsitem = nsitem;
1200 261317 : *namespace = list_make1(nsitem);
1201 261317 : rtr = makeNode(RangeTblRef);
1202 261317 : rtr->rtindex = nsitem->p_rtindex;
1203 261317 : return (Node *) rtr;
1204 : }
1205 99217 : else if (IsA(n, RangeSubselect))
1206 : {
1207 : /* sub-SELECT is like a plain relation */
1208 : RangeTblRef *rtr;
1209 : ParseNamespaceItem *nsitem;
1210 :
1211 13393 : nsitem = transformRangeSubselect(pstate, (RangeSubselect *) n);
1212 13317 : *top_nsitem = nsitem;
1213 13317 : *namespace = list_make1(nsitem);
1214 13317 : rtr = makeNode(RangeTblRef);
1215 13317 : rtr->rtindex = nsitem->p_rtindex;
1216 13317 : return (Node *) rtr;
1217 : }
1218 85824 : else if (IsA(n, RangeFunction))
1219 : {
1220 : /* function is like a plain relation */
1221 : RangeTblRef *rtr;
1222 : ParseNamespaceItem *nsitem;
1223 :
1224 29245 : nsitem = transformRangeFunction(pstate, (RangeFunction *) n);
1225 29099 : *top_nsitem = nsitem;
1226 29099 : *namespace = list_make1(nsitem);
1227 29099 : rtr = makeNode(RangeTblRef);
1228 29099 : rtr->rtindex = nsitem->p_rtindex;
1229 29099 : return (Node *) rtr;
1230 : }
1231 56579 : else if (IsA(n, RangeTableFunc) || IsA(n, JsonTable))
1232 : {
1233 : /* table function is like a plain relation */
1234 : RangeTblRef *rtr;
1235 : ParseNamespaceItem *nsitem;
1236 :
1237 494 : if (IsA(n, JsonTable))
1238 348 : nsitem = transformJsonTable(pstate, (JsonTable *) n);
1239 : else
1240 146 : nsitem = transformRangeTableFunc(pstate, (RangeTableFunc *) n);
1241 :
1242 434 : *top_nsitem = nsitem;
1243 434 : *namespace = list_make1(nsitem);
1244 434 : rtr = makeNode(RangeTblRef);
1245 434 : rtr->rtindex = nsitem->p_rtindex;
1246 434 : return (Node *) rtr;
1247 : }
1248 56085 : else if (IsA(n, RangeGraphTable))
1249 : {
1250 : RangeTblRef *rtr;
1251 : ParseNamespaceItem *nsitem;
1252 :
1253 536 : nsitem = transformRangeGraphTable(pstate, (RangeGraphTable *) n);
1254 488 : *top_nsitem = nsitem;
1255 488 : *namespace = list_make1(nsitem);
1256 488 : rtr = makeNode(RangeTblRef);
1257 488 : rtr->rtindex = nsitem->p_rtindex;
1258 488 : return (Node *) rtr;
1259 : }
1260 55549 : else if (IsA(n, RangeTableSample))
1261 : {
1262 : /* TABLESAMPLE clause (wrapping some other valid FROM node) */
1263 166 : RangeTableSample *rts = (RangeTableSample *) n;
1264 : Node *rel;
1265 : RangeTblEntry *rte;
1266 :
1267 : /* Recursively transform the contained relation */
1268 166 : rel = transformFromClauseItem(pstate, rts->relation,
1269 : top_nsitem, namespace);
1270 166 : rte = (*top_nsitem)->p_rte;
1271 : /* We only support this on plain relations and matviews */
1272 166 : if (rte->rtekind != RTE_RELATION ||
1273 162 : (rte->relkind != RELKIND_RELATION &&
1274 16 : rte->relkind != RELKIND_MATVIEW &&
1275 16 : rte->relkind != RELKIND_PARTITIONED_TABLE))
1276 8 : ereport(ERROR,
1277 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1278 : errmsg("TABLESAMPLE clause can only be applied to tables and materialized views"),
1279 : parser_errposition(pstate, exprLocation(rts->relation))));
1280 :
1281 : /* Transform TABLESAMPLE details and attach to the RTE */
1282 158 : rte->tablesample = transformRangeTableSample(pstate, rts);
1283 152 : return rel;
1284 : }
1285 55383 : else if (IsA(n, JoinExpr))
1286 : {
1287 : /* A newfangled join expression */
1288 55383 : JoinExpr *j = (JoinExpr *) n;
1289 : ParseNamespaceItem *nsitem;
1290 : ParseNamespaceItem *l_nsitem;
1291 : ParseNamespaceItem *r_nsitem;
1292 : List *l_namespace,
1293 : *r_namespace,
1294 : *my_namespace,
1295 : *l_colnames,
1296 : *r_colnames,
1297 : *res_colnames,
1298 : *l_colnos,
1299 : *r_colnos,
1300 : *res_colvars;
1301 : ParseNamespaceColumn *l_nscolumns,
1302 : *r_nscolumns,
1303 : *res_nscolumns;
1304 : int res_colindex;
1305 : bool lateral_ok;
1306 : int sv_namespace_length;
1307 : int k;
1308 :
1309 : /*
1310 : * Recursively process the left subtree, then the right. We must do
1311 : * it in this order for correct visibility of LATERAL references.
1312 : */
1313 55383 : j->larg = transformFromClauseItem(pstate, j->larg,
1314 : &l_nsitem,
1315 : &l_namespace);
1316 :
1317 : /*
1318 : * Make the left-side RTEs available for LATERAL access within the
1319 : * right side, by temporarily adding them to the pstate's namespace
1320 : * list. Per SQL:2008, if the join type is not INNER or LEFT then the
1321 : * left-side names must still be exposed, but it's an error to
1322 : * reference them. (Stupid design, but that's what it says.) Hence,
1323 : * we always push them into the namespace, but mark them as not
1324 : * lateral_ok if the jointype is wrong.
1325 : *
1326 : * Notice that we don't require the merged namespace list to be
1327 : * conflict-free. See the comments for scanNameSpaceForRefname().
1328 : */
1329 55383 : lateral_ok = (j->jointype == JOIN_INNER || j->jointype == JOIN_LEFT);
1330 55383 : setNamespaceLateralState(l_namespace, true, lateral_ok);
1331 :
1332 55383 : sv_namespace_length = list_length(pstate->p_namespace);
1333 55383 : pstate->p_namespace = list_concat(pstate->p_namespace, l_namespace);
1334 :
1335 : /* And now we can process the RHS */
1336 55383 : j->rarg = transformFromClauseItem(pstate, j->rarg,
1337 : &r_nsitem,
1338 : &r_namespace);
1339 :
1340 : /* Remove the left-side RTEs from the namespace list again */
1341 55359 : pstate->p_namespace = list_truncate(pstate->p_namespace,
1342 : sv_namespace_length);
1343 :
1344 : /*
1345 : * Check for conflicting refnames in left and right subtrees. Must do
1346 : * this because higher levels will assume I hand back a self-
1347 : * consistent namespace list.
1348 : */
1349 55359 : checkNameSpaceConflicts(pstate, l_namespace, r_namespace);
1350 :
1351 : /*
1352 : * Generate combined namespace info for possible use below.
1353 : */
1354 55359 : my_namespace = list_concat(l_namespace, r_namespace);
1355 :
1356 : /*
1357 : * We'll work from the nscolumns data and eref alias column names for
1358 : * each of the input nsitems. Note that these include dropped
1359 : * columns, which is helpful because we can keep track of physical
1360 : * input column numbers more easily.
1361 : */
1362 55359 : l_nscolumns = l_nsitem->p_nscolumns;
1363 55359 : l_colnames = l_nsitem->p_names->colnames;
1364 55359 : r_nscolumns = r_nsitem->p_nscolumns;
1365 55359 : r_colnames = r_nsitem->p_names->colnames;
1366 :
1367 : /*
1368 : * Natural join does not explicitly specify columns; must generate
1369 : * columns to join. Need to run through the list of columns from each
1370 : * table or join result and match up the column names. Use the first
1371 : * table, and check every column in the second table for a match.
1372 : * (We'll check that the matches were unique later on.) The result of
1373 : * this step is a list of column names just like an explicitly-written
1374 : * USING list.
1375 : */
1376 55359 : if (j->isNatural)
1377 : {
1378 172 : List *rlist = NIL;
1379 : ListCell *lx,
1380 : *rx;
1381 :
1382 : Assert(j->usingClause == NIL); /* shouldn't have USING() too */
1383 :
1384 760 : foreach(lx, l_colnames)
1385 : {
1386 588 : char *l_colname = strVal(lfirst(lx));
1387 588 : String *m_name = NULL;
1388 :
1389 588 : if (l_colname[0] == '\0')
1390 8 : continue; /* ignore dropped columns */
1391 :
1392 1608 : foreach(rx, r_colnames)
1393 : {
1394 1232 : char *r_colname = strVal(lfirst(rx));
1395 :
1396 1232 : if (strcmp(l_colname, r_colname) == 0)
1397 : {
1398 204 : m_name = makeString(l_colname);
1399 204 : break;
1400 : }
1401 : }
1402 :
1403 : /* matched a right column? then keep as join column... */
1404 580 : if (m_name != NULL)
1405 204 : rlist = lappend(rlist, m_name);
1406 : }
1407 :
1408 172 : j->usingClause = rlist;
1409 : }
1410 :
1411 : /*
1412 : * If a USING clause alias was specified, save the USING columns as
1413 : * its column list.
1414 : */
1415 55359 : if (j->join_using_alias)
1416 56 : j->join_using_alias->colnames = j->usingClause;
1417 :
1418 : /*
1419 : * Now transform the join qualifications, if any.
1420 : */
1421 55359 : l_colnos = NIL;
1422 55359 : r_colnos = NIL;
1423 55359 : res_colnames = NIL;
1424 55359 : res_colvars = NIL;
1425 :
1426 : /* this may be larger than needed, but it's not worth being exact */
1427 : res_nscolumns = (ParseNamespaceColumn *)
1428 55359 : palloc0((list_length(l_colnames) + list_length(r_colnames)) *
1429 : sizeof(ParseNamespaceColumn));
1430 55359 : res_colindex = 0;
1431 :
1432 55359 : if (j->usingClause)
1433 : {
1434 : /*
1435 : * JOIN/USING (or NATURAL JOIN, as transformed above). Transform
1436 : * the list into an explicit ON-condition.
1437 : */
1438 1025 : List *ucols = j->usingClause;
1439 1025 : List *l_usingvars = NIL;
1440 1025 : List *r_usingvars = NIL;
1441 : ListCell *ucol;
1442 :
1443 : Assert(j->quals == NULL); /* shouldn't have ON() too */
1444 :
1445 2190 : foreach(ucol, ucols)
1446 : {
1447 1165 : char *u_colname = strVal(lfirst(ucol));
1448 : ListCell *col;
1449 : int ndx;
1450 1165 : int l_index = -1;
1451 1165 : int r_index = -1;
1452 : Var *l_colvar,
1453 : *r_colvar;
1454 :
1455 : Assert(u_colname[0] != '\0');
1456 :
1457 : /* Check for USING(foo,foo) */
1458 1323 : foreach(col, res_colnames)
1459 : {
1460 158 : char *res_colname = strVal(lfirst(col));
1461 :
1462 158 : if (strcmp(res_colname, u_colname) == 0)
1463 0 : ereport(ERROR,
1464 : (errcode(ERRCODE_DUPLICATE_COLUMN),
1465 : errmsg("column name \"%s\" appears more than once in USING clause",
1466 : u_colname)));
1467 : }
1468 :
1469 : /* Find it in left input */
1470 1165 : ndx = 0;
1471 5737 : foreach(col, l_colnames)
1472 : {
1473 4572 : char *l_colname = strVal(lfirst(col));
1474 :
1475 4572 : if (strcmp(l_colname, u_colname) == 0)
1476 : {
1477 1165 : if (l_index >= 0)
1478 0 : ereport(ERROR,
1479 : (errcode(ERRCODE_AMBIGUOUS_COLUMN),
1480 : errmsg("common column name \"%s\" appears more than once in left table",
1481 : u_colname)));
1482 1165 : l_index = ndx;
1483 : }
1484 4572 : ndx++;
1485 : }
1486 1165 : if (l_index < 0)
1487 0 : ereport(ERROR,
1488 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1489 : errmsg("column \"%s\" specified in USING clause does not exist in left table",
1490 : u_colname)));
1491 1165 : l_colnos = lappend_int(l_colnos, l_index + 1);
1492 :
1493 : /* Find it in right input */
1494 1165 : ndx = 0;
1495 5587 : foreach(col, r_colnames)
1496 : {
1497 4422 : char *r_colname = strVal(lfirst(col));
1498 :
1499 4422 : if (strcmp(r_colname, u_colname) == 0)
1500 : {
1501 1165 : if (r_index >= 0)
1502 0 : ereport(ERROR,
1503 : (errcode(ERRCODE_AMBIGUOUS_COLUMN),
1504 : errmsg("common column name \"%s\" appears more than once in right table",
1505 : u_colname)));
1506 1165 : r_index = ndx;
1507 : }
1508 4422 : ndx++;
1509 : }
1510 1165 : if (r_index < 0)
1511 0 : ereport(ERROR,
1512 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1513 : errmsg("column \"%s\" specified in USING clause does not exist in right table",
1514 : u_colname)));
1515 1165 : r_colnos = lappend_int(r_colnos, r_index + 1);
1516 :
1517 : /* Build Vars to use in the generated JOIN ON clause */
1518 1165 : l_colvar = buildVarFromNSColumn(pstate, l_nscolumns + l_index);
1519 1165 : l_usingvars = lappend(l_usingvars, l_colvar);
1520 1165 : r_colvar = buildVarFromNSColumn(pstate, r_nscolumns + r_index);
1521 1165 : r_usingvars = lappend(r_usingvars, r_colvar);
1522 :
1523 : /*
1524 : * While we're here, add column names to the res_colnames
1525 : * list. It's a bit ugly to do this here while the
1526 : * corresponding res_colvars entries are not made till later,
1527 : * but doing this later would require an additional traversal
1528 : * of the usingClause list.
1529 : */
1530 1165 : res_colnames = lappend(res_colnames, lfirst(ucol));
1531 : }
1532 :
1533 : /* Construct the generated JOIN ON clause */
1534 1025 : j->quals = transformJoinUsingClause(pstate,
1535 : l_usingvars,
1536 : r_usingvars);
1537 : }
1538 54334 : else if (j->quals)
1539 : {
1540 : /* User-written ON-condition; transform it */
1541 53988 : j->quals = transformJoinOnClause(pstate, j, my_namespace);
1542 : }
1543 : else
1544 : {
1545 : /* CROSS JOIN: no quals */
1546 : }
1547 :
1548 : /*
1549 : * If this is an outer join, now mark the appropriate child RTEs as
1550 : * being nulled by this join. We have finished processing the child
1551 : * join expressions as well as the current join's quals, which deal in
1552 : * non-nulled input columns. All future references to those RTEs will
1553 : * see possibly-nulled values, and we should mark generated Vars to
1554 : * account for that. In particular, the join alias Vars that we're
1555 : * about to build should reflect the nulling effects of this join.
1556 : *
1557 : * A difficulty with doing this is that we need the join's RT index,
1558 : * which we don't officially have yet. However, no other RTE can get
1559 : * made between here and the addRangeTableEntryForJoin call, so we can
1560 : * predict what the assignment will be. (Alternatively, we could call
1561 : * addRangeTableEntryForJoin before we have all the data computed, but
1562 : * this seems less ugly.)
1563 : */
1564 55347 : j->rtindex = list_length(pstate->p_rtable) + 1;
1565 :
1566 55347 : switch (j->jointype)
1567 : {
1568 28252 : case JOIN_INNER:
1569 28252 : break;
1570 26186 : case JOIN_LEFT:
1571 26186 : markRelsAsNulledBy(pstate, j->rarg, j->rtindex);
1572 26186 : break;
1573 669 : case JOIN_FULL:
1574 669 : markRelsAsNulledBy(pstate, j->larg, j->rtindex);
1575 669 : markRelsAsNulledBy(pstate, j->rarg, j->rtindex);
1576 669 : break;
1577 240 : case JOIN_RIGHT:
1578 240 : markRelsAsNulledBy(pstate, j->larg, j->rtindex);
1579 240 : break;
1580 0 : default:
1581 : /* shouldn't see any other types here */
1582 0 : elog(ERROR, "unrecognized join type: %d",
1583 : (int) j->jointype);
1584 : break;
1585 : }
1586 :
1587 : /*
1588 : * Now we can construct join alias expressions for the USING columns.
1589 : */
1590 55347 : if (j->usingClause)
1591 : {
1592 : ListCell *lc1,
1593 : *lc2;
1594 :
1595 : /* Scan the colnos lists to recover info from the previous loop */
1596 2190 : forboth(lc1, l_colnos, lc2, r_colnos)
1597 : {
1598 1165 : int l_index = lfirst_int(lc1) - 1;
1599 1165 : int r_index = lfirst_int(lc2) - 1;
1600 : Var *l_colvar,
1601 : *r_colvar;
1602 : Node *u_colvar;
1603 : ParseNamespaceColumn *res_nscolumn;
1604 :
1605 : /*
1606 : * Note we re-build these Vars: they might have different
1607 : * varnullingrels than the ones made in the previous loop.
1608 : */
1609 1165 : l_colvar = buildVarFromNSColumn(pstate, l_nscolumns + l_index);
1610 1165 : r_colvar = buildVarFromNSColumn(pstate, r_nscolumns + r_index);
1611 :
1612 : /* Construct the join alias Var for this column */
1613 1165 : u_colvar = buildMergedJoinVar(pstate,
1614 : j->jointype,
1615 : l_colvar,
1616 : r_colvar);
1617 1165 : res_colvars = lappend(res_colvars, u_colvar);
1618 :
1619 : /* Construct column's res_nscolumns[] entry */
1620 1165 : res_nscolumn = res_nscolumns + res_colindex;
1621 1165 : res_colindex++;
1622 1165 : if (u_colvar == (Node *) l_colvar)
1623 : {
1624 : /* Merged column is equivalent to left input */
1625 845 : *res_nscolumn = l_nscolumns[l_index];
1626 : }
1627 320 : else if (u_colvar == (Node *) r_colvar)
1628 : {
1629 : /* Merged column is equivalent to right input */
1630 28 : *res_nscolumn = r_nscolumns[r_index];
1631 : }
1632 : else
1633 : {
1634 : /*
1635 : * Merged column is not semantically equivalent to either
1636 : * input, so it needs to be referenced as the join output
1637 : * column.
1638 : */
1639 292 : res_nscolumn->p_varno = j->rtindex;
1640 292 : res_nscolumn->p_varattno = res_colindex;
1641 292 : res_nscolumn->p_vartype = exprType(u_colvar);
1642 292 : res_nscolumn->p_vartypmod = exprTypmod(u_colvar);
1643 292 : res_nscolumn->p_varcollid = exprCollation(u_colvar);
1644 292 : res_nscolumn->p_varnosyn = j->rtindex;
1645 292 : res_nscolumn->p_varattnosyn = res_colindex;
1646 : }
1647 : }
1648 : }
1649 :
1650 : /* Add remaining columns from each side to the output columns */
1651 55347 : res_colindex +=
1652 55347 : extractRemainingColumns(pstate,
1653 : l_nscolumns, l_colnames, &l_colnos,
1654 : &res_colnames, &res_colvars,
1655 55347 : res_nscolumns + res_colindex);
1656 55347 : res_colindex +=
1657 55347 : extractRemainingColumns(pstate,
1658 : r_nscolumns, r_colnames, &r_colnos,
1659 : &res_colnames, &res_colvars,
1660 55347 : res_nscolumns + res_colindex);
1661 :
1662 : /* If join has an alias, it syntactically hides all inputs */
1663 55347 : if (j->alias)
1664 : {
1665 668 : for (k = 0; k < res_colindex; k++)
1666 : {
1667 548 : ParseNamespaceColumn *nscol = res_nscolumns + k;
1668 :
1669 548 : nscol->p_varnosyn = j->rtindex;
1670 548 : nscol->p_varattnosyn = k + 1;
1671 : }
1672 : }
1673 :
1674 : /*
1675 : * Now build an RTE and nsitem for the result of the join.
1676 : */
1677 55347 : nsitem = addRangeTableEntryForJoin(pstate,
1678 : res_colnames,
1679 : res_nscolumns,
1680 : j->jointype,
1681 55347 : list_length(j->usingClause),
1682 : res_colvars,
1683 : l_colnos,
1684 : r_colnos,
1685 : j->join_using_alias,
1686 : j->alias,
1687 : true);
1688 :
1689 : /* Verify that we correctly predicted the join's RT index */
1690 : Assert(j->rtindex == nsitem->p_rtindex);
1691 : /* Cross-check number of columns, too */
1692 : Assert(res_colindex == list_length(nsitem->p_names->colnames));
1693 :
1694 : /*
1695 : * Save a link to the JoinExpr in the proper element of p_joinexprs.
1696 : * Since we maintain that list lazily, it may be necessary to fill in
1697 : * empty entries before we can add the JoinExpr in the right place.
1698 : */
1699 145704 : for (k = list_length(pstate->p_joinexprs) + 1; k < j->rtindex; k++)
1700 90361 : pstate->p_joinexprs = lappend(pstate->p_joinexprs, NULL);
1701 55343 : pstate->p_joinexprs = lappend(pstate->p_joinexprs, j);
1702 : Assert(list_length(pstate->p_joinexprs) == j->rtindex);
1703 :
1704 : /*
1705 : * If the join has a USING alias, build a ParseNamespaceItem for that
1706 : * and add it to the list of nsitems in the join's input.
1707 : */
1708 55343 : if (j->join_using_alias)
1709 : {
1710 : ParseNamespaceItem *jnsitem;
1711 :
1712 56 : jnsitem = palloc_object(ParseNamespaceItem);
1713 56 : jnsitem->p_names = j->join_using_alias;
1714 56 : jnsitem->p_rte = nsitem->p_rte;
1715 56 : jnsitem->p_rtindex = nsitem->p_rtindex;
1716 56 : jnsitem->p_perminfo = NULL;
1717 : /* no need to copy the first N columns, just use res_nscolumns */
1718 56 : jnsitem->p_nscolumns = res_nscolumns;
1719 : /* set default visibility flags; might get changed later */
1720 56 : jnsitem->p_rel_visible = true;
1721 56 : jnsitem->p_cols_visible = true;
1722 56 : jnsitem->p_lateral_only = false;
1723 56 : jnsitem->p_lateral_ok = true;
1724 56 : jnsitem->p_returning_type = VAR_RETURNING_DEFAULT;
1725 : /* Per SQL, we must check for alias conflicts */
1726 56 : checkNameSpaceConflicts(pstate, list_make1(jnsitem), my_namespace);
1727 52 : my_namespace = lappend(my_namespace, jnsitem);
1728 : }
1729 :
1730 : /*
1731 : * Prepare returned namespace list. If the JOIN has an alias then it
1732 : * hides the contained RTEs completely; otherwise, the contained RTEs
1733 : * are still visible as table names, but are not visible for
1734 : * unqualified column-name access.
1735 : *
1736 : * Note: if there are nested alias-less JOINs, the lower-level ones
1737 : * will remain in the list although they have neither p_rel_visible
1738 : * nor p_cols_visible set. We could delete such list items, but it's
1739 : * unclear that it's worth expending cycles to do so.
1740 : */
1741 55339 : if (j->alias != NULL)
1742 116 : my_namespace = NIL;
1743 : else
1744 55223 : setNamespaceColumnVisibility(my_namespace, false);
1745 :
1746 : /*
1747 : * The join RTE itself is always made visible for unqualified column
1748 : * names. It's visible as a relation name only if it has an alias.
1749 : */
1750 55339 : nsitem->p_rel_visible = (j->alias != NULL);
1751 55339 : nsitem->p_cols_visible = true;
1752 55339 : nsitem->p_lateral_only = false;
1753 55339 : nsitem->p_lateral_ok = true;
1754 :
1755 55339 : *top_nsitem = nsitem;
1756 55339 : *namespace = lappend(my_namespace, nsitem);
1757 :
1758 55339 : return (Node *) j;
1759 : }
1760 : else
1761 0 : elog(ERROR, "unrecognized node type: %d", (int) nodeTag(n));
1762 : return NULL; /* can't get here, keep compiler quiet */
1763 : }
1764 :
1765 : /*
1766 : * buildVarFromNSColumn -
1767 : * build a Var node using ParseNamespaceColumn data
1768 : *
1769 : * This is used to construct joinaliasvars entries.
1770 : * We can assume varlevelsup should be 0, and no location is specified.
1771 : * Note also that no column SELECT privilege is requested here; that would
1772 : * happen only if the column is actually referenced in the query.
1773 : */
1774 : static Var *
1775 2131398 : buildVarFromNSColumn(ParseState *pstate, ParseNamespaceColumn *nscol)
1776 : {
1777 : Var *var;
1778 :
1779 : Assert(nscol->p_varno > 0); /* i.e., not deleted column */
1780 2131398 : var = makeVar(nscol->p_varno,
1781 2131398 : nscol->p_varattno,
1782 : nscol->p_vartype,
1783 : nscol->p_vartypmod,
1784 : nscol->p_varcollid,
1785 : 0);
1786 : /* makeVar doesn't offer parameters for these, so set by hand: */
1787 2131398 : var->varreturningtype = nscol->p_varreturningtype;
1788 2131398 : var->varnosyn = nscol->p_varnosyn;
1789 2131398 : var->varattnosyn = nscol->p_varattnosyn;
1790 :
1791 : /* ... and update varnullingrels */
1792 2131398 : markNullableIfNeeded(pstate, var);
1793 :
1794 2131398 : return var;
1795 : }
1796 :
1797 : /*
1798 : * buildMergedJoinVar -
1799 : * generate a suitable replacement expression for a merged join column
1800 : */
1801 : static Node *
1802 1165 : buildMergedJoinVar(ParseState *pstate, JoinType jointype,
1803 : Var *l_colvar, Var *r_colvar)
1804 : {
1805 : Oid outcoltype;
1806 : int32 outcoltypmod;
1807 : Node *l_node,
1808 : *r_node,
1809 : *res_node;
1810 :
1811 1165 : outcoltype = select_common_type(pstate,
1812 1165 : list_make2(l_colvar, r_colvar),
1813 : "JOIN/USING",
1814 : NULL);
1815 1165 : outcoltypmod = select_common_typmod(pstate,
1816 1165 : list_make2(l_colvar, r_colvar),
1817 : outcoltype);
1818 :
1819 : /*
1820 : * Insert coercion functions if needed. Note that a difference in typmod
1821 : * can only happen if input has typmod but outcoltypmod is -1. In that
1822 : * case we insert a RelabelType to clearly mark that result's typmod is
1823 : * not same as input. We never need coerce_type_typmod.
1824 : */
1825 1165 : if (l_colvar->vartype != outcoltype)
1826 60 : l_node = coerce_type(pstate, (Node *) l_colvar, l_colvar->vartype,
1827 : outcoltype, outcoltypmod,
1828 : COERCION_IMPLICIT, COERCE_IMPLICIT_CAST, -1);
1829 1105 : else if (l_colvar->vartypmod != outcoltypmod)
1830 0 : l_node = (Node *) makeRelabelType((Expr *) l_colvar,
1831 : outcoltype, outcoltypmod,
1832 : InvalidOid, /* fixed below */
1833 : COERCE_IMPLICIT_CAST);
1834 : else
1835 1105 : l_node = (Node *) l_colvar;
1836 :
1837 1165 : if (r_colvar->vartype != outcoltype)
1838 20 : r_node = coerce_type(pstate, (Node *) r_colvar, r_colvar->vartype,
1839 : outcoltype, outcoltypmod,
1840 : COERCION_IMPLICIT, COERCE_IMPLICIT_CAST, -1);
1841 1145 : else if (r_colvar->vartypmod != outcoltypmod)
1842 0 : r_node = (Node *) makeRelabelType((Expr *) r_colvar,
1843 : outcoltype, outcoltypmod,
1844 : InvalidOid, /* fixed below */
1845 : COERCE_IMPLICIT_CAST);
1846 : else
1847 1145 : r_node = (Node *) r_colvar;
1848 :
1849 : /*
1850 : * Choose what to emit
1851 : */
1852 1165 : switch (jointype)
1853 : {
1854 753 : case JOIN_INNER:
1855 :
1856 : /*
1857 : * We can use either var; prefer non-coerced one if available.
1858 : */
1859 753 : if (IsA(l_node, Var))
1860 733 : res_node = l_node;
1861 20 : else if (IsA(r_node, Var))
1862 20 : res_node = r_node;
1863 : else
1864 0 : res_node = l_node;
1865 753 : break;
1866 152 : case JOIN_LEFT:
1867 : /* Always use left var */
1868 152 : res_node = l_node;
1869 152 : break;
1870 8 : case JOIN_RIGHT:
1871 : /* Always use right var */
1872 8 : res_node = r_node;
1873 8 : break;
1874 252 : case JOIN_FULL:
1875 : {
1876 : /*
1877 : * Here we must build a COALESCE expression to ensure that the
1878 : * join output is non-null if either input is.
1879 : */
1880 252 : CoalesceExpr *c = makeNode(CoalesceExpr);
1881 :
1882 252 : c->coalescetype = outcoltype;
1883 : /* coalescecollid will get set below */
1884 252 : c->args = list_make2(l_node, r_node);
1885 252 : c->location = -1;
1886 252 : res_node = (Node *) c;
1887 252 : break;
1888 : }
1889 0 : default:
1890 0 : elog(ERROR, "unrecognized join type: %d", (int) jointype);
1891 : res_node = NULL; /* keep compiler quiet */
1892 : break;
1893 : }
1894 :
1895 : /*
1896 : * Apply assign_expr_collations to fix up the collation info in the
1897 : * coercion and CoalesceExpr nodes, if we made any. This must be done now
1898 : * so that the join node's alias vars show correct collation info.
1899 : */
1900 1165 : assign_expr_collations(pstate, res_node);
1901 :
1902 1165 : return res_node;
1903 : }
1904 :
1905 : /*
1906 : * markRelsAsNulledBy -
1907 : * Mark the given jointree node and its children as nulled by join jindex
1908 : */
1909 : static void
1910 30014 : markRelsAsNulledBy(ParseState *pstate, Node *n, int jindex)
1911 : {
1912 : int varno;
1913 : ListCell *lc;
1914 :
1915 : /* Note: we can't see FromExpr here */
1916 30014 : if (IsA(n, RangeTblRef))
1917 : {
1918 28889 : varno = ((RangeTblRef *) n)->rtindex;
1919 : }
1920 1125 : else if (IsA(n, JoinExpr))
1921 : {
1922 1125 : JoinExpr *j = (JoinExpr *) n;
1923 :
1924 : /* recurse to children */
1925 1125 : markRelsAsNulledBy(pstate, j->larg, jindex);
1926 1125 : markRelsAsNulledBy(pstate, j->rarg, jindex);
1927 1125 : varno = j->rtindex;
1928 : }
1929 : else
1930 : {
1931 0 : elog(ERROR, "unrecognized node type: %d", (int) nodeTag(n));
1932 : varno = 0; /* keep compiler quiet */
1933 : }
1934 :
1935 : /*
1936 : * Now add jindex to the p_nullingrels set for relation varno. Since we
1937 : * maintain the p_nullingrels list lazily, we might need to extend it to
1938 : * make the varno'th entry exist.
1939 : */
1940 97468 : while (list_length(pstate->p_nullingrels) < varno)
1941 67454 : pstate->p_nullingrels = lappend(pstate->p_nullingrels, NULL);
1942 30014 : lc = list_nth_cell(pstate->p_nullingrels, varno - 1);
1943 30014 : lfirst(lc) = bms_add_member((Bitmapset *) lfirst(lc), jindex);
1944 30014 : }
1945 :
1946 : /*
1947 : * setNamespaceColumnVisibility -
1948 : * Convenience subroutine to update cols_visible flags in a namespace list.
1949 : */
1950 : static void
1951 55223 : setNamespaceColumnVisibility(List *namespace, bool cols_visible)
1952 : {
1953 : ListCell *lc;
1954 :
1955 232985 : foreach(lc, namespace)
1956 : {
1957 177762 : ParseNamespaceItem *nsitem = (ParseNamespaceItem *) lfirst(lc);
1958 :
1959 177762 : nsitem->p_cols_visible = cols_visible;
1960 : }
1961 55223 : }
1962 :
1963 : /*
1964 : * setNamespaceLateralState -
1965 : * Convenience subroutine to update LATERAL flags in a namespace list.
1966 : */
1967 : static void
1968 672975 : setNamespaceLateralState(List *namespace, bool lateral_only, bool lateral_ok)
1969 : {
1970 : ListCell *lc;
1971 :
1972 1699241 : foreach(lc, namespace)
1973 : {
1974 1026266 : ParseNamespaceItem *nsitem = (ParseNamespaceItem *) lfirst(lc);
1975 :
1976 1026266 : nsitem->p_lateral_only = lateral_only;
1977 1026266 : nsitem->p_lateral_ok = lateral_ok;
1978 : }
1979 672975 : }
1980 :
1981 :
1982 : /*
1983 : * transformWhereClause -
1984 : * Transform the qualification and make sure it is of type boolean.
1985 : * Used for WHERE and allied clauses.
1986 : *
1987 : * constructName does not affect the semantics, but is used in error messages
1988 : */
1989 : Node *
1990 667538 : transformWhereClause(ParseState *pstate, Node *clause,
1991 : ParseExprKind exprKind, const char *constructName)
1992 : {
1993 : Node *qual;
1994 :
1995 667538 : if (clause == NULL)
1996 460919 : return NULL;
1997 :
1998 206619 : qual = transformExpr(pstate, clause, exprKind);
1999 :
2000 206480 : qual = coerce_to_boolean(pstate, qual, constructName);
2001 :
2002 206476 : return qual;
2003 : }
2004 :
2005 :
2006 : /*
2007 : * transformLimitClause -
2008 : * Transform the expression and make sure it is of type bigint.
2009 : * Used for LIMIT and allied clauses.
2010 : *
2011 : * Note: as of Postgres 8.2, LIMIT expressions are expected to yield int8,
2012 : * rather than int4 as before.
2013 : *
2014 : * constructName does not affect the semantics, but is used in error messages
2015 : */
2016 : Node *
2017 623846 : transformLimitClause(ParseState *pstate, Node *clause,
2018 : ParseExprKind exprKind, const char *constructName,
2019 : LimitOption limitOption)
2020 : {
2021 : Node *qual;
2022 :
2023 623846 : if (clause == NULL)
2024 620419 : return NULL;
2025 :
2026 3427 : qual = transformExpr(pstate, clause, exprKind);
2027 :
2028 3423 : qual = coerce_to_specific_type(pstate, qual, INT8OID, constructName);
2029 :
2030 : /* LIMIT can't refer to any variables of the current query */
2031 3423 : checkExprIsVarFree(pstate, qual, constructName);
2032 :
2033 : /*
2034 : * Don't allow NULLs in FETCH FIRST .. WITH TIES. This test is ugly and
2035 : * extremely simplistic, in that you can pass a NULL anyway by hiding it
2036 : * inside an expression -- but this protects ruleutils against emitting an
2037 : * unadorned NULL that's not accepted back by the grammar.
2038 : */
2039 3423 : if (exprKind == EXPR_KIND_LIMIT && limitOption == LIMIT_OPTION_WITH_TIES &&
2040 38 : IsA(clause, A_Const) && castNode(A_Const, clause)->isnull)
2041 4 : ereport(ERROR,
2042 : (errcode(ERRCODE_INVALID_ROW_COUNT_IN_LIMIT_CLAUSE),
2043 : errmsg("row count cannot be null in FETCH FIRST ... WITH TIES clause")));
2044 :
2045 3419 : return qual;
2046 : }
2047 :
2048 : /*
2049 : * checkExprIsVarFree
2050 : * Check that given expr has no Vars of the current query level
2051 : * (aggregates and window functions should have been rejected already).
2052 : *
2053 : * This is used to check expressions that have to have a consistent value
2054 : * across all rows of the query, such as a LIMIT. Arguably it should reject
2055 : * volatile functions, too, but we don't do that --- whatever value the
2056 : * function gives on first execution is what you get.
2057 : *
2058 : * constructName does not affect the semantics, but is used in error messages
2059 : */
2060 : static void
2061 4697 : checkExprIsVarFree(ParseState *pstate, Node *n, const char *constructName)
2062 : {
2063 4697 : if (contain_vars_of_level(n, 0))
2064 : {
2065 4 : ereport(ERROR,
2066 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
2067 : /* translator: %s is name of a SQL construct, eg LIMIT */
2068 : errmsg("argument of %s must not contain variables",
2069 : constructName),
2070 : parser_errposition(pstate,
2071 : locate_var_of_level(n, 0))));
2072 : }
2073 4693 : }
2074 :
2075 :
2076 : /*
2077 : * checkTargetlistEntrySQL92 -
2078 : * Validate a targetlist entry found by findTargetlistEntrySQL92
2079 : *
2080 : * When we select a pre-existing tlist entry as a result of syntax such
2081 : * as "GROUP BY 1", we have to make sure it is acceptable for use in the
2082 : * indicated clause type; transformExpr() will have treated it as a regular
2083 : * targetlist item.
2084 : */
2085 : static void
2086 50300 : checkTargetlistEntrySQL92(ParseState *pstate, TargetEntry *tle,
2087 : ParseExprKind exprKind)
2088 : {
2089 50300 : switch (exprKind)
2090 : {
2091 518 : case EXPR_KIND_GROUP_BY:
2092 : /* reject aggregates and window functions */
2093 928 : if (pstate->p_hasAggs &&
2094 410 : contain_aggs_of_level((Node *) tle->expr, 0))
2095 0 : ereport(ERROR,
2096 : (errcode(ERRCODE_GROUPING_ERROR),
2097 : /* translator: %s is name of a SQL construct, eg GROUP BY */
2098 : errmsg("aggregate functions are not allowed in %s",
2099 : ParseExprKindName(exprKind)),
2100 : parser_errposition(pstate,
2101 : locate_agg_of_level((Node *) tle->expr, 0))));
2102 526 : if (pstate->p_hasWindowFuncs &&
2103 8 : contain_windowfuncs((Node *) tle->expr))
2104 4 : ereport(ERROR,
2105 : (errcode(ERRCODE_WINDOWING_ERROR),
2106 : /* translator: %s is name of a SQL construct, eg GROUP BY */
2107 : errmsg("window functions are not allowed in %s",
2108 : ParseExprKindName(exprKind)),
2109 : parser_errposition(pstate,
2110 : locate_windowfunc((Node *) tle->expr))));
2111 514 : break;
2112 49614 : case EXPR_KIND_ORDER_BY:
2113 : /* no extra checks needed */
2114 49614 : break;
2115 168 : case EXPR_KIND_DISTINCT_ON:
2116 : /* no extra checks needed */
2117 168 : break;
2118 0 : default:
2119 0 : elog(ERROR, "unexpected exprKind in checkTargetlistEntrySQL92");
2120 : break;
2121 : }
2122 50296 : }
2123 :
2124 : /*
2125 : * findTargetlistEntrySQL92 -
2126 : * Returns the targetlist entry matching the given (untransformed) node.
2127 : * If no matching entry exists, one is created and appended to the target
2128 : * list as a "resjunk" node.
2129 : *
2130 : * This function supports the old SQL92 ORDER BY interpretation, where the
2131 : * expression is an output column name or number. If we fail to find a
2132 : * match of that sort, we fall through to the SQL99 rules. For historical
2133 : * reasons, Postgres also allows this interpretation for GROUP BY, though
2134 : * the standard never did. However, for GROUP BY we prefer a SQL99 match.
2135 : * This function is *not* used for WINDOW definitions.
2136 : *
2137 : * node the ORDER BY, GROUP BY, or DISTINCT ON expression to be matched
2138 : * tlist the target list (passed by reference so we can append to it)
2139 : * exprKind identifies clause type being processed
2140 : */
2141 : static TargetEntry *
2142 76745 : findTargetlistEntrySQL92(ParseState *pstate, Node *node, List **tlist,
2143 : ParseExprKind exprKind)
2144 : {
2145 : ListCell *tl;
2146 :
2147 : /*----------
2148 : * Handle two special cases as mandated by the SQL92 spec:
2149 : *
2150 : * 1. Bare ColumnName (no qualifier or subscripts)
2151 : * For a bare identifier, we search for a matching column name
2152 : * in the existing target list. Multiple matches are an error
2153 : * unless they refer to identical values; for example,
2154 : * we allow SELECT a, a FROM table ORDER BY a
2155 : * but not SELECT a AS b, b FROM table ORDER BY b
2156 : * If no match is found, we fall through and treat the identifier
2157 : * as an expression.
2158 : * For GROUP BY, it is incorrect to match the grouping item against
2159 : * targetlist entries: according to SQL92, an identifier in GROUP BY
2160 : * is a reference to a column name exposed by FROM, not to a target
2161 : * list column. However, many implementations (including pre-7.0
2162 : * PostgreSQL) accept this anyway. So for GROUP BY, we look first
2163 : * to see if the identifier matches any FROM column name, and only
2164 : * try for a targetlist name if it doesn't. This ensures that we
2165 : * adhere to the spec in the case where the name could be both.
2166 : * DISTINCT ON isn't in the standard, so we can do what we like there;
2167 : * we choose to make it work like ORDER BY, on the rather flimsy
2168 : * grounds that ordinary DISTINCT works on targetlist entries.
2169 : *
2170 : * 2. IntegerConstant
2171 : * This means to use the n'th item in the existing target list.
2172 : * Note that it would make no sense to order/group/distinct by an
2173 : * actual constant, so this does not create a conflict with SQL99.
2174 : * GROUP BY column-number is not allowed by SQL92, but since
2175 : * the standard has no other behavior defined for this syntax,
2176 : * we may as well accept this common extension.
2177 : *
2178 : * Note that pre-existing resjunk targets must not be used in either case,
2179 : * since the user didn't write them in his SELECT list.
2180 : *
2181 : * If neither special case applies, fall through to treat the item as
2182 : * an expression per SQL99.
2183 : *----------
2184 : */
2185 120323 : if (IsA(node, ColumnRef) &&
2186 43578 : list_length(((ColumnRef *) node)->fields) == 1 &&
2187 31188 : IsA(linitial(((ColumnRef *) node)->fields), String))
2188 : {
2189 31188 : char *name = strVal(linitial(((ColumnRef *) node)->fields));
2190 31188 : int location = ((ColumnRef *) node)->location;
2191 :
2192 31188 : if (exprKind == EXPR_KIND_GROUP_BY)
2193 : {
2194 : /*
2195 : * In GROUP BY, we must prefer a match against a FROM-clause
2196 : * column to one against the targetlist. Look to see if there is
2197 : * a matching column. If so, fall through to use SQL99 rules.
2198 : * NOTE: if name could refer ambiguously to more than one column
2199 : * name exposed by FROM, colNameToVar will ereport(ERROR). That's
2200 : * just what we want here.
2201 : *
2202 : * Small tweak for 7.4.3: ignore matches in upper query levels.
2203 : * This effectively changes the search order for bare names to (1)
2204 : * local FROM variables, (2) local targetlist aliases, (3) outer
2205 : * FROM variables, whereas before it was (1) (3) (2). SQL92 and
2206 : * SQL99 do not allow GROUPing BY an outer reference, so this
2207 : * breaks no cases that are legal per spec, and it seems a more
2208 : * self-consistent behavior.
2209 : */
2210 3344 : if (colNameToVar(pstate, name, true, location) != NULL)
2211 3252 : name = NULL;
2212 : }
2213 :
2214 31188 : if (name != NULL)
2215 : {
2216 27936 : TargetEntry *target_result = NULL;
2217 :
2218 154808 : foreach(tl, *tlist)
2219 : {
2220 126872 : TargetEntry *tle = (TargetEntry *) lfirst(tl);
2221 :
2222 126872 : if (!tle->resjunk &&
2223 126231 : strcmp(tle->resname, name) == 0)
2224 : {
2225 24136 : if (target_result != NULL)
2226 : {
2227 6 : if (!equal(target_result->expr, tle->expr))
2228 0 : ereport(ERROR,
2229 : (errcode(ERRCODE_AMBIGUOUS_COLUMN),
2230 :
2231 : /*------
2232 : translator: first %s is name of a SQL construct, eg ORDER BY */
2233 : errmsg("%s \"%s\" is ambiguous",
2234 : ParseExprKindName(exprKind),
2235 : name),
2236 : parser_errposition(pstate, location)));
2237 : }
2238 : else
2239 24130 : target_result = tle;
2240 : /* Stay in loop to check for ambiguity */
2241 : }
2242 : }
2243 27936 : if (target_result != NULL)
2244 : {
2245 : /* return the first match, after suitable validation */
2246 24130 : checkTargetlistEntrySQL92(pstate, target_result, exprKind);
2247 24130 : return target_result;
2248 : }
2249 : }
2250 : }
2251 52615 : if (IsA(node, A_Const))
2252 : {
2253 26174 : A_Const *aconst = castNode(A_Const, node);
2254 26174 : int targetlist_pos = 0;
2255 : int target_pos;
2256 :
2257 26174 : if (!IsA(&aconst->val, Integer))
2258 0 : ereport(ERROR,
2259 : (errcode(ERRCODE_SYNTAX_ERROR),
2260 : /* translator: %s is name of a SQL construct, eg ORDER BY */
2261 : errmsg("non-integer constant in %s",
2262 : ParseExprKindName(exprKind)),
2263 : parser_errposition(pstate, aconst->location)));
2264 :
2265 26174 : target_pos = intVal(&aconst->val);
2266 44958 : foreach(tl, *tlist)
2267 : {
2268 44954 : TargetEntry *tle = (TargetEntry *) lfirst(tl);
2269 :
2270 44954 : if (!tle->resjunk)
2271 : {
2272 44954 : if (++targetlist_pos == target_pos)
2273 : {
2274 : /* return the unique match, after suitable validation */
2275 26170 : checkTargetlistEntrySQL92(pstate, tle, exprKind);
2276 26166 : return tle;
2277 : }
2278 : }
2279 : }
2280 4 : ereport(ERROR,
2281 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
2282 : /* translator: %s is name of a SQL construct, eg ORDER BY */
2283 : errmsg("%s position %d is not in select list",
2284 : ParseExprKindName(exprKind), target_pos),
2285 : parser_errposition(pstate, aconst->location)));
2286 : }
2287 :
2288 : /*
2289 : * Otherwise, we have an expression, so process it per SQL99 rules.
2290 : */
2291 26441 : return findTargetlistEntrySQL99(pstate, node, tlist, exprKind);
2292 : }
2293 :
2294 : /*
2295 : * findTargetlistEntrySQL99 -
2296 : * Returns the targetlist entry matching the given (untransformed) node.
2297 : * If no matching entry exists, one is created and appended to the target
2298 : * list as a "resjunk" node.
2299 : *
2300 : * This function supports the SQL99 interpretation, wherein the expression
2301 : * is just an ordinary expression referencing input column names.
2302 : *
2303 : * node the ORDER BY, GROUP BY, etc expression to be matched
2304 : * tlist the target list (passed by reference so we can append to it)
2305 : * exprKind identifies clause type being processed
2306 : */
2307 : static TargetEntry *
2308 30458 : findTargetlistEntrySQL99(ParseState *pstate, Node *node, List **tlist,
2309 : ParseExprKind exprKind)
2310 : {
2311 : TargetEntry *target_result;
2312 : ListCell *tl;
2313 : Node *expr;
2314 :
2315 : /*
2316 : * Convert the untransformed node to a transformed expression, and search
2317 : * for a match in the tlist. NOTE: it doesn't really matter whether there
2318 : * is more than one match. Also, we are willing to match an existing
2319 : * resjunk target here, though the SQL92 cases above must ignore resjunk
2320 : * targets.
2321 : */
2322 30458 : expr = transformExpr(pstate, node, exprKind);
2323 :
2324 113473 : foreach(tl, *tlist)
2325 : {
2326 95178 : TargetEntry *tle = (TargetEntry *) lfirst(tl);
2327 : Node *texpr;
2328 :
2329 : /*
2330 : * Ignore any implicit cast on the existing tlist expression.
2331 : *
2332 : * This essentially allows the ORDER/GROUP/etc item to adopt the same
2333 : * datatype previously selected for a textually-equivalent tlist item.
2334 : * There can't be any implicit cast at top level in an ordinary SELECT
2335 : * tlist at this stage, but the case does arise with ORDER BY in an
2336 : * aggregate function.
2337 : */
2338 95178 : texpr = strip_implicit_coercions((Node *) tle->expr);
2339 :
2340 95178 : if (equal(expr, texpr))
2341 12127 : return tle;
2342 : }
2343 :
2344 : /*
2345 : * If no matches, construct a new target entry which is appended to the
2346 : * end of the target list. This target is given resjunk = true so that it
2347 : * will not be projected into the final tuple.
2348 : */
2349 18295 : target_result = transformTargetEntry(pstate, node, expr, exprKind,
2350 : NULL, true);
2351 :
2352 18295 : *tlist = lappend(*tlist, target_result);
2353 :
2354 18295 : return target_result;
2355 : }
2356 :
2357 : /*-------------------------------------------------------------------------
2358 : * Flatten out parenthesized sublists in grouping lists, and some cases
2359 : * of nested grouping sets.
2360 : *
2361 : * Inside a grouping set (ROLLUP, CUBE, or GROUPING SETS), we expect the
2362 : * content to be nested no more than 2 deep: i.e. ROLLUP((a,b),(c,d)) is
2363 : * ok, but ROLLUP((a,(b,c)),d) is flattened to ((a,b,c),d), which we then
2364 : * (later) normalize to ((a,b,c),(d)).
2365 : *
2366 : * CUBE or ROLLUP can be nested inside GROUPING SETS (but not the reverse),
2367 : * and we leave that alone if we find it. But if we see GROUPING SETS inside
2368 : * GROUPING SETS, we can flatten and normalize as follows:
2369 : * GROUPING SETS (a, (b,c), GROUPING SETS ((c,d),(e)), (f,g))
2370 : * becomes
2371 : * GROUPING SETS ((a), (b,c), (c,d), (e), (f,g))
2372 : *
2373 : * This is per the spec's syntax transformations, but these are the only such
2374 : * transformations we do in parse analysis, so that queries retain the
2375 : * originally specified grouping set syntax for CUBE and ROLLUP as much as
2376 : * possible when deparsed. (Full expansion of the result into a list of
2377 : * grouping sets is left to the planner.)
2378 : *
2379 : * When we're done, the resulting list should contain only these possible
2380 : * elements:
2381 : * - an expression
2382 : * - a CUBE or ROLLUP with a list of expressions nested 2 deep
2383 : * - a GROUPING SET containing any of:
2384 : * - expression lists
2385 : * - empty grouping sets
2386 : * - CUBE or ROLLUP nodes with lists nested 2 deep
2387 : * The return is a new list, but doesn't deep-copy the old nodes except for
2388 : * GroupingSet nodes.
2389 : *
2390 : * As a side effect, flag whether the list has any GroupingSet nodes.
2391 : *-------------------------------------------------------------------------
2392 : */
2393 : static Node *
2394 307766 : flatten_grouping_sets(Node *expr, bool toplevel, bool *hasGroupingSets)
2395 : {
2396 : /* just in case of pathological input */
2397 307766 : check_stack_depth();
2398 :
2399 307766 : if (expr == (Node *) NIL)
2400 296096 : return (Node *) NIL;
2401 :
2402 11670 : switch (expr->type)
2403 : {
2404 242 : case T_RowExpr:
2405 : {
2406 242 : RowExpr *r = (RowExpr *) expr;
2407 :
2408 242 : if (r->row_format == COERCE_IMPLICIT_CAST)
2409 242 : return flatten_grouping_sets((Node *) r->args,
2410 : false, NULL);
2411 : }
2412 0 : break;
2413 1036 : case T_GroupingSet:
2414 : {
2415 1036 : GroupingSet *gset = (GroupingSet *) expr;
2416 : ListCell *l2;
2417 1036 : List *result_set = NIL;
2418 :
2419 1036 : if (hasGroupingSets)
2420 754 : *hasGroupingSets = true;
2421 :
2422 : /*
2423 : * at the top level, we skip over all empty grouping sets; the
2424 : * caller can supply the canonical GROUP BY () if nothing is
2425 : * left.
2426 : */
2427 :
2428 1036 : if (toplevel && gset->kind == GROUPING_SET_EMPTY)
2429 24 : return (Node *) NIL;
2430 :
2431 2636 : foreach(l2, gset->content)
2432 : {
2433 1624 : Node *n1 = lfirst(l2);
2434 1624 : Node *n2 = flatten_grouping_sets(n1, false, NULL);
2435 :
2436 1624 : if (IsA(n1, GroupingSet) &&
2437 282 : ((GroupingSet *) n1)->kind == GROUPING_SET_SETS)
2438 68 : result_set = list_concat(result_set, (List *) n2);
2439 : else
2440 1556 : result_set = lappend(result_set, n2);
2441 : }
2442 :
2443 : /*
2444 : * At top level, keep the grouping set node; but if we're in a
2445 : * nested grouping set, then we need to concat the flattened
2446 : * result into the outer list if it's simply nested.
2447 : */
2448 :
2449 1012 : if (toplevel || (gset->kind != GROUPING_SET_SETS))
2450 : {
2451 944 : return (Node *) makeGroupingSet(gset->kind, result_set, gset->location);
2452 : }
2453 : else
2454 68 : return (Node *) result_set;
2455 : }
2456 4055 : case T_List:
2457 : {
2458 4055 : List *result = NIL;
2459 : ListCell *l;
2460 :
2461 10046 : foreach(l, (List *) expr)
2462 : {
2463 5991 : Node *n = flatten_grouping_sets(lfirst(l), toplevel, hasGroupingSets);
2464 :
2465 5991 : if (n != (Node *) NIL)
2466 : {
2467 5967 : if (IsA(n, List))
2468 30 : result = list_concat(result, (List *) n);
2469 : else
2470 5937 : result = lappend(result, n);
2471 : }
2472 : }
2473 :
2474 4055 : return (Node *) result;
2475 : }
2476 6337 : default:
2477 6337 : break;
2478 : }
2479 :
2480 6337 : return expr;
2481 : }
2482 :
2483 : /*
2484 : * Transform a single expression within a GROUP BY clause or grouping set.
2485 : *
2486 : * The expression is added to the targetlist if not already present, and to the
2487 : * flatresult list (which will become the groupClause) if not already present
2488 : * there. The sortClause is consulted for operator and sort order hints.
2489 : *
2490 : * Returns the ressortgroupref of the expression.
2491 : *
2492 : * flatresult reference to flat list of SortGroupClause nodes
2493 : * seen_local bitmapset of sortgrouprefs already seen at the local level
2494 : * pstate ParseState
2495 : * gexpr node to transform
2496 : * targetlist reference to TargetEntry list
2497 : * sortClause ORDER BY clause (SortGroupClause nodes)
2498 : * exprKind expression kind
2499 : * useSQL99 SQL99 rather than SQL92 syntax
2500 : * toplevel false if within any grouping set
2501 : */
2502 : static Index
2503 6337 : transformGroupClauseExpr(List **flatresult, Bitmapset *seen_local,
2504 : ParseState *pstate, Node *gexpr,
2505 : List **targetlist, List *sortClause,
2506 : ParseExprKind exprKind, bool useSQL99, bool toplevel)
2507 : {
2508 : TargetEntry *tle;
2509 6337 : bool found = false;
2510 :
2511 6337 : if (useSQL99)
2512 756 : tle = findTargetlistEntrySQL99(pstate, gexpr,
2513 : targetlist, exprKind);
2514 : else
2515 5581 : tle = findTargetlistEntrySQL92(pstate, gexpr,
2516 : targetlist, exprKind);
2517 :
2518 6321 : if (tle->ressortgroupref > 0)
2519 : {
2520 : ListCell *sl;
2521 :
2522 : /*
2523 : * Eliminate duplicates (GROUP BY x, x) but only at local level.
2524 : * (Duplicates in grouping sets can affect the number of returned
2525 : * rows, so can't be dropped indiscriminately.)
2526 : *
2527 : * Since we don't care about anything except the sortgroupref, we can
2528 : * use a bitmapset rather than scanning lists.
2529 : */
2530 1976 : if (bms_is_member(tle->ressortgroupref, seen_local))
2531 16 : return 0;
2532 :
2533 : /*
2534 : * If we're already in the flat clause list, we don't need to consider
2535 : * adding ourselves again.
2536 : */
2537 1960 : found = targetIsInSortList(tle, InvalidOid, *flatresult);
2538 1960 : if (found)
2539 174 : return tle->ressortgroupref;
2540 :
2541 : /*
2542 : * If the GROUP BY tlist entry also appears in ORDER BY, copy operator
2543 : * info from the (first) matching ORDER BY item. This means that if
2544 : * you write something like "GROUP BY foo ORDER BY foo USING <<<", the
2545 : * GROUP BY operation silently takes on the equality semantics implied
2546 : * by the ORDER BY. There are two reasons to do this: it improves the
2547 : * odds that we can implement both GROUP BY and ORDER BY with a single
2548 : * sort step, and it allows the user to choose the equality semantics
2549 : * used by GROUP BY, should she be working with a datatype that has
2550 : * more than one equality operator.
2551 : *
2552 : * If we're in a grouping set, though, we force our requested ordering
2553 : * to be NULLS LAST, because if we have any hope of using a sorted agg
2554 : * for the job, we're going to be tacking on generated NULL values
2555 : * after the corresponding groups. If the user demands nulls first,
2556 : * another sort step is going to be inevitable, but that's the
2557 : * planner's problem.
2558 : */
2559 :
2560 2441 : foreach(sl, sortClause)
2561 : {
2562 2312 : SortGroupClause *sc = (SortGroupClause *) lfirst(sl);
2563 :
2564 2312 : if (sc->tleSortGroupRef == tle->ressortgroupref)
2565 : {
2566 1657 : SortGroupClause *grpc = copyObject(sc);
2567 :
2568 1657 : if (!toplevel)
2569 466 : grpc->nulls_first = false;
2570 1657 : *flatresult = lappend(*flatresult, grpc);
2571 1657 : found = true;
2572 1657 : break;
2573 : }
2574 : }
2575 : }
2576 :
2577 : /*
2578 : * If no match in ORDER BY, just add it to the result using default
2579 : * sort/group semantics.
2580 : */
2581 6131 : if (!found)
2582 4474 : *flatresult = addTargetToGroupList(pstate, tle,
2583 : *flatresult, *targetlist,
2584 : exprLocation(gexpr));
2585 :
2586 : /*
2587 : * _something_ must have assigned us a sortgroupref by now...
2588 : */
2589 :
2590 6131 : return tle->ressortgroupref;
2591 : }
2592 :
2593 : /*
2594 : * Transform a list of expressions within a GROUP BY clause or grouping set.
2595 : *
2596 : * The list of expressions belongs to a single clause within which duplicates
2597 : * can be safely eliminated.
2598 : *
2599 : * Returns an integer list of ressortgroupref values.
2600 : *
2601 : * flatresult reference to flat list of SortGroupClause nodes
2602 : * pstate ParseState
2603 : * list nodes to transform
2604 : * targetlist reference to TargetEntry list
2605 : * sortClause ORDER BY clause (SortGroupClause nodes)
2606 : * exprKind expression kind
2607 : * useSQL99 SQL99 rather than SQL92 syntax
2608 : * toplevel false if within any grouping set
2609 : */
2610 : static List *
2611 212 : transformGroupClauseList(List **flatresult,
2612 : ParseState *pstate, List *list,
2613 : List **targetlist, List *sortClause,
2614 : ParseExprKind exprKind, bool useSQL99, bool toplevel)
2615 : {
2616 212 : Bitmapset *seen_local = NULL;
2617 212 : List *result = NIL;
2618 : ListCell *gl;
2619 :
2620 652 : foreach(gl, list)
2621 : {
2622 440 : Node *gexpr = (Node *) lfirst(gl);
2623 :
2624 440 : Index ref = transformGroupClauseExpr(flatresult,
2625 : seen_local,
2626 : pstate,
2627 : gexpr,
2628 : targetlist,
2629 : sortClause,
2630 : exprKind,
2631 : useSQL99,
2632 : toplevel);
2633 :
2634 440 : if (ref > 0)
2635 : {
2636 432 : seen_local = bms_add_member(seen_local, ref);
2637 432 : result = lappend_int(result, ref);
2638 : }
2639 : }
2640 :
2641 212 : return result;
2642 : }
2643 :
2644 : /*
2645 : * Transform a grouping set and (recursively) its content.
2646 : *
2647 : * The grouping set might be a GROUPING SETS node with other grouping sets
2648 : * inside it, but SETS within SETS have already been flattened out before
2649 : * reaching here.
2650 : *
2651 : * Returns the transformed node, which now contains SIMPLE nodes with lists
2652 : * of ressortgrouprefs rather than expressions.
2653 : *
2654 : * flatresult reference to flat list of SortGroupClause nodes
2655 : * pstate ParseState
2656 : * gset grouping set to transform
2657 : * targetlist reference to TargetEntry list
2658 : * sortClause ORDER BY clause (SortGroupClause nodes)
2659 : * exprKind expression kind
2660 : * useSQL99 SQL99 rather than SQL92 syntax
2661 : * toplevel false if within any grouping set
2662 : */
2663 : static Node *
2664 944 : transformGroupingSet(List **flatresult,
2665 : ParseState *pstate, GroupingSet *gset,
2666 : List **targetlist, List *sortClause,
2667 : ParseExprKind exprKind, bool useSQL99, bool toplevel)
2668 : {
2669 : ListCell *gl;
2670 944 : List *content = NIL;
2671 :
2672 : Assert(toplevel || gset->kind != GROUPING_SET_SETS);
2673 :
2674 2500 : foreach(gl, gset->content)
2675 : {
2676 1556 : Node *n = lfirst(gl);
2677 :
2678 1556 : if (IsA(n, List))
2679 : {
2680 212 : List *l = transformGroupClauseList(flatresult,
2681 : pstate, (List *) n,
2682 : targetlist, sortClause,
2683 : exprKind, useSQL99, false);
2684 :
2685 212 : content = lappend(content, makeGroupingSet(GROUPING_SET_SIMPLE,
2686 : l,
2687 : exprLocation(n)));
2688 : }
2689 1344 : else if (IsA(n, GroupingSet))
2690 : {
2691 214 : GroupingSet *gset2 = (GroupingSet *) lfirst(gl);
2692 :
2693 214 : content = lappend(content, transformGroupingSet(flatresult,
2694 : pstate, gset2,
2695 : targetlist, sortClause,
2696 : exprKind, useSQL99, false));
2697 : }
2698 : else
2699 : {
2700 1130 : Index ref = transformGroupClauseExpr(flatresult,
2701 : NULL,
2702 : pstate,
2703 : n,
2704 : targetlist,
2705 : sortClause,
2706 : exprKind,
2707 : useSQL99,
2708 : false);
2709 :
2710 2260 : content = lappend(content, makeGroupingSet(GROUPING_SET_SIMPLE,
2711 1130 : list_make1_int(ref),
2712 : exprLocation(n)));
2713 : }
2714 : }
2715 :
2716 : /* Arbitrarily cap the size of CUBE, which has exponential growth */
2717 944 : if (gset->kind == GROUPING_SET_CUBE)
2718 : {
2719 122 : if (list_length(content) > 12)
2720 0 : ereport(ERROR,
2721 : (errcode(ERRCODE_TOO_MANY_COLUMNS),
2722 : errmsg("CUBE is limited to 12 elements"),
2723 : parser_errposition(pstate, gset->location)));
2724 : }
2725 :
2726 944 : return (Node *) makeGroupingSet(gset->kind, content, gset->location);
2727 : }
2728 :
2729 :
2730 : /*
2731 : * transformGroupClause -
2732 : * transform a GROUP BY clause
2733 : *
2734 : * GROUP BY items will be added to the targetlist (as resjunk columns)
2735 : * if not already present, so the targetlist must be passed by reference.
2736 : *
2737 : * If GROUP BY ALL is specified, the groupClause will be inferred to be all
2738 : * non-aggregate, non-window expressions in the targetlist.
2739 : *
2740 : * This is also used for window PARTITION BY clauses (which act almost the
2741 : * same, but are always interpreted per SQL99 rules).
2742 : *
2743 : * Grouping sets make this a lot more complex than it was. Our goal here is
2744 : * twofold: we make a flat list of SortGroupClause nodes referencing each
2745 : * distinct expression used for grouping, with those expressions added to the
2746 : * targetlist if needed. At the same time, we build the groupingSets tree,
2747 : * which stores only ressortgrouprefs as integer lists inside GroupingSet nodes
2748 : * (possibly nested, but limited in depth: a GROUPING_SET_SETS node can contain
2749 : * nested SIMPLE, CUBE or ROLLUP nodes, but not more sets - we flatten that
2750 : * out; while CUBE and ROLLUP can contain only SIMPLE nodes).
2751 : *
2752 : * We skip much of the hard work if there are no grouping sets.
2753 : *
2754 : * One subtlety is that the groupClause list can end up empty while the
2755 : * groupingSets list is not; this happens if there are only empty grouping
2756 : * sets, or an explicit GROUP BY (). This has the same effect as specifying
2757 : * aggregates or a HAVING clause with no GROUP BY; the output is one row per
2758 : * grouping set even if the input is empty.
2759 : *
2760 : * Returns the transformed (flat) groupClause.
2761 : *
2762 : * pstate ParseState
2763 : * grouplist clause to transform
2764 : * groupByAll is this a GROUP BY ALL statement?
2765 : * groupingSets reference to list to contain the grouping set tree
2766 : * targetlist reference to TargetEntry list
2767 : * sortClause ORDER BY clause (SortGroupClause nodes)
2768 : * exprKind expression kind
2769 : * useSQL99 SQL99 rather than SQL92 syntax
2770 : */
2771 : List *
2772 299945 : transformGroupClause(ParseState *pstate, List *grouplist, bool groupByAll,
2773 : List **groupingSets,
2774 : List **targetlist, List *sortClause,
2775 : ParseExprKind exprKind, bool useSQL99)
2776 : {
2777 299945 : List *result = NIL;
2778 : List *flat_grouplist;
2779 299945 : List *gsets = NIL;
2780 : ListCell *gl;
2781 299945 : bool hasGroupingSets = false;
2782 299945 : Bitmapset *seen_local = NULL;
2783 :
2784 : /* Handle GROUP BY ALL */
2785 299945 : if (groupByAll)
2786 : {
2787 : /* There cannot have been any explicit grouplist items */
2788 : Assert(grouplist == NIL);
2789 :
2790 : /* Iterate over targets, adding acceptable ones to the result list */
2791 176 : foreach_ptr(TargetEntry, tle, *targetlist)
2792 : {
2793 : /* Ignore junk TLEs */
2794 88 : if (tle->resjunk)
2795 0 : continue;
2796 :
2797 : /*
2798 : * TLEs containing aggregates are not okay to add to GROUP BY
2799 : * (compare checkTargetlistEntrySQL92). But the SQL standard
2800 : * directs us to skip them, so it's fine.
2801 : */
2802 164 : if (pstate->p_hasAggs &&
2803 76 : contain_aggs_of_level((Node *) tle->expr, 0))
2804 36 : continue;
2805 :
2806 : /*
2807 : * Likewise, TLEs containing window functions are not okay to add
2808 : * to GROUP BY. At this writing, the SQL standard is silent on
2809 : * what to do with them, but by analogy to aggregates we'll just
2810 : * skip them.
2811 : */
2812 60 : if (pstate->p_hasWindowFuncs &&
2813 8 : contain_windowfuncs((Node *) tle->expr))
2814 4 : continue;
2815 :
2816 : /*
2817 : * Otherwise, add the TLE to the result using default sort/group
2818 : * semantics. We specify the parse location as the TLE's
2819 : * location, despite the comment for addTargetToGroupList
2820 : * discouraging that. The only other thing we could point to is
2821 : * the ALL keyword, which seems unhelpful when there are multiple
2822 : * TLEs.
2823 : */
2824 48 : result = addTargetToGroupList(pstate, tle,
2825 : result, *targetlist,
2826 48 : exprLocation((Node *) tle->expr));
2827 : }
2828 :
2829 : /* If we found any acceptable targets, we're done */
2830 44 : if (result != NIL)
2831 36 : return result;
2832 :
2833 : /*
2834 : * Otherwise, the SQL standard says to treat it like "GROUP BY ()".
2835 : * Build a representation of that, and let the rest of this function
2836 : * handle it.
2837 : */
2838 8 : grouplist = list_make1(makeGroupingSet(GROUPING_SET_EMPTY, NIL, -1));
2839 : }
2840 :
2841 : /*
2842 : * Recursively flatten implicit RowExprs. (Technically this is only needed
2843 : * for GROUP BY, per the syntax rules for grouping sets, but we do it
2844 : * anyway.)
2845 : */
2846 299909 : flat_grouplist = (List *) flatten_grouping_sets((Node *) grouplist,
2847 : true,
2848 : &hasGroupingSets);
2849 :
2850 : /*
2851 : * If the list is now empty, but hasGroupingSets is true, it's because we
2852 : * elided redundant empty grouping sets. Restore a single empty grouping
2853 : * set to leave a canonical form: GROUP BY ()
2854 : */
2855 :
2856 299909 : if (flat_grouplist == NIL && hasGroupingSets)
2857 : {
2858 24 : flat_grouplist = list_make1(makeGroupingSet(GROUPING_SET_EMPTY,
2859 : NIL,
2860 : exprLocation((Node *) grouplist)));
2861 : }
2862 :
2863 305414 : foreach(gl, flat_grouplist)
2864 : {
2865 5521 : Node *gexpr = (Node *) lfirst(gl);
2866 :
2867 5521 : if (IsA(gexpr, GroupingSet))
2868 : {
2869 754 : GroupingSet *gset = (GroupingSet *) gexpr;
2870 :
2871 754 : switch (gset->kind)
2872 : {
2873 24 : case GROUPING_SET_EMPTY:
2874 24 : gsets = lappend(gsets, gset);
2875 24 : break;
2876 0 : case GROUPING_SET_SIMPLE:
2877 : /* can't happen */
2878 : Assert(false);
2879 0 : break;
2880 730 : case GROUPING_SET_SETS:
2881 : case GROUPING_SET_CUBE:
2882 : case GROUPING_SET_ROLLUP:
2883 730 : gsets = lappend(gsets,
2884 730 : transformGroupingSet(&result,
2885 : pstate, gset,
2886 : targetlist, sortClause,
2887 : exprKind, useSQL99, true));
2888 730 : break;
2889 : }
2890 : }
2891 : else
2892 : {
2893 4767 : Index ref = transformGroupClauseExpr(&result, seen_local,
2894 : pstate, gexpr,
2895 : targetlist, sortClause,
2896 : exprKind, useSQL99, true);
2897 :
2898 4751 : if (ref > 0)
2899 : {
2900 4743 : seen_local = bms_add_member(seen_local, ref);
2901 4743 : if (hasGroupingSets)
2902 32 : gsets = lappend(gsets,
2903 64 : makeGroupingSet(GROUPING_SET_SIMPLE,
2904 32 : list_make1_int(ref),
2905 : exprLocation(gexpr)));
2906 : }
2907 : }
2908 : }
2909 :
2910 : /* parser should prevent this */
2911 : Assert(gsets == NIL || groupingSets != NULL);
2912 :
2913 299893 : if (groupingSets)
2914 297849 : *groupingSets = gsets;
2915 :
2916 299893 : return result;
2917 : }
2918 :
2919 : /*
2920 : * transformSortClause -
2921 : * transform an ORDER BY clause
2922 : *
2923 : * ORDER BY items will be added to the targetlist (as resjunk columns)
2924 : * if not already present, so the targetlist must be passed by reference.
2925 : *
2926 : * This is also used for window and aggregate ORDER BY clauses (which act
2927 : * almost the same, but are always interpreted per SQL99 rules).
2928 : */
2929 : List *
2930 344111 : transformSortClause(ParseState *pstate,
2931 : List *orderlist,
2932 : List **targetlist,
2933 : ParseExprKind exprKind,
2934 : bool useSQL99)
2935 : {
2936 344111 : List *sortlist = NIL;
2937 : ListCell *olitem;
2938 :
2939 418265 : foreach(olitem, orderlist)
2940 : {
2941 74182 : SortBy *sortby = (SortBy *) lfirst(olitem);
2942 : TargetEntry *tle;
2943 :
2944 74182 : if (useSQL99)
2945 3261 : tle = findTargetlistEntrySQL99(pstate, sortby->node,
2946 : targetlist, exprKind);
2947 : else
2948 70921 : tle = findTargetlistEntrySQL92(pstate, sortby->node,
2949 : targetlist, exprKind);
2950 :
2951 74158 : sortlist = addTargetToSortList(pstate, tle,
2952 : sortlist, *targetlist, sortby);
2953 : }
2954 :
2955 344083 : return sortlist;
2956 : }
2957 :
2958 : /*
2959 : * transformWindowDefinitions -
2960 : * transform window definitions (WindowDef to WindowClause)
2961 : */
2962 : List *
2963 297869 : transformWindowDefinitions(ParseState *pstate,
2964 : List *windowdefs,
2965 : List **targetlist)
2966 : {
2967 297869 : List *result = NIL;
2968 297869 : Index winref = 0;
2969 : ListCell *lc;
2970 :
2971 299877 : foreach(lc, windowdefs)
2972 : {
2973 2052 : WindowDef *windef = (WindowDef *) lfirst(lc);
2974 2052 : WindowClause *refwc = NULL;
2975 : List *partitionClause;
2976 : List *orderClause;
2977 2052 : Oid rangeopfamily = InvalidOid;
2978 2052 : Oid rangeopcintype = InvalidOid;
2979 : WindowClause *wc;
2980 :
2981 2052 : winref++;
2982 :
2983 : /*
2984 : * Check for duplicate window names.
2985 : */
2986 2478 : if (windef->name &&
2987 426 : findWindowClause(result, windef->name) != NULL)
2988 4 : ereport(ERROR,
2989 : (errcode(ERRCODE_WINDOWING_ERROR),
2990 : errmsg("window \"%s\" is already defined", windef->name),
2991 : parser_errposition(pstate, windef->location)));
2992 :
2993 : /*
2994 : * If it references a previous window, look that up.
2995 : */
2996 2048 : if (windef->refname)
2997 : {
2998 28 : refwc = findWindowClause(result, windef->refname);
2999 28 : if (refwc == NULL)
3000 0 : ereport(ERROR,
3001 : (errcode(ERRCODE_UNDEFINED_OBJECT),
3002 : errmsg("window \"%s\" does not exist",
3003 : windef->refname),
3004 : parser_errposition(pstate, windef->location)));
3005 : }
3006 :
3007 : /*
3008 : * Transform PARTITION and ORDER specs, if any. These are treated
3009 : * almost exactly like top-level GROUP BY and ORDER BY clauses,
3010 : * including the special handling of nondefault operator semantics.
3011 : */
3012 2048 : orderClause = transformSortClause(pstate,
3013 : windef->orderClause,
3014 : targetlist,
3015 : EXPR_KIND_WINDOW_ORDER,
3016 : true /* force SQL99 rules */ );
3017 2044 : partitionClause = transformGroupClause(pstate,
3018 : windef->partitionClause,
3019 : false /* not GROUP BY ALL */ ,
3020 : NULL,
3021 : targetlist,
3022 : orderClause,
3023 : EXPR_KIND_WINDOW_PARTITION,
3024 : true /* force SQL99 rules */ );
3025 :
3026 : /*
3027 : * And prepare the new WindowClause.
3028 : */
3029 2044 : wc = makeNode(WindowClause);
3030 2044 : wc->name = windef->name;
3031 2044 : wc->refname = windef->refname;
3032 :
3033 : /*
3034 : * Per spec, a windowdef that references a previous one copies the
3035 : * previous partition clause (and mustn't specify its own). It can
3036 : * specify its own ordering clause, but only if the previous one had
3037 : * none. It always specifies its own frame clause, and the previous
3038 : * one must not have a frame clause. Yeah, it's bizarre that each of
3039 : * these cases works differently, but SQL:2008 says so; see 7.11
3040 : * <window clause> syntax rule 10 and general rule 1. The frame
3041 : * clause rule is especially bizarre because it makes "OVER foo"
3042 : * different from "OVER (foo)", and requires the latter to throw an
3043 : * error if foo has a nondefault frame clause. Well, ours not to
3044 : * reason why, but we do go out of our way to throw a useful error
3045 : * message for such cases.
3046 : */
3047 2044 : if (refwc)
3048 : {
3049 28 : if (partitionClause)
3050 0 : ereport(ERROR,
3051 : (errcode(ERRCODE_WINDOWING_ERROR),
3052 : errmsg("cannot override PARTITION BY clause of window \"%s\"",
3053 : windef->refname),
3054 : parser_errposition(pstate, windef->location)));
3055 28 : wc->partitionClause = copyObject(refwc->partitionClause);
3056 : }
3057 : else
3058 2016 : wc->partitionClause = partitionClause;
3059 2044 : if (refwc)
3060 : {
3061 28 : if (orderClause && refwc->orderClause)
3062 0 : ereport(ERROR,
3063 : (errcode(ERRCODE_WINDOWING_ERROR),
3064 : errmsg("cannot override ORDER BY clause of window \"%s\"",
3065 : windef->refname),
3066 : parser_errposition(pstate, windef->location)));
3067 28 : if (orderClause)
3068 : {
3069 12 : wc->orderClause = orderClause;
3070 12 : wc->copiedOrder = false;
3071 : }
3072 : else
3073 : {
3074 16 : wc->orderClause = copyObject(refwc->orderClause);
3075 16 : wc->copiedOrder = true;
3076 : }
3077 : }
3078 : else
3079 : {
3080 2016 : wc->orderClause = orderClause;
3081 2016 : wc->copiedOrder = false;
3082 : }
3083 2044 : if (refwc && refwc->frameOptions != FRAMEOPTION_DEFAULTS)
3084 : {
3085 : /*
3086 : * Use this message if this is a WINDOW clause, or if it's an OVER
3087 : * clause that includes ORDER BY or framing clauses. (We already
3088 : * rejected PARTITION BY above, so no need to check that.)
3089 : */
3090 0 : if (windef->name ||
3091 0 : orderClause || windef->frameOptions != FRAMEOPTION_DEFAULTS)
3092 0 : ereport(ERROR,
3093 : (errcode(ERRCODE_WINDOWING_ERROR),
3094 : errmsg("cannot copy window \"%s\" because it has a frame clause",
3095 : windef->refname),
3096 : parser_errposition(pstate, windef->location)));
3097 : /* Else this clause is just OVER (foo), so say this: */
3098 0 : ereport(ERROR,
3099 : (errcode(ERRCODE_WINDOWING_ERROR),
3100 : errmsg("cannot copy window \"%s\" because it has a frame clause",
3101 : windef->refname),
3102 : errhint("Omit the parentheses in this OVER clause."),
3103 : parser_errposition(pstate, windef->location)));
3104 : }
3105 2044 : wc->frameOptions = windef->frameOptions;
3106 :
3107 : /*
3108 : * RANGE offset PRECEDING/FOLLOWING requires exactly one ORDER BY
3109 : * column; check that and get its sort opfamily info.
3110 : */
3111 2044 : if ((wc->frameOptions & FRAMEOPTION_RANGE) &&
3112 1478 : (wc->frameOptions & (FRAMEOPTION_START_OFFSET |
3113 : FRAMEOPTION_END_OFFSET)))
3114 : {
3115 : SortGroupClause *sortcl;
3116 : Node *sortkey;
3117 : CompareType rangecmptype;
3118 :
3119 424 : if (list_length(wc->orderClause) != 1)
3120 12 : ereport(ERROR,
3121 : (errcode(ERRCODE_WINDOWING_ERROR),
3122 : errmsg("RANGE with offset PRECEDING/FOLLOWING requires exactly one ORDER BY column"),
3123 : parser_errposition(pstate, windef->location)));
3124 412 : sortcl = linitial_node(SortGroupClause, wc->orderClause);
3125 412 : sortkey = get_sortgroupclause_expr(sortcl, *targetlist);
3126 : /* Find the sort operator in pg_amop */
3127 412 : if (!get_ordering_op_properties(sortcl->sortop,
3128 : &rangeopfamily,
3129 : &rangeopcintype,
3130 : &rangecmptype))
3131 0 : elog(ERROR, "operator %u is not a valid ordering operator",
3132 : sortcl->sortop);
3133 : /* Record properties of sort ordering */
3134 412 : wc->inRangeColl = exprCollation(sortkey);
3135 412 : wc->inRangeAsc = !sortcl->reverse_sort;
3136 412 : wc->inRangeNullsFirst = sortcl->nulls_first;
3137 : }
3138 :
3139 : /* Per spec, GROUPS mode requires an ORDER BY clause */
3140 2032 : if (wc->frameOptions & FRAMEOPTION_GROUPS)
3141 : {
3142 116 : if (wc->orderClause == NIL)
3143 4 : ereport(ERROR,
3144 : (errcode(ERRCODE_WINDOWING_ERROR),
3145 : errmsg("GROUPS mode requires an ORDER BY clause"),
3146 : parser_errposition(pstate, windef->location)));
3147 : }
3148 :
3149 : /* Process frame offset expressions */
3150 2028 : wc->startOffset = transformFrameOffset(pstate, wc->frameOptions,
3151 : rangeopfamily, rangeopcintype,
3152 : &wc->startInRangeFunc,
3153 : windef->startOffset);
3154 2012 : wc->endOffset = transformFrameOffset(pstate, wc->frameOptions,
3155 : rangeopfamily, rangeopcintype,
3156 : &wc->endInRangeFunc,
3157 : windef->endOffset);
3158 2008 : wc->winref = winref;
3159 :
3160 2008 : result = lappend(result, wc);
3161 : }
3162 :
3163 297825 : return result;
3164 : }
3165 :
3166 : /*
3167 : * transformDistinctClause -
3168 : * transform a DISTINCT clause
3169 : *
3170 : * Since we may need to add items to the query's targetlist, that list
3171 : * is passed by reference.
3172 : *
3173 : * As with GROUP BY, we absorb the sorting semantics of ORDER BY as much as
3174 : * possible into the distinctClause. This avoids a possible need to re-sort,
3175 : * and allows the user to choose the equality semantics used by DISTINCT,
3176 : * should she be working with a datatype that has more than one equality
3177 : * operator.
3178 : *
3179 : * is_agg is true if we are transforming an aggregate(DISTINCT ...)
3180 : * function call. This does not affect any behavior, only the phrasing
3181 : * of error messages.
3182 : */
3183 : List *
3184 2588 : transformDistinctClause(ParseState *pstate,
3185 : List **targetlist, List *sortClause, bool is_agg)
3186 : {
3187 2588 : List *result = NIL;
3188 : ListCell *slitem;
3189 : ListCell *tlitem;
3190 :
3191 : /*
3192 : * The distinctClause should consist of all ORDER BY items followed by all
3193 : * other non-resjunk targetlist items. There must not be any resjunk
3194 : * ORDER BY items --- that would imply that we are sorting by a value that
3195 : * isn't necessarily unique within a DISTINCT group, so the results
3196 : * wouldn't be well-defined. This construction ensures we follow the rule
3197 : * that sortClause and distinctClause match; in fact the sortClause will
3198 : * always be a prefix of distinctClause.
3199 : *
3200 : * Note a corner case: the same TLE could be in the ORDER BY list multiple
3201 : * times with different sortops. We have to include it in the
3202 : * distinctClause the same way to preserve the prefix property. The net
3203 : * effect will be that the TLE value will be made unique according to both
3204 : * sortops.
3205 : */
3206 3003 : foreach(slitem, sortClause)
3207 : {
3208 439 : SortGroupClause *scl = (SortGroupClause *) lfirst(slitem);
3209 439 : TargetEntry *tle = get_sortgroupclause_tle(scl, *targetlist);
3210 :
3211 439 : if (tle->resjunk)
3212 24 : ereport(ERROR,
3213 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3214 : is_agg ?
3215 : errmsg("in an aggregate with DISTINCT, ORDER BY expressions must appear in argument list") :
3216 : errmsg("for SELECT DISTINCT, ORDER BY expressions must appear in select list"),
3217 : parser_errposition(pstate,
3218 : exprLocation((Node *) tle->expr))));
3219 415 : result = lappend(result, copyObject(scl));
3220 : }
3221 :
3222 : /*
3223 : * Now add any remaining non-resjunk tlist items, using default sort/group
3224 : * semantics for their data types.
3225 : */
3226 10214 : foreach(tlitem, *targetlist)
3227 : {
3228 7650 : TargetEntry *tle = (TargetEntry *) lfirst(tlitem);
3229 :
3230 7650 : if (tle->resjunk)
3231 2 : continue; /* ignore junk */
3232 7648 : result = addTargetToGroupList(pstate, tle,
3233 : result, *targetlist,
3234 7648 : exprLocation((Node *) tle->expr));
3235 : }
3236 :
3237 : /*
3238 : * Complain if we found nothing to make DISTINCT. Returning an empty list
3239 : * would cause the parsed Query to look like it didn't have DISTINCT, with
3240 : * results that would probably surprise the user. Note: this case is
3241 : * presently impossible for aggregates because of grammar restrictions,
3242 : * but we check anyway.
3243 : */
3244 2564 : if (result == NIL)
3245 0 : ereport(ERROR,
3246 : (errcode(ERRCODE_SYNTAX_ERROR),
3247 : is_agg ?
3248 : errmsg("an aggregate with DISTINCT must have at least one argument") :
3249 : errmsg("SELECT DISTINCT must have at least one column")));
3250 :
3251 2564 : return result;
3252 : }
3253 :
3254 : /*
3255 : * transformDistinctOnClause -
3256 : * transform a DISTINCT ON clause
3257 : *
3258 : * Since we may need to add items to the query's targetlist, that list
3259 : * is passed by reference.
3260 : *
3261 : * As with GROUP BY, we absorb the sorting semantics of ORDER BY as much as
3262 : * possible into the distinctClause. This avoids a possible need to re-sort,
3263 : * and allows the user to choose the equality semantics used by DISTINCT,
3264 : * should she be working with a datatype that has more than one equality
3265 : * operator.
3266 : */
3267 : List *
3268 159 : transformDistinctOnClause(ParseState *pstate, List *distinctlist,
3269 : List **targetlist, List *sortClause)
3270 : {
3271 159 : List *result = NIL;
3272 159 : List *sortgrouprefs = NIL;
3273 : bool skipped_sortitem;
3274 : ListCell *lc;
3275 : ListCell *lc2;
3276 :
3277 : /*
3278 : * Add all the DISTINCT ON expressions to the tlist (if not already
3279 : * present, they are added as resjunk items). Assign sortgroupref numbers
3280 : * to them, and make a list of these numbers. (NB: we rely below on the
3281 : * sortgrouprefs list being one-for-one with the original distinctlist.
3282 : * Also notice that we could have duplicate DISTINCT ON expressions and
3283 : * hence duplicate entries in sortgrouprefs.)
3284 : */
3285 398 : foreach(lc, distinctlist)
3286 : {
3287 243 : Node *dexpr = (Node *) lfirst(lc);
3288 : int sortgroupref;
3289 : TargetEntry *tle;
3290 :
3291 243 : tle = findTargetlistEntrySQL92(pstate, dexpr, targetlist,
3292 : EXPR_KIND_DISTINCT_ON);
3293 239 : sortgroupref = assignSortGroupRef(tle, *targetlist);
3294 239 : sortgrouprefs = lappend_int(sortgrouprefs, sortgroupref);
3295 : }
3296 :
3297 : /*
3298 : * If the user writes both DISTINCT ON and ORDER BY, adopt the sorting
3299 : * semantics from ORDER BY items that match DISTINCT ON items, and also
3300 : * adopt their column sort order. We insist that the distinctClause and
3301 : * sortClause match, so throw error if we find the need to add any more
3302 : * distinctClause items after we've skipped an ORDER BY item that wasn't
3303 : * in DISTINCT ON.
3304 : */
3305 155 : skipped_sortitem = false;
3306 358 : foreach(lc, sortClause)
3307 : {
3308 207 : SortGroupClause *scl = (SortGroupClause *) lfirst(lc);
3309 :
3310 207 : if (list_member_int(sortgrouprefs, scl->tleSortGroupRef))
3311 : {
3312 159 : if (skipped_sortitem)
3313 4 : ereport(ERROR,
3314 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3315 : errmsg("SELECT DISTINCT ON expressions must match initial ORDER BY expressions"),
3316 : parser_errposition(pstate,
3317 : get_matching_location(scl->tleSortGroupRef,
3318 : sortgrouprefs,
3319 : distinctlist))));
3320 : else
3321 155 : result = lappend(result, copyObject(scl));
3322 : }
3323 : else
3324 48 : skipped_sortitem = true;
3325 : }
3326 :
3327 : /*
3328 : * Now add any remaining DISTINCT ON items, using default sort/group
3329 : * semantics for their data types. (Note: this is pretty questionable; if
3330 : * the ORDER BY list doesn't include all the DISTINCT ON items and more
3331 : * besides, you certainly aren't using DISTINCT ON in the intended way,
3332 : * and you probably aren't going to get consistent results. It might be
3333 : * better to throw an error or warning here. But historically we've
3334 : * allowed it, so keep doing so.)
3335 : */
3336 382 : forboth(lc, distinctlist, lc2, sortgrouprefs)
3337 : {
3338 231 : Node *dexpr = (Node *) lfirst(lc);
3339 231 : int sortgroupref = lfirst_int(lc2);
3340 231 : TargetEntry *tle = get_sortgroupref_tle(sortgroupref, *targetlist);
3341 :
3342 231 : if (targetIsInSortList(tle, InvalidOid, result))
3343 151 : continue; /* already in list (with some semantics) */
3344 80 : if (skipped_sortitem)
3345 0 : ereport(ERROR,
3346 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3347 : errmsg("SELECT DISTINCT ON expressions must match initial ORDER BY expressions"),
3348 : parser_errposition(pstate, exprLocation(dexpr))));
3349 80 : result = addTargetToGroupList(pstate, tle,
3350 : result, *targetlist,
3351 : exprLocation(dexpr));
3352 : }
3353 :
3354 : /*
3355 : * An empty result list is impossible here because of grammar
3356 : * restrictions.
3357 : */
3358 : Assert(result != NIL);
3359 :
3360 151 : return result;
3361 : }
3362 :
3363 : /*
3364 : * get_matching_location
3365 : * Get the exprLocation of the exprs member corresponding to the
3366 : * (first) member of sortgrouprefs that equals sortgroupref.
3367 : *
3368 : * This is used so that we can point at a troublesome DISTINCT ON entry.
3369 : * (Note that we need to use the original untransformed DISTINCT ON list
3370 : * item, as whatever TLE it corresponds to will very possibly have a
3371 : * parse location pointing to some matching entry in the SELECT list
3372 : * or ORDER BY list.)
3373 : */
3374 : static int
3375 4 : get_matching_location(int sortgroupref, List *sortgrouprefs, List *exprs)
3376 : {
3377 : ListCell *lcs;
3378 : ListCell *lce;
3379 :
3380 8 : forboth(lcs, sortgrouprefs, lce, exprs)
3381 : {
3382 8 : if (lfirst_int(lcs) == sortgroupref)
3383 4 : return exprLocation((Node *) lfirst(lce));
3384 : }
3385 : /* if no match, caller blew it */
3386 0 : elog(ERROR, "get_matching_location: no matching sortgroupref");
3387 : return -1; /* keep compiler quiet */
3388 : }
3389 :
3390 : /*
3391 : * resolve_unique_index_expr
3392 : * Infer a unique index from a list of indexElems, for ON
3393 : * CONFLICT clause
3394 : *
3395 : * Perform parse analysis of expressions and columns appearing within ON
3396 : * CONFLICT clause. During planning, the returned list of expressions is used
3397 : * to infer which unique index to use.
3398 : */
3399 : static List *
3400 1246 : resolve_unique_index_expr(ParseState *pstate, InferClause *infer,
3401 : Relation heapRel)
3402 : {
3403 1246 : List *result = NIL;
3404 : ListCell *l;
3405 :
3406 2782 : foreach(l, infer->indexElems)
3407 : {
3408 1552 : IndexElem *ielem = (IndexElem *) lfirst(l);
3409 1552 : InferenceElem *pInfer = makeNode(InferenceElem);
3410 : Node *parse;
3411 :
3412 : /*
3413 : * Raw grammar re-uses CREATE INDEX infrastructure for unique index
3414 : * inference clause, and so will accept opclasses by name and so on.
3415 : *
3416 : * Make no attempt to match ASC or DESC ordering, NULLS FIRST/NULLS
3417 : * LAST ordering or opclass options, since those are not significant
3418 : * for inference purposes (any unique index matching the inference
3419 : * specification in other regards is accepted indifferently). Actively
3420 : * reject this as wrong-headed.
3421 : */
3422 1552 : if (ielem->ordering != SORTBY_DEFAULT)
3423 4 : ereport(ERROR,
3424 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3425 : errmsg("%s is not allowed in ON CONFLICT clause",
3426 : "ASC/DESC"),
3427 : parser_errposition(pstate, ielem->location)));
3428 1548 : if (ielem->nulls_ordering != SORTBY_NULLS_DEFAULT)
3429 4 : ereport(ERROR,
3430 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3431 : errmsg("%s is not allowed in ON CONFLICT clause",
3432 : "NULLS FIRST/LAST"),
3433 : parser_errposition(pstate, ielem->location)));
3434 1544 : if (ielem->opclassopts)
3435 4 : ereport(ERROR,
3436 : errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3437 : errmsg("operator class options are not allowed in ON CONFLICT clause"),
3438 : parser_errposition(pstate, ielem->location));
3439 :
3440 1540 : if (!ielem->expr)
3441 : {
3442 : /* Simple index attribute */
3443 : ColumnRef *n;
3444 :
3445 : /*
3446 : * Grammar won't have built raw expression for us in event of
3447 : * plain column reference. Create one directly, and perform
3448 : * expression transformation. Planner expects this, and performs
3449 : * its own normalization for the purposes of matching against
3450 : * pg_index.
3451 : */
3452 1426 : n = makeNode(ColumnRef);
3453 1426 : n->fields = list_make1(makeString(ielem->name));
3454 : /* Location is approximately that of inference specification */
3455 1426 : n->location = infer->location;
3456 1426 : parse = (Node *) n;
3457 : }
3458 : else
3459 : {
3460 : /* Do parse transformation of the raw expression */
3461 114 : parse = (Node *) ielem->expr;
3462 : }
3463 :
3464 : /*
3465 : * transformExpr() will reject subqueries, aggregates, window
3466 : * functions, and SRFs, based on being passed
3467 : * EXPR_KIND_INDEX_EXPRESSION. So we needn't worry about those
3468 : * further ... not that they would match any available index
3469 : * expression anyway.
3470 : */
3471 1540 : pInfer->expr = transformExpr(pstate, parse, EXPR_KIND_INDEX_EXPRESSION);
3472 :
3473 : /* Perform lookup of collation and operator class as required */
3474 1536 : if (!ielem->collation)
3475 1508 : pInfer->infercollid = InvalidOid;
3476 : else
3477 28 : pInfer->infercollid = LookupCollation(pstate, ielem->collation,
3478 : ielem->location);
3479 :
3480 1536 : if (!ielem->opclass)
3481 1508 : pInfer->inferopclass = InvalidOid;
3482 : else
3483 28 : pInfer->inferopclass = get_opclass_oid(BTREE_AM_OID,
3484 : ielem->opclass, false);
3485 :
3486 1536 : result = lappend(result, pInfer);
3487 : }
3488 :
3489 1230 : return result;
3490 : }
3491 :
3492 : /*
3493 : * transformOnConflictArbiter -
3494 : * transform arbiter expressions in an ON CONFLICT clause.
3495 : *
3496 : * Transformed expressions used to infer one unique index relation to serve as
3497 : * an ON CONFLICT arbiter. Partial unique indexes may be inferred using WHERE
3498 : * clause from inference specification clause.
3499 : */
3500 : void
3501 1536 : transformOnConflictArbiter(ParseState *pstate,
3502 : OnConflictClause *onConflictClause,
3503 : List **arbiterExpr, Node **arbiterWhere,
3504 : Oid *constraint)
3505 : {
3506 1536 : InferClause *infer = onConflictClause->infer;
3507 :
3508 1536 : *arbiterExpr = NIL;
3509 1536 : *arbiterWhere = NULL;
3510 1536 : *constraint = InvalidOid;
3511 :
3512 1536 : if ((onConflictClause->action == ONCONFLICT_UPDATE ||
3513 1536 : onConflictClause->action == ONCONFLICT_SELECT) && !infer)
3514 4 : ereport(ERROR,
3515 : errcode(ERRCODE_SYNTAX_ERROR),
3516 : errmsg("ON CONFLICT DO %s requires inference specification or constraint name",
3517 : onConflictClause->action == ONCONFLICT_UPDATE ? "UPDATE" : "SELECT"),
3518 : errhint("For example, ON CONFLICT (column_name)."),
3519 : parser_errposition(pstate,
3520 : exprLocation((Node *) onConflictClause)));
3521 :
3522 : /*
3523 : * To simplify certain aspects of its design, speculative insertion into
3524 : * system catalogs is disallowed
3525 : */
3526 1532 : if (IsCatalogRelation(pstate->p_target_relation))
3527 0 : ereport(ERROR,
3528 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3529 : errmsg("ON CONFLICT is not supported with system catalog tables"),
3530 : parser_errposition(pstate,
3531 : exprLocation((Node *) onConflictClause))));
3532 :
3533 : /* Same applies to table used by logical decoding as catalog table */
3534 1532 : if (RelationIsUsedAsCatalogTable(pstate->p_target_relation))
3535 0 : ereport(ERROR,
3536 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3537 : errmsg("ON CONFLICT is not supported on table \"%s\" used as a catalog table",
3538 : RelationGetRelationName(pstate->p_target_relation)),
3539 : parser_errposition(pstate,
3540 : exprLocation((Node *) onConflictClause))));
3541 :
3542 : /* ON CONFLICT DO NOTHING does not require an inference clause */
3543 1532 : if (infer)
3544 : {
3545 1384 : if (infer->indexElems)
3546 1246 : *arbiterExpr = resolve_unique_index_expr(pstate, infer,
3547 : pstate->p_target_relation);
3548 :
3549 : /*
3550 : * Handling inference WHERE clause (for partial unique index
3551 : * inference)
3552 : */
3553 1368 : if (infer->whereClause)
3554 34 : *arbiterWhere = transformExpr(pstate, infer->whereClause,
3555 : EXPR_KIND_INDEX_PREDICATE);
3556 :
3557 : /*
3558 : * If the arbiter is specified by constraint name, get the constraint
3559 : * OID and mark the constrained columns as requiring SELECT privilege,
3560 : * in the same way as would have happened if the arbiter had been
3561 : * specified by explicit reference to the constraint's index columns.
3562 : */
3563 1368 : if (infer->conname)
3564 : {
3565 138 : Oid relid = RelationGetRelid(pstate->p_target_relation);
3566 138 : RTEPermissionInfo *perminfo = pstate->p_target_nsitem->p_perminfo;
3567 : Bitmapset *conattnos;
3568 :
3569 138 : conattnos = get_relation_constraint_attnos(relid, infer->conname,
3570 : false, constraint);
3571 :
3572 : /* Make sure the rel as a whole is marked for SELECT access */
3573 138 : perminfo->requiredPerms |= ACL_SELECT;
3574 : /* Mark the constrained columns as requiring SELECT access */
3575 138 : perminfo->selectedCols = bms_add_members(perminfo->selectedCols,
3576 : conattnos);
3577 : }
3578 : }
3579 :
3580 : /*
3581 : * It's convenient to form a list of expressions based on the
3582 : * representation used by CREATE INDEX, since the same restrictions are
3583 : * appropriate (e.g. on subqueries). However, from here on, a dedicated
3584 : * primnode representation is used for inference elements, and so
3585 : * assign_query_collations() can be trusted to do the right thing with the
3586 : * post parse analysis query tree inference clause representation.
3587 : */
3588 1516 : }
3589 :
3590 : /*
3591 : * addTargetToSortList
3592 : * If the given targetlist entry isn't already in the SortGroupClause
3593 : * list, add it to the end of the list, using the given sort ordering
3594 : * info.
3595 : *
3596 : * Returns the updated SortGroupClause list.
3597 : */
3598 : List *
3599 74401 : addTargetToSortList(ParseState *pstate, TargetEntry *tle,
3600 : List *sortlist, List *targetlist, SortBy *sortby)
3601 : {
3602 74401 : Oid restype = exprType((Node *) tle->expr);
3603 : Oid sortop;
3604 : Oid eqop;
3605 : bool hashable;
3606 : bool reverse;
3607 : int location;
3608 : ParseCallbackState pcbstate;
3609 :
3610 : /* if tlist item is an UNKNOWN literal, change it to TEXT */
3611 74401 : if (restype == UNKNOWNOID)
3612 : {
3613 8 : tle->expr = (Expr *) coerce_type(pstate, (Node *) tle->expr,
3614 : restype, TEXTOID, -1,
3615 : COERCION_IMPLICIT,
3616 : COERCE_IMPLICIT_CAST,
3617 : -1);
3618 8 : restype = TEXTOID;
3619 : }
3620 :
3621 : /*
3622 : * Rather than clutter the API of get_sort_group_operators and the other
3623 : * functions we're about to use, make use of error context callback to
3624 : * mark any error reports with a parse position. We point to the operator
3625 : * location if present, else to the expression being sorted. (NB: use the
3626 : * original untransformed expression here; the TLE entry might well point
3627 : * at a duplicate expression in the regular SELECT list.)
3628 : */
3629 74401 : location = sortby->location;
3630 74401 : if (location < 0)
3631 74259 : location = exprLocation(sortby->node);
3632 74401 : setup_parser_errposition_callback(&pcbstate, pstate, location);
3633 :
3634 : /* determine the sortop, eqop, and directionality */
3635 74401 : switch (sortby->sortby_dir)
3636 : {
3637 71916 : case SORTBY_DEFAULT:
3638 : case SORTBY_ASC:
3639 71916 : get_sort_group_operators(restype,
3640 : true, true, false,
3641 : &sortop, &eqop, NULL,
3642 : &hashable);
3643 71912 : reverse = false;
3644 71912 : break;
3645 2343 : case SORTBY_DESC:
3646 2343 : get_sort_group_operators(restype,
3647 : false, true, true,
3648 : NULL, &eqop, &sortop,
3649 : &hashable);
3650 2343 : reverse = true;
3651 2343 : break;
3652 142 : case SORTBY_USING:
3653 : Assert(sortby->useOp != NIL);
3654 142 : sortop = compatible_oper_opid(sortby->useOp,
3655 : restype,
3656 : restype,
3657 : false);
3658 :
3659 : /*
3660 : * Verify it's a valid ordering operator, fetch the corresponding
3661 : * equality operator, and determine whether to consider it like
3662 : * ASC or DESC.
3663 : */
3664 142 : eqop = get_equality_op_for_ordering_op(sortop, &reverse);
3665 142 : if (!OidIsValid(eqop))
3666 0 : ereport(ERROR,
3667 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
3668 : errmsg("operator %s is not a valid ordering operator",
3669 : strVal(llast(sortby->useOp))),
3670 : errhint("Ordering operators must be \"<\" or \">\" members of btree operator families.")));
3671 :
3672 : /*
3673 : * Also see if the equality operator is hashable.
3674 : */
3675 142 : hashable = op_hashjoinable(eqop, restype);
3676 142 : break;
3677 0 : default:
3678 0 : elog(ERROR, "unrecognized sortby_dir: %d", sortby->sortby_dir);
3679 : sortop = InvalidOid; /* keep compiler quiet */
3680 : eqop = InvalidOid;
3681 : hashable = false;
3682 : reverse = false;
3683 : break;
3684 : }
3685 :
3686 74397 : cancel_parser_errposition_callback(&pcbstate);
3687 :
3688 : /* avoid making duplicate sortlist entries */
3689 74397 : if (!targetIsInSortList(tle, sortop, sortlist))
3690 : {
3691 74397 : SortGroupClause *sortcl = makeNode(SortGroupClause);
3692 :
3693 74397 : sortcl->tleSortGroupRef = assignSortGroupRef(tle, targetlist);
3694 :
3695 74397 : sortcl->eqop = eqop;
3696 74397 : sortcl->sortop = sortop;
3697 74397 : sortcl->hashable = hashable;
3698 74397 : sortcl->reverse_sort = reverse;
3699 :
3700 74397 : switch (sortby->sortby_nulls)
3701 : {
3702 73306 : case SORTBY_NULLS_DEFAULT:
3703 : /* NULLS FIRST is default for DESC; other way for ASC */
3704 73306 : sortcl->nulls_first = reverse;
3705 73306 : break;
3706 203 : case SORTBY_NULLS_FIRST:
3707 203 : sortcl->nulls_first = true;
3708 203 : break;
3709 888 : case SORTBY_NULLS_LAST:
3710 888 : sortcl->nulls_first = false;
3711 888 : break;
3712 0 : default:
3713 0 : elog(ERROR, "unrecognized sortby_nulls: %d",
3714 : sortby->sortby_nulls);
3715 : break;
3716 : }
3717 :
3718 74397 : sortlist = lappend(sortlist, sortcl);
3719 : }
3720 :
3721 74397 : return sortlist;
3722 : }
3723 :
3724 : /*
3725 : * addTargetToGroupList
3726 : * If the given targetlist entry isn't already in the SortGroupClause
3727 : * list, add it to the end of the list, using default sort/group
3728 : * semantics.
3729 : *
3730 : * This is very similar to addTargetToSortList, except that we allow the
3731 : * case where only a grouping (equality) operator can be found, and that
3732 : * the TLE is considered "already in the list" if it appears there with any
3733 : * sorting semantics.
3734 : *
3735 : * location is the parse location to be fingered in event of trouble. Note
3736 : * that we can't rely on exprLocation(tle->expr), because that might point
3737 : * to a SELECT item that matches the GROUP BY item; it'd be pretty confusing
3738 : * to report such a location.
3739 : *
3740 : * Returns the updated SortGroupClause list.
3741 : */
3742 : static List *
3743 12250 : addTargetToGroupList(ParseState *pstate, TargetEntry *tle,
3744 : List *grouplist, List *targetlist, int location)
3745 : {
3746 12250 : Oid restype = exprType((Node *) tle->expr);
3747 :
3748 : /* if tlist item is an UNKNOWN literal, change it to TEXT */
3749 12250 : if (restype == UNKNOWNOID)
3750 : {
3751 10 : tle->expr = (Expr *) coerce_type(pstate, (Node *) tle->expr,
3752 : restype, TEXTOID, -1,
3753 : COERCION_IMPLICIT,
3754 : COERCE_IMPLICIT_CAST,
3755 : -1);
3756 10 : restype = TEXTOID;
3757 : }
3758 :
3759 : /* avoid making duplicate grouplist entries */
3760 12250 : if (!targetIsInSortList(tle, InvalidOid, grouplist))
3761 : {
3762 11851 : SortGroupClause *grpcl = makeNode(SortGroupClause);
3763 : Oid sortop;
3764 : Oid eqop;
3765 : bool hashable;
3766 : ParseCallbackState pcbstate;
3767 :
3768 11851 : setup_parser_errposition_callback(&pcbstate, pstate, location);
3769 :
3770 : /* determine the eqop and optional sortop */
3771 11851 : get_sort_group_operators(restype,
3772 : false, true, false,
3773 : &sortop, &eqop, NULL,
3774 : &hashable);
3775 :
3776 11851 : cancel_parser_errposition_callback(&pcbstate);
3777 :
3778 11851 : grpcl->tleSortGroupRef = assignSortGroupRef(tle, targetlist);
3779 11851 : grpcl->eqop = eqop;
3780 11851 : grpcl->sortop = sortop;
3781 11851 : grpcl->reverse_sort = false; /* sortop is "less than", or
3782 : * InvalidOid */
3783 11851 : grpcl->nulls_first = false; /* OK with or without sortop */
3784 11851 : grpcl->hashable = hashable;
3785 :
3786 11851 : grouplist = lappend(grouplist, grpcl);
3787 : }
3788 :
3789 12250 : return grouplist;
3790 : }
3791 :
3792 : /*
3793 : * assignSortGroupRef
3794 : * Assign the targetentry an unused ressortgroupref, if it doesn't
3795 : * already have one. Return the assigned or pre-existing refnumber.
3796 : *
3797 : * 'tlist' is the targetlist containing (or to contain) the given targetentry.
3798 : */
3799 : Index
3800 126622 : assignSortGroupRef(TargetEntry *tle, List *tlist)
3801 : {
3802 : Index maxRef;
3803 : ListCell *l;
3804 :
3805 126622 : if (tle->ressortgroupref) /* already has one? */
3806 4353 : return tle->ressortgroupref;
3807 :
3808 : /* easiest way to pick an unused refnumber: max used + 1 */
3809 122269 : maxRef = 0;
3810 706740 : foreach(l, tlist)
3811 : {
3812 584471 : Index ref = ((TargetEntry *) lfirst(l))->ressortgroupref;
3813 :
3814 584471 : if (ref > maxRef)
3815 99603 : maxRef = ref;
3816 : }
3817 122269 : tle->ressortgroupref = maxRef + 1;
3818 122269 : return tle->ressortgroupref;
3819 : }
3820 :
3821 : /*
3822 : * targetIsInSortList
3823 : * Is the given target item already in the sortlist?
3824 : * If sortop is not InvalidOid, also test for a match to the sortop.
3825 : *
3826 : * It is not an oversight that this function ignores the nulls_first flag.
3827 : * We check sortop when determining if an ORDER BY item is redundant with
3828 : * earlier ORDER BY items, because it's conceivable that "ORDER BY
3829 : * foo USING <, foo USING <<<" is not redundant, if <<< distinguishes
3830 : * values that < considers equal. We need not check nulls_first
3831 : * however, because a lower-order column with the same sortop but
3832 : * opposite nulls direction is redundant. Also, we can consider
3833 : * ORDER BY foo ASC, foo DESC redundant, so check for a commutator match.
3834 : *
3835 : * Works for both ordering and grouping lists (sortop would normally be
3836 : * InvalidOid when considering grouping). Note that the main reason we need
3837 : * this routine (and not just a quick test for nonzeroness of ressortgroupref)
3838 : * is that a TLE might be in only one of the lists.
3839 : */
3840 : bool
3841 90468 : targetIsInSortList(TargetEntry *tle, Oid sortop, List *sortList)
3842 : {
3843 90468 : Index ref = tle->ressortgroupref;
3844 : ListCell *l;
3845 :
3846 : /* no need to scan list if tle has no marker */
3847 90468 : if (ref == 0)
3848 86697 : return false;
3849 :
3850 4673 : foreach(l, sortList)
3851 : {
3852 2204 : SortGroupClause *scl = (SortGroupClause *) lfirst(l);
3853 :
3854 2204 : if (scl->tleSortGroupRef == ref &&
3855 0 : (sortop == InvalidOid ||
3856 0 : sortop == scl->sortop ||
3857 0 : sortop == get_commutator(scl->sortop)))
3858 1302 : return true;
3859 : }
3860 2469 : return false;
3861 : }
3862 :
3863 : /*
3864 : * findWindowClause
3865 : * Find the named WindowClause in the list, or return NULL if not there
3866 : */
3867 : static WindowClause *
3868 454 : findWindowClause(List *wclist, const char *name)
3869 : {
3870 : ListCell *l;
3871 :
3872 470 : foreach(l, wclist)
3873 : {
3874 48 : WindowClause *wc = (WindowClause *) lfirst(l);
3875 :
3876 48 : if (wc->name && strcmp(wc->name, name) == 0)
3877 32 : return wc;
3878 : }
3879 :
3880 422 : return NULL;
3881 : }
3882 :
3883 : /*
3884 : * transformFrameOffset
3885 : * Process a window frame offset expression
3886 : *
3887 : * In RANGE mode, rangeopfamily is the sort opfamily for the input ORDER BY
3888 : * column, and rangeopcintype is the input data type the sort operator is
3889 : * registered with. We expect the in_range function to be registered with
3890 : * that same type. (In binary-compatible cases, it might be different from
3891 : * the input column's actual type, so we can't use that for the lookups.)
3892 : * We'll return the OID of the in_range function to *inRangeFunc.
3893 : */
3894 : static Node *
3895 4040 : transformFrameOffset(ParseState *pstate, int frameOptions,
3896 : Oid rangeopfamily, Oid rangeopcintype, Oid *inRangeFunc,
3897 : Node *clause)
3898 : {
3899 4040 : const char *constructName = NULL;
3900 : Node *node;
3901 :
3902 4040 : *inRangeFunc = InvalidOid; /* default result */
3903 :
3904 : /* Quick exit if no offset expression */
3905 4040 : if (clause == NULL)
3906 2750 : return NULL;
3907 :
3908 1290 : if (frameOptions & FRAMEOPTION_ROWS)
3909 : {
3910 : /* Transform the raw expression tree */
3911 334 : node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_ROWS);
3912 :
3913 : /*
3914 : * Like LIMIT clause, simply coerce to int8
3915 : */
3916 334 : constructName = "ROWS";
3917 334 : node = coerce_to_specific_type(pstate, node, INT8OID, constructName);
3918 : }
3919 956 : else if (frameOptions & FRAMEOPTION_RANGE)
3920 : {
3921 : /*
3922 : * We must look up the in_range support function that's to be used,
3923 : * possibly choosing one of several, and coerce the "offset" value to
3924 : * the appropriate input type.
3925 : */
3926 : Oid nodeType;
3927 : Oid preferredType;
3928 768 : int nfuncs = 0;
3929 768 : int nmatches = 0;
3930 768 : Oid selectedType = InvalidOid;
3931 768 : Oid selectedFunc = InvalidOid;
3932 : CatCList *proclist;
3933 : int i;
3934 :
3935 : /* Transform the raw expression tree */
3936 768 : node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_RANGE);
3937 768 : nodeType = exprType(node);
3938 :
3939 : /*
3940 : * If there are multiple candidates, we'll prefer the one that exactly
3941 : * matches nodeType; or if nodeType is as yet unknown, prefer the one
3942 : * that exactly matches the sort column type. (The second rule is
3943 : * like what we do for "known_type operator unknown".)
3944 : */
3945 768 : preferredType = (nodeType != UNKNOWNOID) ? nodeType : rangeopcintype;
3946 :
3947 : /* Find the in_range support functions applicable to this case */
3948 768 : proclist = SearchSysCacheList2(AMPROCNUM,
3949 : ObjectIdGetDatum(rangeopfamily),
3950 : ObjectIdGetDatum(rangeopcintype));
3951 5340 : for (i = 0; i < proclist->n_members; i++)
3952 : {
3953 4572 : HeapTuple proctup = &proclist->members[i]->tuple;
3954 4572 : Form_pg_amproc procform = (Form_pg_amproc) GETSTRUCT(proctup);
3955 :
3956 : /* The search will find all support proc types; ignore others */
3957 4572 : if (procform->amprocnum != BTINRANGE_PROC)
3958 3400 : continue;
3959 1172 : nfuncs++;
3960 :
3961 : /* Ignore function if given value can't be coerced to that type */
3962 1172 : if (!can_coerce_type(1, &nodeType, &procform->amprocrighttype,
3963 : COERCION_IMPLICIT))
3964 220 : continue;
3965 952 : nmatches++;
3966 :
3967 : /* Remember preferred match, or any match if didn't find that */
3968 952 : if (selectedType != preferredType)
3969 : {
3970 912 : selectedType = procform->amprocrighttype;
3971 912 : selectedFunc = procform->amproc;
3972 : }
3973 : }
3974 768 : ReleaseCatCacheList(proclist);
3975 :
3976 : /*
3977 : * Throw error if needed. It seems worth taking the trouble to
3978 : * distinguish "no support at all" from "you didn't match any
3979 : * available offset type".
3980 : */
3981 768 : if (nfuncs == 0)
3982 4 : ereport(ERROR,
3983 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3984 : errmsg("RANGE with offset PRECEDING/FOLLOWING is not supported for column type %s",
3985 : format_type_be(rangeopcintype)),
3986 : parser_errposition(pstate, exprLocation(node))));
3987 764 : if (nmatches == 0)
3988 12 : ereport(ERROR,
3989 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3990 : errmsg("RANGE with offset PRECEDING/FOLLOWING is not supported for column type %s and offset type %s",
3991 : format_type_be(rangeopcintype),
3992 : format_type_be(nodeType)),
3993 : errhint("Cast the offset value to an appropriate type."),
3994 : parser_errposition(pstate, exprLocation(node))));
3995 752 : if (nmatches != 1 && selectedType != preferredType)
3996 0 : ereport(ERROR,
3997 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3998 : errmsg("RANGE with offset PRECEDING/FOLLOWING has multiple interpretations for column type %s and offset type %s",
3999 : format_type_be(rangeopcintype),
4000 : format_type_be(nodeType)),
4001 : errhint("Cast the offset value to the exact intended type."),
4002 : parser_errposition(pstate, exprLocation(node))));
4003 :
4004 : /* OK, coerce the offset to the right type */
4005 752 : constructName = "RANGE";
4006 752 : node = coerce_to_specific_type(pstate, node,
4007 : selectedType, constructName);
4008 752 : *inRangeFunc = selectedFunc;
4009 : }
4010 188 : else if (frameOptions & FRAMEOPTION_GROUPS)
4011 : {
4012 : /* Transform the raw expression tree */
4013 188 : node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_GROUPS);
4014 :
4015 : /*
4016 : * Like LIMIT clause, simply coerce to int8
4017 : */
4018 188 : constructName = "GROUPS";
4019 188 : node = coerce_to_specific_type(pstate, node, INT8OID, constructName);
4020 : }
4021 : else
4022 : {
4023 : Assert(false);
4024 0 : node = NULL;
4025 : }
4026 :
4027 : /* Disallow variables in frame offsets */
4028 1274 : checkExprIsVarFree(pstate, node, constructName);
4029 :
4030 1270 : return node;
4031 : }
|
