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 321757 : 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 574845 : foreach(fl, frmList)
130 : {
131 253611 : Node *n = lfirst(fl);
132 : ParseNamespaceItem *nsitem;
133 : List *namespace;
134 :
135 253611 : n = transformFromClauseItem(pstate, n,
136 : &nsitem,
137 : &namespace);
138 :
139 253092 : checkNameSpaceConflicts(pstate, pstate->p_namespace, namespace);
140 :
141 : /* Mark the new namespace items as visible only to LATERAL */
142 253088 : setNamespaceLateralState(namespace, true, true);
143 :
144 253088 : pstate->p_joinlist = lappend(pstate->p_joinlist, n);
145 253088 : 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 321234 : setNamespaceLateralState(pstate->p_namespace, false, true);
155 321234 : }
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 59151 : 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 112317 : if (relation->schemaname == NULL &&
192 53166 : 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 59147 : 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 59147 : pstate->p_target_relation = parserOpenTable(pstate, relation,
210 : RowExclusiveLock);
211 :
212 : /*
213 : * Now build an RTE and a ParseNamespaceItem.
214 : */
215 59130 : 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 59130 : 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 59130 : nsitem->p_perminfo->requiredPerms = requiredPerms;
232 :
233 : /*
234 : * If UPDATE/DELETE, add table to joinlist and namespace.
235 : */
236 59130 : if (alsoSource)
237 12599 : addNSItemToQuery(pstate, nsitem, true, true, true);
238 :
239 59130 : 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 114016 : 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 114016 : 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 114016 : prevcols = NULL;
276 116382 : foreach(lc, *src_colnos)
277 : {
278 2366 : prevcols = bms_add_member(prevcols, lfirst_int(lc));
279 : }
280 :
281 114016 : attnum = 0;
282 2315351 : foreach(lc, src_colnames)
283 : {
284 2201335 : char *colname = strVal(lfirst(lc));
285 :
286 2201335 : attnum++;
287 : /* Non-dropped and not already merged? */
288 2201335 : if (colname[0] != '\0' && !bms_is_member(attnum, prevcols))
289 : {
290 : /* Yes, so emit it as next output column */
291 2198650 : *src_colnos = lappend_int(*src_colnos, attnum);
292 2198650 : *res_colnames = lappend(*res_colnames, lfirst(lc));
293 2198650 : *res_colvars = lappend(*res_colvars,
294 2198650 : buildVarFromNSColumn(pstate,
295 2198650 : src_nscolumns + attnum - 1));
296 : /* Copy the input relation's nscolumn data for this column */
297 2198650 : res_nscolumns[colcount] = src_nscolumns[attnum - 1];
298 2198650 : colcount++;
299 : }
300 : }
301 114016 : return colcount;
302 : }
303 :
304 : /*
305 : * transformJoinUsingClause()
306 : * Build a complete ON clause from a partially-transformed USING list.
307 : * We are given lists of nodes representing left and right match columns.
308 : * Result is a transformed qualification expression.
309 : */
310 : static Node *
311 1028 : transformJoinUsingClause(ParseState *pstate,
312 : List *leftVars, List *rightVars)
313 : {
314 : Node *result;
315 1028 : List *andargs = NIL;
316 : ListCell *lvars,
317 : *rvars;
318 :
319 : /*
320 : * We cheat a little bit here by building an untransformed operator tree
321 : * whose leaves are the already-transformed Vars. This requires collusion
322 : * from transformExpr(), which normally could be expected to complain
323 : * about already-transformed subnodes. However, this does mean that we
324 : * have to mark the columns as requiring SELECT privilege for ourselves;
325 : * transformExpr() won't do it.
326 : */
327 2211 : forboth(lvars, leftVars, rvars, rightVars)
328 : {
329 1183 : Var *lvar = (Var *) lfirst(lvars);
330 1183 : Var *rvar = (Var *) lfirst(rvars);
331 : A_Expr *e;
332 :
333 : /* Require read access to the join variables */
334 1183 : markVarForSelectPriv(pstate, lvar);
335 1183 : markVarForSelectPriv(pstate, rvar);
336 :
337 : /* Now create the lvar = rvar join condition */
338 1183 : e = makeSimpleA_Expr(AEXPR_OP, "=",
339 1183 : (Node *) copyObject(lvar), (Node *) copyObject(rvar),
340 : -1);
341 :
342 : /* Prepare to combine into an AND clause, if multiple join columns */
343 1183 : andargs = lappend(andargs, e);
344 : }
345 :
346 : /* Only need an AND if there's more than one join column */
347 1028 : if (list_length(andargs) == 1)
348 896 : result = (Node *) linitial(andargs);
349 : else
350 132 : result = (Node *) makeBoolExpr(AND_EXPR, andargs, -1);
351 :
352 : /*
353 : * Since the references are already Vars, and are certainly from the input
354 : * relations, we don't have to go through the same pushups that
355 : * transformJoinOnClause() does. Just invoke transformExpr() to fix up
356 : * the operators, and we're done.
357 : */
358 1028 : result = transformExpr(pstate, result, EXPR_KIND_JOIN_USING);
359 :
360 1028 : result = coerce_to_boolean(pstate, result, "JOIN/USING");
361 :
362 1028 : return result;
363 : }
364 :
365 : /*
366 : * transformJoinOnClause()
367 : * Transform the qual conditions for JOIN/ON.
368 : * Result is a transformed qualification expression.
369 : */
370 : static Node *
371 55641 : transformJoinOnClause(ParseState *pstate, JoinExpr *j, List *namespace)
372 : {
373 : Node *result;
374 : List *save_namespace;
375 :
376 : /*
377 : * The namespace that the join expression should see is just the two
378 : * subtrees of the JOIN plus any outer references from upper pstate
379 : * levels. Temporarily set this pstate's namespace accordingly. (We need
380 : * not check for refname conflicts, because transformFromClauseItem()
381 : * already did.) All namespace items are marked visible regardless of
382 : * LATERAL state.
383 : */
384 55641 : setNamespaceLateralState(namespace, false, true);
385 :
386 55641 : save_namespace = pstate->p_namespace;
387 55641 : pstate->p_namespace = namespace;
388 :
389 55641 : result = transformWhereClause(pstate, j->quals,
390 : EXPR_KIND_JOIN_ON, "JOIN/ON");
391 :
392 55629 : pstate->p_namespace = save_namespace;
393 :
394 55629 : return result;
395 : }
396 :
397 : /*
398 : * transformTableEntry --- transform a RangeVar (simple relation reference)
399 : */
400 : static ParseNamespaceItem *
401 260702 : transformTableEntry(ParseState *pstate, RangeVar *r)
402 : {
403 : /* addRangeTableEntry does all the work */
404 260702 : return addRangeTableEntry(pstate, r, r->alias, r->inh, true);
405 : }
406 :
407 : /*
408 : * transformRangeSubselect --- transform a sub-SELECT appearing in FROM
409 : */
410 : static ParseNamespaceItem *
411 13938 : transformRangeSubselect(ParseState *pstate, RangeSubselect *r)
412 : {
413 : Query *query;
414 :
415 : /*
416 : * Set p_expr_kind to show this parse level is recursing to a subselect.
417 : * We can't be nested within any expression, so don't need save-restore
418 : * logic here.
419 : */
420 : Assert(pstate->p_expr_kind == EXPR_KIND_NONE);
421 13938 : pstate->p_expr_kind = EXPR_KIND_FROM_SUBSELECT;
422 :
423 : /*
424 : * If the subselect is LATERAL, make lateral_only names of this level
425 : * visible to it. (LATERAL can't nest within a single pstate level, so we
426 : * don't need save/restore logic here.)
427 : */
428 : Assert(!pstate->p_lateral_active);
429 13938 : pstate->p_lateral_active = r->lateral;
430 :
431 : /*
432 : * Analyze and transform the subquery. Note that if the subquery doesn't
433 : * have an alias, it can't be explicitly selected for locking, but locking
434 : * might still be required (if there is an all-tables locking clause).
435 : */
436 13938 : query = parse_sub_analyze(r->subquery, pstate, NULL,
437 13938 : isLockedRefname(pstate,
438 13938 : r->alias == NULL ? NULL :
439 13768 : r->alias->aliasname),
440 : true);
441 :
442 : /* Restore state */
443 13866 : pstate->p_lateral_active = false;
444 13866 : pstate->p_expr_kind = EXPR_KIND_NONE;
445 :
446 : /*
447 : * Check that we got a SELECT. Anything else should be impossible given
448 : * restrictions of the grammar, but check anyway.
449 : */
450 13866 : if (!IsA(query, Query) ||
451 13866 : query->commandType != CMD_SELECT)
452 0 : elog(ERROR, "unexpected non-SELECT command in subquery in FROM");
453 :
454 : /*
455 : * OK, build an RTE and nsitem for the subquery.
456 : */
457 27728 : return addRangeTableEntryForSubquery(pstate,
458 : query,
459 : r->alias,
460 13866 : r->lateral,
461 : true);
462 : }
463 :
464 :
465 : /*
466 : * transformRangeFunction --- transform a function call appearing in FROM
467 : */
468 : static ParseNamespaceItem *
469 30201 : transformRangeFunction(ParseState *pstate, RangeFunction *r)
470 : {
471 30201 : List *funcexprs = NIL;
472 30201 : List *funcnames = NIL;
473 30201 : List *coldeflists = NIL;
474 : bool is_lateral;
475 : ListCell *lc;
476 :
477 : /*
478 : * We make lateral_only names of this level visible, whether or not the
479 : * RangeFunction is explicitly marked LATERAL. This is needed for SQL
480 : * spec compliance in the case of UNNEST(), and seems useful on
481 : * convenience grounds for all functions in FROM.
482 : *
483 : * (LATERAL can't nest within a single pstate level, so we don't need
484 : * save/restore logic here.)
485 : */
486 : Assert(!pstate->p_lateral_active);
487 30201 : pstate->p_lateral_active = true;
488 :
489 : /*
490 : * Transform the raw expressions.
491 : *
492 : * While transforming, also save function names for possible use as alias
493 : * and column names. We use the same transformation rules as for a SELECT
494 : * output expression. For a FuncCall node, the result will be the
495 : * function name, but it is possible for the grammar to hand back other
496 : * node types.
497 : *
498 : * We have to get this info now, because FigureColname only works on raw
499 : * parsetrees. Actually deciding what to do with the names is left up to
500 : * addRangeTableEntryForFunction.
501 : *
502 : * Likewise, collect column definition lists if there were any. But
503 : * complain if we find one here and the RangeFunction has one too.
504 : */
505 60413 : foreach(lc, r->functions)
506 : {
507 30325 : List *pair = (List *) lfirst(lc);
508 : Node *fexpr;
509 : List *coldeflist;
510 : Node *newfexpr;
511 : Node *last_srf;
512 :
513 : /* Disassemble the function-call/column-def-list pairs */
514 : Assert(list_length(pair) == 2);
515 30325 : fexpr = (Node *) linitial(pair);
516 30325 : coldeflist = (List *) lsecond(pair);
517 :
518 : /*
519 : * If we find a function call unnest() with more than one argument and
520 : * no special decoration, transform it into separate unnest() calls on
521 : * each argument. This is a kluge, for sure, but it's less nasty than
522 : * other ways of implementing the SQL-standard UNNEST() syntax.
523 : *
524 : * If there is any decoration (including a coldeflist), we don't
525 : * transform, which probably means a no-such-function error later. We
526 : * could alternatively throw an error right now, but that doesn't seem
527 : * tremendously helpful. If someone is using any such decoration,
528 : * then they're not using the SQL-standard syntax, and they're more
529 : * likely expecting an un-tweaked function call.
530 : *
531 : * Note: the transformation changes a non-schema-qualified unnest()
532 : * function name into schema-qualified pg_catalog.unnest(). This
533 : * choice is also a bit debatable, but it seems reasonable to force
534 : * use of built-in unnest() when we make this transformation.
535 : */
536 30325 : if (IsA(fexpr, FuncCall))
537 : {
538 30229 : FuncCall *fc = (FuncCall *) fexpr;
539 :
540 30229 : if (list_length(fc->funcname) == 1 &&
541 21772 : strcmp(strVal(linitial(fc->funcname)), "unnest") == 0 &&
542 2100 : list_length(fc->args) > 1 &&
543 43 : fc->agg_order == NIL &&
544 43 : fc->agg_filter == NULL &&
545 43 : fc->over == NULL &&
546 43 : !fc->agg_star &&
547 43 : !fc->agg_distinct &&
548 43 : !fc->func_variadic &&
549 : coldeflist == NIL)
550 43 : {
551 : ListCell *lc2;
552 :
553 155 : foreach(lc2, fc->args)
554 : {
555 112 : Node *arg = (Node *) lfirst(lc2);
556 : FuncCall *newfc;
557 :
558 112 : last_srf = pstate->p_last_srf;
559 :
560 112 : newfc = makeFuncCall(SystemFuncName("unnest"),
561 : list_make1(arg),
562 : COERCE_EXPLICIT_CALL,
563 : fc->location);
564 :
565 112 : newfexpr = transformExpr(pstate, (Node *) newfc,
566 : EXPR_KIND_FROM_FUNCTION);
567 :
568 : /* nodeFunctionscan.c requires SRFs to be at top level */
569 112 : if (pstate->p_last_srf != last_srf &&
570 112 : pstate->p_last_srf != newfexpr)
571 0 : ereport(ERROR,
572 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
573 : errmsg("set-returning functions must appear at top level of FROM"),
574 : parser_errposition(pstate,
575 : exprLocation(pstate->p_last_srf))));
576 :
577 112 : funcexprs = lappend(funcexprs, newfexpr);
578 :
579 112 : funcnames = lappend(funcnames,
580 112 : FigureColname((Node *) newfc));
581 :
582 : /* coldeflist is empty, so no error is possible */
583 :
584 112 : coldeflists = lappend(coldeflists, coldeflist);
585 : }
586 43 : continue; /* done with this function item */
587 : }
588 : }
589 :
590 : /* normal case ... */
591 30282 : last_srf = pstate->p_last_srf;
592 :
593 30282 : newfexpr = transformExpr(pstate, fexpr,
594 : EXPR_KIND_FROM_FUNCTION);
595 :
596 : /* nodeFunctionscan.c requires SRFs to be at top level */
597 30173 : if (pstate->p_last_srf != last_srf &&
598 24421 : pstate->p_last_srf != newfexpr)
599 4 : ereport(ERROR,
600 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
601 : errmsg("set-returning functions must appear at top level of FROM"),
602 : parser_errposition(pstate,
603 : exprLocation(pstate->p_last_srf))));
604 :
605 30169 : funcexprs = lappend(funcexprs, newfexpr);
606 :
607 30169 : funcnames = lappend(funcnames,
608 30169 : FigureColname(fexpr));
609 :
610 30169 : if (coldeflist && r->coldeflist)
611 0 : ereport(ERROR,
612 : (errcode(ERRCODE_SYNTAX_ERROR),
613 : errmsg("multiple column definition lists are not allowed for the same function"),
614 : parser_errposition(pstate,
615 : exprLocation((Node *) r->coldeflist))));
616 :
617 30169 : coldeflists = lappend(coldeflists, coldeflist);
618 : }
619 :
620 30088 : pstate->p_lateral_active = false;
621 :
622 : /*
623 : * We must assign collations now so that the RTE exposes correct collation
624 : * info for Vars created from it.
625 : */
626 30088 : assign_list_collations(pstate, funcexprs);
627 :
628 : /*
629 : * Install the top-level coldeflist if there was one (we already checked
630 : * that there was no conflicting per-function coldeflist).
631 : *
632 : * We only allow this when there's a single function (even after UNNEST
633 : * expansion) and no WITH ORDINALITY. The reason for the latter
634 : * restriction is that it's not real clear whether the ordinality column
635 : * should be in the coldeflist, and users are too likely to make mistakes
636 : * in one direction or the other. Putting the coldeflist inside ROWS
637 : * FROM() is much clearer in this case.
638 : */
639 30088 : if (r->coldeflist)
640 : {
641 460 : if (list_length(funcexprs) != 1)
642 : {
643 0 : if (r->is_rowsfrom)
644 0 : ereport(ERROR,
645 : (errcode(ERRCODE_SYNTAX_ERROR),
646 : errmsg("ROWS FROM() with multiple functions cannot have a column definition list"),
647 : errhint("Put a separate column definition list for each function inside ROWS FROM()."),
648 : parser_errposition(pstate,
649 : exprLocation((Node *) r->coldeflist))));
650 : else
651 0 : ereport(ERROR,
652 : (errcode(ERRCODE_SYNTAX_ERROR),
653 : errmsg("UNNEST() with multiple arguments cannot have a column definition list"),
654 : errhint("Use separate UNNEST() calls inside ROWS FROM(), and attach a column definition list to each one."),
655 : parser_errposition(pstate,
656 : exprLocation((Node *) r->coldeflist))));
657 : }
658 460 : if (r->ordinality)
659 0 : ereport(ERROR,
660 : (errcode(ERRCODE_SYNTAX_ERROR),
661 : errmsg("WITH ORDINALITY cannot be used with a column definition list"),
662 : errhint("Put the column definition list inside ROWS FROM()."),
663 : parser_errposition(pstate,
664 : exprLocation((Node *) r->coldeflist))));
665 :
666 460 : coldeflists = list_make1(r->coldeflist);
667 : }
668 :
669 : /*
670 : * Mark the RTE as LATERAL if the user said LATERAL explicitly, or if
671 : * there are any lateral cross-references in it.
672 : */
673 30088 : is_lateral = r->lateral || contain_vars_of_level((Node *) funcexprs, 0);
674 :
675 : /*
676 : * OK, build an RTE and nsitem for the function.
677 : */
678 30088 : return addRangeTableEntryForFunction(pstate,
679 : funcnames, funcexprs, coldeflists,
680 : r, is_lateral, true);
681 : }
682 :
683 : /*
684 : * transformRangeTableFunc -
685 : * Transform a raw RangeTableFunc into TableFunc.
686 : *
687 : * Transform the namespace clauses, the document-generating expression, the
688 : * row-generating expression, the column-generating expressions, and the
689 : * default value expressions.
690 : */
691 : static ParseNamespaceItem *
692 146 : transformRangeTableFunc(ParseState *pstate, RangeTableFunc *rtf)
693 : {
694 146 : TableFunc *tf = makeNode(TableFunc);
695 : const char *constructName;
696 : Oid docType;
697 : bool is_lateral;
698 : ListCell *col;
699 : char **names;
700 : int colno;
701 :
702 : /*
703 : * Currently we only support XMLTABLE here. See transformJsonTable() for
704 : * JSON_TABLE support.
705 : */
706 146 : tf->functype = TFT_XMLTABLE;
707 146 : constructName = "XMLTABLE";
708 146 : docType = XMLOID;
709 :
710 : /*
711 : * We make lateral_only names of this level visible, whether or not the
712 : * RangeTableFunc is explicitly marked LATERAL. This is needed for SQL
713 : * spec compliance and seems useful on convenience grounds for all
714 : * functions in FROM.
715 : *
716 : * (LATERAL can't nest within a single pstate level, so we don't need
717 : * save/restore logic here.)
718 : */
719 : Assert(!pstate->p_lateral_active);
720 146 : pstate->p_lateral_active = true;
721 :
722 : /* Transform and apply typecast to the row-generating expression ... */
723 : Assert(rtf->rowexpr != NULL);
724 146 : tf->rowexpr = coerce_to_specific_type(pstate,
725 : transformExpr(pstate, rtf->rowexpr, EXPR_KIND_FROM_FUNCTION),
726 : TEXTOID,
727 : constructName);
728 146 : assign_expr_collations(pstate, tf->rowexpr);
729 :
730 : /* ... and to the document itself */
731 : Assert(rtf->docexpr != NULL);
732 146 : tf->docexpr = coerce_to_specific_type(pstate,
733 : transformExpr(pstate, rtf->docexpr, EXPR_KIND_FROM_FUNCTION),
734 : docType,
735 : constructName);
736 146 : assign_expr_collations(pstate, tf->docexpr);
737 :
738 : /* undef ordinality column number */
739 146 : tf->ordinalitycol = -1;
740 :
741 : /* Process column specs */
742 146 : names = palloc_array(char *, list_length(rtf->columns));
743 :
744 146 : colno = 0;
745 643 : foreach(col, rtf->columns)
746 : {
747 497 : RangeTableFuncCol *rawc = (RangeTableFuncCol *) lfirst(col);
748 : Oid typid;
749 : int32 typmod;
750 : Node *colexpr;
751 : Node *coldefexpr;
752 : int j;
753 :
754 497 : tf->colnames = lappend(tf->colnames,
755 497 : makeString(pstrdup(rawc->colname)));
756 :
757 : /*
758 : * Determine the type and typmod for the new column. FOR ORDINALITY
759 : * columns are INTEGER per spec; the others are user-specified.
760 : */
761 497 : if (rawc->for_ordinality)
762 : {
763 41 : if (tf->ordinalitycol != -1)
764 0 : ereport(ERROR,
765 : (errcode(ERRCODE_SYNTAX_ERROR),
766 : errmsg("only one FOR ORDINALITY column is allowed"),
767 : parser_errposition(pstate, rawc->location)));
768 :
769 41 : typid = INT4OID;
770 41 : typmod = -1;
771 41 : tf->ordinalitycol = colno;
772 : }
773 : else
774 : {
775 456 : if (rawc->typeName->setof)
776 0 : ereport(ERROR,
777 : (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
778 : errmsg("column \"%s\" cannot be declared SETOF",
779 : rawc->colname),
780 : parser_errposition(pstate, rawc->location)));
781 :
782 456 : typenameTypeIdAndMod(pstate, rawc->typeName,
783 : &typid, &typmod);
784 : }
785 :
786 497 : tf->coltypes = lappend_oid(tf->coltypes, typid);
787 497 : tf->coltypmods = lappend_int(tf->coltypmods, typmod);
788 497 : tf->colcollations = lappend_oid(tf->colcollations,
789 : get_typcollation(typid));
790 :
791 : /* Transform the PATH and DEFAULT expressions */
792 497 : if (rawc->colexpr)
793 : {
794 324 : colexpr = coerce_to_specific_type(pstate,
795 : transformExpr(pstate, rawc->colexpr,
796 : EXPR_KIND_FROM_FUNCTION),
797 : TEXTOID,
798 : constructName);
799 324 : assign_expr_collations(pstate, colexpr);
800 : }
801 : else
802 173 : colexpr = NULL;
803 :
804 497 : if (rawc->coldefexpr)
805 : {
806 37 : coldefexpr = coerce_to_specific_type_typmod(pstate,
807 : transformExpr(pstate, rawc->coldefexpr,
808 : EXPR_KIND_FROM_FUNCTION),
809 : typid, typmod,
810 : constructName);
811 37 : assign_expr_collations(pstate, coldefexpr);
812 : }
813 : else
814 460 : coldefexpr = NULL;
815 :
816 497 : tf->colexprs = lappend(tf->colexprs, colexpr);
817 497 : tf->coldefexprs = lappend(tf->coldefexprs, coldefexpr);
818 :
819 497 : if (rawc->is_not_null)
820 37 : tf->notnulls = bms_add_member(tf->notnulls, colno);
821 :
822 : /* make sure column names are unique */
823 1677 : for (j = 0; j < colno; j++)
824 1180 : if (strcmp(names[j], rawc->colname) == 0)
825 0 : ereport(ERROR,
826 : (errcode(ERRCODE_SYNTAX_ERROR),
827 : errmsg("column name \"%s\" is not unique",
828 : rawc->colname),
829 : parser_errposition(pstate, rawc->location)));
830 497 : names[colno] = rawc->colname;
831 :
832 497 : colno++;
833 : }
834 146 : pfree(names);
835 :
836 : /* Namespaces, if any, also need to be transformed */
837 146 : if (rtf->namespaces != NIL)
838 : {
839 : ListCell *ns;
840 : ListCell *lc2;
841 13 : List *ns_uris = NIL;
842 13 : List *ns_names = NIL;
843 13 : bool default_ns_seen = false;
844 :
845 26 : foreach(ns, rtf->namespaces)
846 : {
847 13 : ResTarget *r = (ResTarget *) lfirst(ns);
848 : Node *ns_uri;
849 :
850 : Assert(IsA(r, ResTarget));
851 13 : ns_uri = transformExpr(pstate, r->val, EXPR_KIND_FROM_FUNCTION);
852 13 : ns_uri = coerce_to_specific_type(pstate, ns_uri,
853 : TEXTOID, constructName);
854 13 : assign_expr_collations(pstate, ns_uri);
855 13 : ns_uris = lappend(ns_uris, ns_uri);
856 :
857 : /* Verify consistency of name list: no dupes, only one DEFAULT */
858 13 : if (r->name != NULL)
859 : {
860 9 : foreach(lc2, ns_names)
861 : {
862 0 : String *ns_node = lfirst_node(String, lc2);
863 :
864 0 : if (ns_node == NULL)
865 0 : continue;
866 0 : if (strcmp(strVal(ns_node), r->name) == 0)
867 0 : ereport(ERROR,
868 : (errcode(ERRCODE_SYNTAX_ERROR),
869 : errmsg("namespace name \"%s\" is not unique",
870 : r->name),
871 : parser_errposition(pstate, r->location)));
872 : }
873 : }
874 : else
875 : {
876 4 : if (default_ns_seen)
877 0 : ereport(ERROR,
878 : (errcode(ERRCODE_SYNTAX_ERROR),
879 : errmsg("only one default namespace is allowed"),
880 : parser_errposition(pstate, r->location)));
881 4 : default_ns_seen = true;
882 : }
883 :
884 : /* We represent DEFAULT by a null pointer */
885 13 : ns_names = lappend(ns_names,
886 13 : r->name ? makeString(r->name) : NULL);
887 : }
888 :
889 13 : tf->ns_uris = ns_uris;
890 13 : tf->ns_names = ns_names;
891 : }
892 :
893 146 : tf->location = rtf->location;
894 :
895 146 : pstate->p_lateral_active = false;
896 :
897 : /*
898 : * Mark the RTE as LATERAL if the user said LATERAL explicitly, or if
899 : * there are any lateral cross-references in it.
900 : */
901 146 : is_lateral = rtf->lateral || contain_vars_of_level((Node *) tf, 0);
902 :
903 146 : return addRangeTableEntryForTableFunc(pstate,
904 : tf, rtf->alias, is_lateral, true);
905 : }
906 :
907 : /*
908 : * Similar to parserOpenTable() but for property graphs.
909 : */
910 : static Relation
911 574 : parserOpenPropGraph(ParseState *pstate, const RangeVar *relation, LOCKMODE lockmode)
912 : {
913 : Relation rel;
914 : ParseCallbackState pcbstate;
915 :
916 574 : setup_parser_errposition_callback(&pcbstate, pstate, relation->location);
917 :
918 574 : rel = relation_openrv(relation, lockmode);
919 :
920 : /*
921 : * In parserOpenTable(), the relkind check is done inside table_openrv*.
922 : * We do it here since we don't have anything like propgraph_open.
923 : */
924 570 : if (rel->rd_rel->relkind != RELKIND_PROPGRAPH)
925 4 : ereport(ERROR,
926 : errcode(ERRCODE_WRONG_OBJECT_TYPE),
927 : errmsg("\"%s\" is not a property graph",
928 : RelationGetRelationName(rel)));
929 :
930 566 : cancel_parser_errposition_callback(&pcbstate);
931 566 : return rel;
932 : }
933 :
934 : /*
935 : * transformRangeGraphTable -- transform a GRAPH_TABLE clause
936 : */
937 : static ParseNamespaceItem *
938 574 : transformRangeGraphTable(ParseState *pstate, RangeGraphTable *rgt)
939 : {
940 : Relation rel;
941 : Oid graphid;
942 574 : GraphTableParseState *gpstate = palloc0_object(GraphTableParseState);
943 : Node *gp;
944 574 : List *columns = NIL;
945 574 : List *colnames = NIL;
946 : ListCell *lc;
947 574 : int resno = 0;
948 : bool saved_hasSublinks;
949 :
950 574 : rel = parserOpenPropGraph(pstate, rgt->graph_name, AccessShareLock);
951 :
952 566 : graphid = RelationGetRelid(rel);
953 :
954 566 : gpstate->graphid = graphid;
955 :
956 : /*
957 : * The syntax does not allow nested GRAPH_TABLE and this function
958 : * prohibits subquery within GRAPH_TABLE. There should be only one
959 : * GRAPH_TABLE being transformed at a time.
960 : */
961 : Assert(!pstate->p_graph_table_pstate);
962 566 : pstate->p_graph_table_pstate = gpstate;
963 :
964 : Assert(!pstate->p_lateral_active);
965 566 : pstate->p_lateral_active = true;
966 :
967 566 : saved_hasSublinks = pstate->p_hasSubLinks;
968 566 : pstate->p_hasSubLinks = false;
969 :
970 566 : gp = transformGraphPattern(pstate, rgt->graph_pattern);
971 :
972 : /*
973 : * Construct a targetlist representing the COLUMNS specified in the
974 : * GRAPH_TABLE. This uses previously constructed list of element pattern
975 : * variables in the GraphTableParseState.
976 : */
977 1833 : foreach(lc, rgt->columns)
978 : {
979 1331 : ResTarget *rt = lfirst_node(ResTarget, lc);
980 : Node *colexpr;
981 : TargetEntry *te;
982 : char *colname;
983 :
984 1331 : colexpr = transformExpr(pstate, rt->val, EXPR_KIND_SELECT_TARGET);
985 :
986 1319 : if (rt->name)
987 600 : colname = rt->name;
988 : else
989 : {
990 719 : if (IsA(colexpr, GraphPropertyRef))
991 719 : colname = get_propgraph_property_name(castNode(GraphPropertyRef, colexpr)->propid);
992 : else
993 : {
994 0 : ereport(ERROR,
995 : errcode(ERRCODE_SYNTAX_ERROR),
996 : errmsg("complex graph table column must specify an explicit column name"),
997 : parser_errposition(pstate, rt->location));
998 : colname = NULL;
999 : }
1000 : }
1001 :
1002 1319 : colnames = lappend(colnames, makeString(colname));
1003 :
1004 1319 : te = makeTargetEntry((Expr *) colexpr, ++resno, colname, false);
1005 1319 : columns = lappend(columns, te);
1006 : }
1007 :
1008 : /*
1009 : * Assign collations to column expressions now since
1010 : * assign_query_collations() does not process rangetable entries.
1011 : */
1012 502 : assign_list_collations(pstate, columns);
1013 :
1014 502 : table_close(rel, NoLock);
1015 :
1016 502 : pstate->p_graph_table_pstate = NULL;
1017 502 : pstate->p_lateral_active = false;
1018 :
1019 : /*
1020 : * If we support subqueries within GRAPH_TABLE, those need to be
1021 : * propagated to the queries resulting from rewriting graph table RTE. We
1022 : * don't do that right now, hence prohibit it for now.
1023 : */
1024 502 : if (pstate->p_hasSubLinks)
1025 8 : ereport(ERROR,
1026 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1027 : errmsg("subqueries within GRAPH_TABLE reference are not supported")));
1028 494 : pstate->p_hasSubLinks = saved_hasSublinks;
1029 :
1030 494 : return addRangeTableEntryForGraphTable(pstate, graphid, castNode(GraphPattern, gp), columns, colnames, rgt->alias, false, true);
1031 : }
1032 :
1033 : /*
1034 : * transformRangeTableSample --- transform a TABLESAMPLE clause
1035 : *
1036 : * Caller has already transformed rts->relation, we just have to validate
1037 : * the remaining fields and create a TableSampleClause node.
1038 : */
1039 : static TableSampleClause *
1040 160 : transformRangeTableSample(ParseState *pstate, RangeTableSample *rts)
1041 : {
1042 : TableSampleClause *tablesample;
1043 : Oid handlerOid;
1044 : Oid funcargtypes[1];
1045 : TsmRoutine *tsm;
1046 : List *fargs;
1047 : ListCell *larg,
1048 : *ltyp;
1049 :
1050 : /*
1051 : * To validate the sample method name, look up the handler function, which
1052 : * has the same name, one dummy INTERNAL argument, and a result type of
1053 : * tsm_handler. (Note: tablesample method names are not schema-qualified
1054 : * in the SQL standard; but since they are just functions to us, we allow
1055 : * schema qualification to resolve any potential ambiguity.)
1056 : */
1057 160 : funcargtypes[0] = INTERNALOID;
1058 :
1059 160 : handlerOid = LookupFuncName(rts->method, 1, funcargtypes, true);
1060 :
1061 : /* we want error to complain about no-such-method, not no-such-function */
1062 160 : if (!OidIsValid(handlerOid))
1063 4 : ereport(ERROR,
1064 : (errcode(ERRCODE_UNDEFINED_OBJECT),
1065 : errmsg("tablesample method %s does not exist",
1066 : NameListToString(rts->method)),
1067 : parser_errposition(pstate, rts->location)));
1068 :
1069 : /* check that handler has correct return type */
1070 156 : if (get_func_rettype(handlerOid) != TSM_HANDLEROID)
1071 0 : ereport(ERROR,
1072 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
1073 : errmsg("function %s must return type %s",
1074 : NameListToString(rts->method), "tsm_handler"),
1075 : parser_errposition(pstate, rts->location)));
1076 :
1077 : /* OK, run the handler to get TsmRoutine, for argument type info */
1078 156 : tsm = GetTsmRoutine(handlerOid);
1079 :
1080 156 : tablesample = makeNode(TableSampleClause);
1081 156 : tablesample->tsmhandler = handlerOid;
1082 :
1083 : /* check user provided the expected number of arguments */
1084 156 : if (list_length(rts->args) != list_length(tsm->parameterTypes))
1085 0 : ereport(ERROR,
1086 : (errcode(ERRCODE_INVALID_TABLESAMPLE_ARGUMENT),
1087 : errmsg_plural("tablesample method %s requires %d argument, not %d",
1088 : "tablesample method %s requires %d arguments, not %d",
1089 : list_length(tsm->parameterTypes),
1090 : NameListToString(rts->method),
1091 : list_length(tsm->parameterTypes),
1092 : list_length(rts->args)),
1093 : parser_errposition(pstate, rts->location)));
1094 :
1095 : /*
1096 : * Transform the arguments, typecasting them as needed. Note we must also
1097 : * assign collations now, because assign_query_collations() doesn't
1098 : * examine any substructure of RTEs.
1099 : */
1100 156 : fargs = NIL;
1101 312 : forboth(larg, rts->args, ltyp, tsm->parameterTypes)
1102 : {
1103 156 : Node *arg = (Node *) lfirst(larg);
1104 156 : Oid argtype = lfirst_oid(ltyp);
1105 :
1106 156 : arg = transformExpr(pstate, arg, EXPR_KIND_FROM_FUNCTION);
1107 156 : arg = coerce_to_specific_type(pstate, arg, argtype, "TABLESAMPLE");
1108 156 : assign_expr_collations(pstate, arg);
1109 156 : fargs = lappend(fargs, arg);
1110 : }
1111 156 : tablesample->args = fargs;
1112 :
1113 : /* Process REPEATABLE (seed) */
1114 156 : if (rts->repeatable != NULL)
1115 : {
1116 : Node *arg;
1117 :
1118 67 : if (!tsm->repeatable_across_queries)
1119 2 : ereport(ERROR,
1120 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1121 : errmsg("tablesample method %s does not support REPEATABLE",
1122 : NameListToString(rts->method)),
1123 : parser_errposition(pstate, rts->location)));
1124 :
1125 65 : arg = transformExpr(pstate, rts->repeatable, EXPR_KIND_FROM_FUNCTION);
1126 65 : arg = coerce_to_specific_type(pstate, arg, FLOAT8OID, "REPEATABLE");
1127 65 : assign_expr_collations(pstate, arg);
1128 65 : tablesample->repeatable = (Expr *) arg;
1129 : }
1130 : else
1131 89 : tablesample->repeatable = NULL;
1132 :
1133 154 : return tablesample;
1134 : }
1135 :
1136 : /*
1137 : * getNSItemForSpecialRelationTypes
1138 : *
1139 : * If given RangeVar refers to a CTE or an EphemeralNamedRelation,
1140 : * build and return an appropriate ParseNamespaceItem, otherwise return NULL
1141 : */
1142 : static ParseNamespaceItem *
1143 265448 : getNSItemForSpecialRelationTypes(ParseState *pstate, RangeVar *rv)
1144 : {
1145 : ParseNamespaceItem *nsitem;
1146 : CommonTableExpr *cte;
1147 : Index levelsup;
1148 :
1149 : /*
1150 : * if it is a qualified name, it can't be a CTE or tuplestore reference
1151 : */
1152 265448 : if (rv->schemaname)
1153 130076 : return NULL;
1154 :
1155 135372 : cte = scanNameSpaceForCTE(pstate, rv->relname, &levelsup);
1156 135372 : if (cte)
1157 4388 : nsitem = addRangeTableEntryForCTE(pstate, cte, levelsup, rv, true);
1158 130984 : else if (scanNameSpaceForENR(pstate, rv->relname))
1159 358 : nsitem = addRangeTableEntryForENR(pstate, rv, true);
1160 : else
1161 130626 : nsitem = NULL;
1162 :
1163 135364 : return nsitem;
1164 : }
1165 :
1166 : /*
1167 : * transformFromClauseItem -
1168 : * Transform a FROM-clause item, adding any required entries to the
1169 : * range table list being built in the ParseState, and return the
1170 : * transformed item ready to include in the joinlist. Also build a
1171 : * ParseNamespaceItem list describing the names exposed by this item.
1172 : * This routine can recurse to handle SQL92 JOIN expressions.
1173 : *
1174 : * The function return value is the node to add to the jointree (a
1175 : * RangeTblRef or JoinExpr). Additional output parameters are:
1176 : *
1177 : * *top_nsitem: receives the ParseNamespaceItem directly corresponding to the
1178 : * jointree item. (This is only used during internal recursion, not by
1179 : * outside callers.)
1180 : *
1181 : * *namespace: receives a List of ParseNamespaceItems for the RTEs exposed
1182 : * as table/column names by this item. (The lateral_only flags in these items
1183 : * are indeterminate and should be explicitly set by the caller before use.)
1184 : */
1185 : static Node *
1186 367867 : transformFromClauseItem(ParseState *pstate, Node *n,
1187 : ParseNamespaceItem **top_nsitem,
1188 : List **namespace)
1189 : {
1190 : /* Guard against stack overflow due to overly deep subtree */
1191 367867 : check_stack_depth();
1192 :
1193 367867 : if (IsA(n, RangeVar))
1194 : {
1195 : /* Plain relation reference, or perhaps a CTE reference */
1196 265448 : RangeVar *rv = (RangeVar *) n;
1197 : RangeTblRef *rtr;
1198 : ParseNamespaceItem *nsitem;
1199 :
1200 : /* Check if it's a CTE or tuplestore reference */
1201 265448 : nsitem = getNSItemForSpecialRelationTypes(pstate, rv);
1202 :
1203 : /* if not found above, must be a table reference */
1204 265440 : if (!nsitem)
1205 260702 : nsitem = transformTableEntry(pstate, rv);
1206 :
1207 265326 : *top_nsitem = nsitem;
1208 265326 : *namespace = list_make1(nsitem);
1209 265326 : rtr = makeNode(RangeTblRef);
1210 265326 : rtr->rtindex = nsitem->p_rtindex;
1211 265326 : return (Node *) rtr;
1212 : }
1213 102419 : else if (IsA(n, RangeSubselect))
1214 : {
1215 : /* sub-SELECT is like a plain relation */
1216 : RangeTblRef *rtr;
1217 : ParseNamespaceItem *nsitem;
1218 :
1219 13938 : nsitem = transformRangeSubselect(pstate, (RangeSubselect *) n);
1220 13862 : *top_nsitem = nsitem;
1221 13862 : *namespace = list_make1(nsitem);
1222 13862 : rtr = makeNode(RangeTblRef);
1223 13862 : rtr->rtindex = nsitem->p_rtindex;
1224 13862 : return (Node *) rtr;
1225 : }
1226 88481 : else if (IsA(n, RangeFunction))
1227 : {
1228 : /* function is like a plain relation */
1229 : RangeTblRef *rtr;
1230 : ParseNamespaceItem *nsitem;
1231 :
1232 30201 : nsitem = transformRangeFunction(pstate, (RangeFunction *) n);
1233 30054 : *top_nsitem = nsitem;
1234 30054 : *namespace = list_make1(nsitem);
1235 30054 : rtr = makeNode(RangeTblRef);
1236 30054 : rtr->rtindex = nsitem->p_rtindex;
1237 30054 : return (Node *) rtr;
1238 : }
1239 58280 : else if (IsA(n, RangeTableFunc) || IsA(n, JsonTable))
1240 : {
1241 : /* table function is like a plain relation */
1242 : RangeTblRef *rtr;
1243 : ParseNamespaceItem *nsitem;
1244 :
1245 494 : if (IsA(n, JsonTable))
1246 348 : nsitem = transformJsonTable(pstate, (JsonTable *) n);
1247 : else
1248 146 : nsitem = transformRangeTableFunc(pstate, (RangeTableFunc *) n);
1249 :
1250 434 : *top_nsitem = nsitem;
1251 434 : *namespace = list_make1(nsitem);
1252 434 : rtr = makeNode(RangeTblRef);
1253 434 : rtr->rtindex = nsitem->p_rtindex;
1254 434 : return (Node *) rtr;
1255 : }
1256 57786 : else if (IsA(n, RangeGraphTable))
1257 : {
1258 : RangeTblRef *rtr;
1259 : ParseNamespaceItem *nsitem;
1260 :
1261 574 : nsitem = transformRangeGraphTable(pstate, (RangeGraphTable *) n);
1262 494 : *top_nsitem = nsitem;
1263 494 : *namespace = list_make1(nsitem);
1264 494 : rtr = makeNode(RangeTblRef);
1265 494 : rtr->rtindex = nsitem->p_rtindex;
1266 494 : return (Node *) rtr;
1267 : }
1268 57212 : else if (IsA(n, RangeTableSample))
1269 : {
1270 : /* TABLESAMPLE clause (wrapping some other valid FROM node) */
1271 168 : RangeTableSample *rts = (RangeTableSample *) n;
1272 : Node *rel;
1273 : RangeTblEntry *rte;
1274 :
1275 : /* Recursively transform the contained relation */
1276 168 : rel = transformFromClauseItem(pstate, rts->relation,
1277 : top_nsitem, namespace);
1278 168 : rte = (*top_nsitem)->p_rte;
1279 : /* We only support this on plain relations and matviews */
1280 168 : if (rte->rtekind != RTE_RELATION ||
1281 164 : (rte->relkind != RELKIND_RELATION &&
1282 16 : rte->relkind != RELKIND_MATVIEW &&
1283 16 : rte->relkind != RELKIND_PARTITIONED_TABLE))
1284 8 : ereport(ERROR,
1285 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1286 : errmsg("TABLESAMPLE clause can only be applied to tables and materialized views"),
1287 : parser_errposition(pstate, exprLocation(rts->relation))));
1288 :
1289 : /* Transform TABLESAMPLE details and attach to the RTE */
1290 160 : rte->tablesample = transformRangeTableSample(pstate, rts);
1291 154 : return rel;
1292 : }
1293 57044 : else if (IsA(n, JoinExpr))
1294 : {
1295 : /* A newfangled join expression */
1296 57044 : JoinExpr *j = (JoinExpr *) n;
1297 : ParseNamespaceItem *nsitem;
1298 : ParseNamespaceItem *l_nsitem;
1299 : ParseNamespaceItem *r_nsitem;
1300 : List *l_namespace,
1301 : *r_namespace,
1302 : *my_namespace,
1303 : *l_colnames,
1304 : *r_colnames,
1305 : *res_colnames,
1306 : *l_colnos,
1307 : *r_colnos,
1308 : *res_colvars;
1309 : ParseNamespaceColumn *l_nscolumns,
1310 : *r_nscolumns,
1311 : *res_nscolumns;
1312 : int res_colindex;
1313 : bool lateral_ok;
1314 : int sv_namespace_length;
1315 : int k;
1316 :
1317 : /*
1318 : * Recursively process the left subtree, then the right. We must do
1319 : * it in this order for correct visibility of LATERAL references.
1320 : */
1321 57044 : j->larg = transformFromClauseItem(pstate, j->larg,
1322 : &l_nsitem,
1323 : &l_namespace);
1324 :
1325 : /*
1326 : * Make the left-side RTEs available for LATERAL access within the
1327 : * right side, by temporarily adding them to the pstate's namespace
1328 : * list. Per SQL:2008, if the join type is not INNER or LEFT then the
1329 : * left-side names must still be exposed, but it's an error to
1330 : * reference them. (Stupid design, but that's what it says.) Hence,
1331 : * we always push them into the namespace, but mark them as not
1332 : * lateral_ok if the jointype is wrong.
1333 : *
1334 : * Notice that we don't require the merged namespace list to be
1335 : * conflict-free. See the comments for scanNameSpaceForRefname().
1336 : */
1337 57044 : lateral_ok = (j->jointype == JOIN_INNER || j->jointype == JOIN_LEFT);
1338 57044 : setNamespaceLateralState(l_namespace, true, lateral_ok);
1339 :
1340 57044 : sv_namespace_length = list_length(pstate->p_namespace);
1341 57044 : pstate->p_namespace = list_concat(pstate->p_namespace, l_namespace);
1342 :
1343 : /* And now we can process the RHS */
1344 57044 : j->rarg = transformFromClauseItem(pstate, j->rarg,
1345 : &r_nsitem,
1346 : &r_namespace);
1347 :
1348 : /* Remove the left-side RTEs from the namespace list again */
1349 57020 : pstate->p_namespace = list_truncate(pstate->p_namespace,
1350 : sv_namespace_length);
1351 :
1352 : /*
1353 : * Check for conflicting refnames in left and right subtrees. Must do
1354 : * this because higher levels will assume I hand back a self-
1355 : * consistent namespace list.
1356 : */
1357 57020 : checkNameSpaceConflicts(pstate, l_namespace, r_namespace);
1358 :
1359 : /*
1360 : * Generate combined namespace info for possible use below.
1361 : */
1362 57020 : my_namespace = list_concat(l_namespace, r_namespace);
1363 :
1364 : /*
1365 : * We'll work from the nscolumns data and eref alias column names for
1366 : * each of the input nsitems. Note that these include dropped
1367 : * columns, which is helpful because we can keep track of physical
1368 : * input column numbers more easily.
1369 : */
1370 57020 : l_nscolumns = l_nsitem->p_nscolumns;
1371 57020 : l_colnames = l_nsitem->p_names->colnames;
1372 57020 : r_nscolumns = r_nsitem->p_nscolumns;
1373 57020 : r_colnames = r_nsitem->p_names->colnames;
1374 :
1375 : /*
1376 : * Natural join does not explicitly specify columns; must generate
1377 : * columns to join. Need to run through the list of columns from each
1378 : * table or join result and match up the column names. Use the first
1379 : * table, and check every column in the second table for a match.
1380 : * (We'll check that the matches were unique later on.) The result of
1381 : * this step is a list of column names just like an explicitly-written
1382 : * USING list.
1383 : */
1384 57020 : if (j->isNatural)
1385 : {
1386 176 : List *rlist = NIL;
1387 : ListCell *lx,
1388 : *rx;
1389 :
1390 : Assert(j->usingClause == NIL); /* shouldn't have USING() too */
1391 :
1392 780 : foreach(lx, l_colnames)
1393 : {
1394 604 : char *l_colname = strVal(lfirst(lx));
1395 604 : String *m_name = NULL;
1396 :
1397 604 : if (l_colname[0] == '\0')
1398 8 : continue; /* ignore dropped columns */
1399 :
1400 1648 : foreach(rx, r_colnames)
1401 : {
1402 1272 : char *r_colname = strVal(lfirst(rx));
1403 :
1404 1272 : if (strcmp(l_colname, r_colname) == 0)
1405 : {
1406 220 : m_name = makeString(l_colname);
1407 220 : break;
1408 : }
1409 : }
1410 :
1411 : /* matched a right column? then keep as join column... */
1412 596 : if (m_name != NULL)
1413 220 : rlist = lappend(rlist, m_name);
1414 : }
1415 :
1416 176 : j->usingClause = rlist;
1417 : }
1418 :
1419 : /*
1420 : * If a USING clause alias was specified, save the USING columns as
1421 : * its column list.
1422 : */
1423 57020 : if (j->join_using_alias)
1424 56 : j->join_using_alias->colnames = j->usingClause;
1425 :
1426 : /*
1427 : * Now transform the join qualifications, if any.
1428 : */
1429 57020 : l_colnos = NIL;
1430 57020 : r_colnos = NIL;
1431 57020 : res_colnames = NIL;
1432 57020 : res_colvars = NIL;
1433 :
1434 : /* this may be larger than needed, but it's not worth being exact */
1435 : res_nscolumns = (ParseNamespaceColumn *)
1436 57020 : palloc0((list_length(l_colnames) + list_length(r_colnames)) *
1437 : sizeof(ParseNamespaceColumn));
1438 57020 : res_colindex = 0;
1439 :
1440 57020 : if (j->usingClause)
1441 : {
1442 : /*
1443 : * JOIN/USING (or NATURAL JOIN, as transformed above). Transform
1444 : * the list into an explicit ON-condition.
1445 : */
1446 1028 : List *ucols = j->usingClause;
1447 1028 : List *l_usingvars = NIL;
1448 1028 : List *r_usingvars = NIL;
1449 : ListCell *ucol;
1450 :
1451 : Assert(j->quals == NULL); /* shouldn't have ON() too */
1452 :
1453 2211 : foreach(ucol, ucols)
1454 : {
1455 1183 : char *u_colname = strVal(lfirst(ucol));
1456 : ListCell *col;
1457 : int ndx;
1458 1183 : int l_index = -1;
1459 1183 : int r_index = -1;
1460 : Var *l_colvar,
1461 : *r_colvar;
1462 :
1463 : Assert(u_colname[0] != '\0');
1464 :
1465 : /* Check for USING(foo,foo) */
1466 1369 : foreach(col, res_colnames)
1467 : {
1468 186 : char *res_colname = strVal(lfirst(col));
1469 :
1470 186 : if (strcmp(res_colname, u_colname) == 0)
1471 0 : ereport(ERROR,
1472 : (errcode(ERRCODE_DUPLICATE_COLUMN),
1473 : errmsg("column name \"%s\" appears more than once in USING clause",
1474 : u_colname)));
1475 : }
1476 :
1477 : /* Find it in left input */
1478 1183 : ndx = 0;
1479 5736 : foreach(col, l_colnames)
1480 : {
1481 4553 : char *l_colname = strVal(lfirst(col));
1482 :
1483 4553 : if (strcmp(l_colname, u_colname) == 0)
1484 : {
1485 1183 : if (l_index >= 0)
1486 0 : ereport(ERROR,
1487 : (errcode(ERRCODE_AMBIGUOUS_COLUMN),
1488 : errmsg("common column name \"%s\" appears more than once in left table",
1489 : u_colname)));
1490 1183 : l_index = ndx;
1491 : }
1492 4553 : ndx++;
1493 : }
1494 1183 : if (l_index < 0)
1495 0 : ereport(ERROR,
1496 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1497 : errmsg("column \"%s\" specified in USING clause does not exist in left table",
1498 : u_colname)));
1499 1183 : l_colnos = lappend_int(l_colnos, l_index + 1);
1500 :
1501 : /* Find it in right input */
1502 1183 : ndx = 0;
1503 5679 : foreach(col, r_colnames)
1504 : {
1505 4496 : char *r_colname = strVal(lfirst(col));
1506 :
1507 4496 : if (strcmp(r_colname, u_colname) == 0)
1508 : {
1509 1183 : if (r_index >= 0)
1510 0 : ereport(ERROR,
1511 : (errcode(ERRCODE_AMBIGUOUS_COLUMN),
1512 : errmsg("common column name \"%s\" appears more than once in right table",
1513 : u_colname)));
1514 1183 : r_index = ndx;
1515 : }
1516 4496 : ndx++;
1517 : }
1518 1183 : if (r_index < 0)
1519 0 : ereport(ERROR,
1520 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1521 : errmsg("column \"%s\" specified in USING clause does not exist in right table",
1522 : u_colname)));
1523 1183 : r_colnos = lappend_int(r_colnos, r_index + 1);
1524 :
1525 : /* Build Vars to use in the generated JOIN ON clause */
1526 1183 : l_colvar = buildVarFromNSColumn(pstate, l_nscolumns + l_index);
1527 1183 : l_usingvars = lappend(l_usingvars, l_colvar);
1528 1183 : r_colvar = buildVarFromNSColumn(pstate, r_nscolumns + r_index);
1529 1183 : r_usingvars = lappend(r_usingvars, r_colvar);
1530 :
1531 : /*
1532 : * While we're here, add column names to the res_colnames
1533 : * list. It's a bit ugly to do this here while the
1534 : * corresponding res_colvars entries are not made till later,
1535 : * but doing this later would require an additional traversal
1536 : * of the usingClause list.
1537 : */
1538 1183 : res_colnames = lappend(res_colnames, lfirst(ucol));
1539 : }
1540 :
1541 : /* Construct the generated JOIN ON clause */
1542 1028 : j->quals = transformJoinUsingClause(pstate,
1543 : l_usingvars,
1544 : r_usingvars);
1545 : }
1546 55992 : else if (j->quals)
1547 : {
1548 : /* User-written ON-condition; transform it */
1549 55641 : j->quals = transformJoinOnClause(pstate, j, my_namespace);
1550 : }
1551 : else
1552 : {
1553 : /* CROSS JOIN: no quals */
1554 : }
1555 :
1556 : /*
1557 : * If this is an outer join, now mark the appropriate child RTEs as
1558 : * being nulled by this join. We have finished processing the child
1559 : * join expressions as well as the current join's quals, which deal in
1560 : * non-nulled input columns. All future references to those RTEs will
1561 : * see possibly-nulled values, and we should mark generated Vars to
1562 : * account for that. In particular, the join alias Vars that we're
1563 : * about to build should reflect the nulling effects of this join.
1564 : *
1565 : * A difficulty with doing this is that we need the join's RT index,
1566 : * which we don't officially have yet. However, no other RTE can get
1567 : * made between here and the addRangeTableEntryForJoin call, so we can
1568 : * predict what the assignment will be. (Alternatively, we could call
1569 : * addRangeTableEntryForJoin before we have all the data computed, but
1570 : * this seems less ugly.)
1571 : */
1572 57008 : j->rtindex = list_length(pstate->p_rtable) + 1;
1573 :
1574 57008 : switch (j->jointype)
1575 : {
1576 29079 : case JOIN_INNER:
1577 29079 : break;
1578 27010 : case JOIN_LEFT:
1579 27010 : markRelsAsNulledBy(pstate, j->rarg, j->rtindex);
1580 27010 : break;
1581 679 : case JOIN_FULL:
1582 679 : markRelsAsNulledBy(pstate, j->larg, j->rtindex);
1583 679 : markRelsAsNulledBy(pstate, j->rarg, j->rtindex);
1584 679 : break;
1585 240 : case JOIN_RIGHT:
1586 240 : markRelsAsNulledBy(pstate, j->larg, j->rtindex);
1587 240 : break;
1588 0 : default:
1589 : /* shouldn't see any other types here */
1590 0 : elog(ERROR, "unrecognized join type: %d",
1591 : (int) j->jointype);
1592 : break;
1593 : }
1594 :
1595 : /*
1596 : * Now we can construct join alias expressions for the USING columns.
1597 : */
1598 57008 : if (j->usingClause)
1599 : {
1600 : ListCell *lc1,
1601 : *lc2;
1602 :
1603 : /* Scan the colnos lists to recover info from the previous loop */
1604 2211 : forboth(lc1, l_colnos, lc2, r_colnos)
1605 : {
1606 1183 : int l_index = lfirst_int(lc1) - 1;
1607 1183 : int r_index = lfirst_int(lc2) - 1;
1608 : Var *l_colvar,
1609 : *r_colvar;
1610 : Node *u_colvar;
1611 : ParseNamespaceColumn *res_nscolumn;
1612 :
1613 : /*
1614 : * Note we re-build these Vars: they might have different
1615 : * varnullingrels than the ones made in the previous loop.
1616 : */
1617 1183 : l_colvar = buildVarFromNSColumn(pstate, l_nscolumns + l_index);
1618 1183 : r_colvar = buildVarFromNSColumn(pstate, r_nscolumns + r_index);
1619 :
1620 : /* Construct the join alias Var for this column */
1621 1183 : u_colvar = buildMergedJoinVar(pstate,
1622 : j->jointype,
1623 : l_colvar,
1624 : r_colvar);
1625 1183 : res_colvars = lappend(res_colvars, u_colvar);
1626 :
1627 : /* Construct column's res_nscolumns[] entry */
1628 1183 : res_nscolumn = res_nscolumns + res_colindex;
1629 1183 : res_colindex++;
1630 1183 : if (u_colvar == (Node *) l_colvar)
1631 : {
1632 : /* Merged column is equivalent to left input */
1633 858 : *res_nscolumn = l_nscolumns[l_index];
1634 : }
1635 325 : else if (u_colvar == (Node *) r_colvar)
1636 : {
1637 : /* Merged column is equivalent to right input */
1638 28 : *res_nscolumn = r_nscolumns[r_index];
1639 : }
1640 : else
1641 : {
1642 : /*
1643 : * Merged column is not semantically equivalent to either
1644 : * input, so it needs to be referenced as the join output
1645 : * column.
1646 : */
1647 297 : res_nscolumn->p_varno = j->rtindex;
1648 297 : res_nscolumn->p_varattno = res_colindex;
1649 297 : res_nscolumn->p_vartype = exprType(u_colvar);
1650 297 : res_nscolumn->p_vartypmod = exprTypmod(u_colvar);
1651 297 : res_nscolumn->p_varcollid = exprCollation(u_colvar);
1652 297 : res_nscolumn->p_varnosyn = j->rtindex;
1653 297 : res_nscolumn->p_varattnosyn = res_colindex;
1654 : }
1655 : }
1656 : }
1657 :
1658 : /* Add remaining columns from each side to the output columns */
1659 57008 : res_colindex +=
1660 57008 : extractRemainingColumns(pstate,
1661 : l_nscolumns, l_colnames, &l_colnos,
1662 : &res_colnames, &res_colvars,
1663 57008 : res_nscolumns + res_colindex);
1664 57008 : res_colindex +=
1665 57008 : extractRemainingColumns(pstate,
1666 : r_nscolumns, r_colnames, &r_colnos,
1667 : &res_colnames, &res_colvars,
1668 57008 : res_nscolumns + res_colindex);
1669 :
1670 : /* If join has an alias, it syntactically hides all inputs */
1671 57008 : if (j->alias)
1672 : {
1673 668 : for (k = 0; k < res_colindex; k++)
1674 : {
1675 548 : ParseNamespaceColumn *nscol = res_nscolumns + k;
1676 :
1677 548 : nscol->p_varnosyn = j->rtindex;
1678 548 : nscol->p_varattnosyn = k + 1;
1679 : }
1680 : }
1681 :
1682 : /*
1683 : * Now build an RTE and nsitem for the result of the join.
1684 : */
1685 57008 : nsitem = addRangeTableEntryForJoin(pstate,
1686 : res_colnames,
1687 : res_nscolumns,
1688 : j->jointype,
1689 57008 : list_length(j->usingClause),
1690 : res_colvars,
1691 : l_colnos,
1692 : r_colnos,
1693 : j->join_using_alias,
1694 : j->alias,
1695 : true);
1696 :
1697 : /* Verify that we correctly predicted the join's RT index */
1698 : Assert(j->rtindex == nsitem->p_rtindex);
1699 : /* Cross-check number of columns, too */
1700 : Assert(res_colindex == list_length(nsitem->p_names->colnames));
1701 :
1702 : /*
1703 : * Save a link to the JoinExpr in the proper element of p_joinexprs.
1704 : * Since we maintain that list lazily, it may be necessary to fill in
1705 : * empty entries before we can add the JoinExpr in the right place.
1706 : */
1707 150025 : for (k = list_length(pstate->p_joinexprs) + 1; k < j->rtindex; k++)
1708 93021 : pstate->p_joinexprs = lappend(pstate->p_joinexprs, NULL);
1709 57004 : pstate->p_joinexprs = lappend(pstate->p_joinexprs, j);
1710 : Assert(list_length(pstate->p_joinexprs) == j->rtindex);
1711 :
1712 : /*
1713 : * If the join has a USING alias, build a ParseNamespaceItem for that
1714 : * and add it to the list of nsitems in the join's input.
1715 : */
1716 57004 : if (j->join_using_alias)
1717 : {
1718 : ParseNamespaceItem *jnsitem;
1719 :
1720 56 : jnsitem = palloc_object(ParseNamespaceItem);
1721 56 : jnsitem->p_names = j->join_using_alias;
1722 56 : jnsitem->p_rte = nsitem->p_rte;
1723 56 : jnsitem->p_rtindex = nsitem->p_rtindex;
1724 56 : jnsitem->p_perminfo = NULL;
1725 : /* no need to copy the first N columns, just use res_nscolumns */
1726 56 : jnsitem->p_nscolumns = res_nscolumns;
1727 : /* set default visibility flags; might get changed later */
1728 56 : jnsitem->p_rel_visible = true;
1729 56 : jnsitem->p_cols_visible = true;
1730 56 : jnsitem->p_lateral_only = false;
1731 56 : jnsitem->p_lateral_ok = true;
1732 56 : jnsitem->p_returning_type = VAR_RETURNING_DEFAULT;
1733 : /* Per SQL, we must check for alias conflicts */
1734 56 : checkNameSpaceConflicts(pstate, list_make1(jnsitem), my_namespace);
1735 52 : my_namespace = lappend(my_namespace, jnsitem);
1736 : }
1737 :
1738 : /*
1739 : * Prepare returned namespace list. If the JOIN has an alias then it
1740 : * hides the contained RTEs completely; otherwise, the contained RTEs
1741 : * are still visible as table names, but are not visible for
1742 : * unqualified column-name access.
1743 : *
1744 : * Note: if there are nested alias-less JOINs, the lower-level ones
1745 : * will remain in the list although they have neither p_rel_visible
1746 : * nor p_cols_visible set. We could delete such list items, but it's
1747 : * unclear that it's worth expending cycles to do so.
1748 : */
1749 57000 : if (j->alias != NULL)
1750 116 : my_namespace = NIL;
1751 : else
1752 56884 : setNamespaceColumnVisibility(my_namespace, false);
1753 :
1754 : /*
1755 : * The join RTE itself is always made visible for unqualified column
1756 : * names. It's visible as a relation name only if it has an alias.
1757 : */
1758 57000 : nsitem->p_rel_visible = (j->alias != NULL);
1759 57000 : nsitem->p_cols_visible = true;
1760 57000 : nsitem->p_lateral_only = false;
1761 57000 : nsitem->p_lateral_ok = true;
1762 :
1763 57000 : *top_nsitem = nsitem;
1764 57000 : *namespace = lappend(my_namespace, nsitem);
1765 :
1766 57000 : return (Node *) j;
1767 : }
1768 : else
1769 0 : elog(ERROR, "unrecognized node type: %d", (int) nodeTag(n));
1770 : return NULL; /* can't get here, keep compiler quiet */
1771 : }
1772 :
1773 : /*
1774 : * buildVarFromNSColumn -
1775 : * build a Var node using ParseNamespaceColumn data
1776 : *
1777 : * This is used to construct joinaliasvars entries.
1778 : * We can assume varlevelsup should be 0, and no location is specified.
1779 : * Note also that no column SELECT privilege is requested here; that would
1780 : * happen only if the column is actually referenced in the query.
1781 : */
1782 : static Var *
1783 2203382 : buildVarFromNSColumn(ParseState *pstate, ParseNamespaceColumn *nscol)
1784 : {
1785 : Var *var;
1786 :
1787 : Assert(nscol->p_varno > 0); /* i.e., not deleted column */
1788 2203382 : var = makeVar(nscol->p_varno,
1789 2203382 : nscol->p_varattno,
1790 : nscol->p_vartype,
1791 : nscol->p_vartypmod,
1792 : nscol->p_varcollid,
1793 : 0);
1794 : /* makeVar doesn't offer parameters for these, so set by hand: */
1795 2203382 : var->varreturningtype = nscol->p_varreturningtype;
1796 2203382 : var->varnosyn = nscol->p_varnosyn;
1797 2203382 : var->varattnosyn = nscol->p_varattnosyn;
1798 :
1799 : /* ... and update varnullingrels */
1800 2203382 : markNullableIfNeeded(pstate, var);
1801 :
1802 2203382 : return var;
1803 : }
1804 :
1805 : /*
1806 : * buildMergedJoinVar -
1807 : * generate a suitable replacement expression for a merged join column
1808 : */
1809 : static Node *
1810 1183 : buildMergedJoinVar(ParseState *pstate, JoinType jointype,
1811 : Var *l_colvar, Var *r_colvar)
1812 : {
1813 : Oid outcoltype;
1814 : int32 outcoltypmod;
1815 : Node *l_node,
1816 : *r_node,
1817 : *res_node;
1818 :
1819 1183 : outcoltype = select_common_type(pstate,
1820 : list_make2(l_colvar, r_colvar),
1821 : "JOIN/USING",
1822 : NULL);
1823 1183 : outcoltypmod = select_common_typmod(pstate,
1824 : list_make2(l_colvar, r_colvar),
1825 : outcoltype);
1826 :
1827 : /*
1828 : * Insert coercion functions if needed. Note that a difference in typmod
1829 : * can only happen if input has typmod but outcoltypmod is -1. In that
1830 : * case we insert a RelabelType to clearly mark that result's typmod is
1831 : * not same as input. We never need coerce_type_typmod.
1832 : */
1833 1183 : if (l_colvar->vartype != outcoltype)
1834 60 : l_node = coerce_type(pstate, (Node *) l_colvar, l_colvar->vartype,
1835 : outcoltype, outcoltypmod,
1836 : COERCION_IMPLICIT, COERCE_IMPLICIT_CAST, -1);
1837 1123 : else if (l_colvar->vartypmod != outcoltypmod)
1838 0 : l_node = (Node *) makeRelabelType((Expr *) l_colvar,
1839 : outcoltype, outcoltypmod,
1840 : InvalidOid, /* fixed below */
1841 : COERCE_IMPLICIT_CAST);
1842 : else
1843 1123 : l_node = (Node *) l_colvar;
1844 :
1845 1183 : if (r_colvar->vartype != outcoltype)
1846 20 : r_node = coerce_type(pstate, (Node *) r_colvar, r_colvar->vartype,
1847 : outcoltype, outcoltypmod,
1848 : COERCION_IMPLICIT, COERCE_IMPLICIT_CAST, -1);
1849 1163 : else if (r_colvar->vartypmod != outcoltypmod)
1850 0 : r_node = (Node *) makeRelabelType((Expr *) r_colvar,
1851 : outcoltype, outcoltypmod,
1852 : InvalidOid, /* fixed below */
1853 : COERCE_IMPLICIT_CAST);
1854 : else
1855 1163 : r_node = (Node *) r_colvar;
1856 :
1857 : /*
1858 : * Choose what to emit
1859 : */
1860 1183 : switch (jointype)
1861 : {
1862 766 : case JOIN_INNER:
1863 :
1864 : /*
1865 : * We can use either var; prefer non-coerced one if available.
1866 : */
1867 766 : if (IsA(l_node, Var))
1868 746 : res_node = l_node;
1869 20 : else if (IsA(r_node, Var))
1870 20 : res_node = r_node;
1871 : else
1872 0 : res_node = l_node;
1873 766 : break;
1874 152 : case JOIN_LEFT:
1875 : /* Always use left var */
1876 152 : res_node = l_node;
1877 152 : break;
1878 8 : case JOIN_RIGHT:
1879 : /* Always use right var */
1880 8 : res_node = r_node;
1881 8 : break;
1882 257 : case JOIN_FULL:
1883 : {
1884 : /*
1885 : * Here we must build a COALESCE expression to ensure that the
1886 : * join output is non-null if either input is.
1887 : */
1888 257 : CoalesceExpr *c = makeNode(CoalesceExpr);
1889 :
1890 257 : c->coalescetype = outcoltype;
1891 : /* coalescecollid will get set below */
1892 257 : c->args = list_make2(l_node, r_node);
1893 257 : c->location = -1;
1894 257 : res_node = (Node *) c;
1895 257 : break;
1896 : }
1897 0 : default:
1898 0 : elog(ERROR, "unrecognized join type: %d", (int) jointype);
1899 : res_node = NULL; /* keep compiler quiet */
1900 : break;
1901 : }
1902 :
1903 : /*
1904 : * Apply assign_expr_collations to fix up the collation info in the
1905 : * coercion and CoalesceExpr nodes, if we made any. This must be done now
1906 : * so that the join node's alias vars show correct collation info.
1907 : */
1908 1183 : assign_expr_collations(pstate, res_node);
1909 :
1910 1183 : return res_node;
1911 : }
1912 :
1913 : /*
1914 : * markRelsAsNulledBy -
1915 : * Mark the given jointree node and its children as nulled by join jindex
1916 : */
1917 : static void
1918 31010 : markRelsAsNulledBy(ParseState *pstate, Node *n, int jindex)
1919 : {
1920 : int varno;
1921 : ListCell *lc;
1922 :
1923 : /* Note: we can't see FromExpr here */
1924 31010 : if (IsA(n, RangeTblRef))
1925 : {
1926 29809 : varno = ((RangeTblRef *) n)->rtindex;
1927 : }
1928 1201 : else if (IsA(n, JoinExpr))
1929 : {
1930 1201 : JoinExpr *j = (JoinExpr *) n;
1931 :
1932 : /* recurse to children */
1933 1201 : markRelsAsNulledBy(pstate, j->larg, jindex);
1934 1201 : markRelsAsNulledBy(pstate, j->rarg, jindex);
1935 1201 : varno = j->rtindex;
1936 : }
1937 : else
1938 : {
1939 0 : elog(ERROR, "unrecognized node type: %d", (int) nodeTag(n));
1940 : varno = 0; /* keep compiler quiet */
1941 : }
1942 :
1943 : /*
1944 : * Now add jindex to the p_nullingrels set for relation varno. Since we
1945 : * maintain the p_nullingrels list lazily, we might need to extend it to
1946 : * make the varno'th entry exist.
1947 : */
1948 100896 : while (list_length(pstate->p_nullingrels) < varno)
1949 69886 : pstate->p_nullingrels = lappend(pstate->p_nullingrels, NULL);
1950 31010 : lc = list_nth_cell(pstate->p_nullingrels, varno - 1);
1951 31010 : lfirst(lc) = bms_add_member((Bitmapset *) lfirst(lc), jindex);
1952 31010 : }
1953 :
1954 : /*
1955 : * setNamespaceColumnVisibility -
1956 : * Convenience subroutine to update cols_visible flags in a namespace list.
1957 : */
1958 : static void
1959 56884 : setNamespaceColumnVisibility(List *namespace, bool cols_visible)
1960 : {
1961 : ListCell *lc;
1962 :
1963 240570 : foreach(lc, namespace)
1964 : {
1965 183686 : ParseNamespaceItem *nsitem = (ParseNamespaceItem *) lfirst(lc);
1966 :
1967 183686 : nsitem->p_cols_visible = cols_visible;
1968 : }
1969 56884 : }
1970 :
1971 : /*
1972 : * setNamespaceLateralState -
1973 : * Convenience subroutine to update LATERAL flags in a namespace list.
1974 : */
1975 : static void
1976 687007 : setNamespaceLateralState(List *namespace, bool lateral_only, bool lateral_ok)
1977 : {
1978 : ListCell *lc;
1979 :
1980 1738675 : foreach(lc, namespace)
1981 : {
1982 1051668 : ParseNamespaceItem *nsitem = (ParseNamespaceItem *) lfirst(lc);
1983 :
1984 1051668 : nsitem->p_lateral_only = lateral_only;
1985 1051668 : nsitem->p_lateral_ok = lateral_ok;
1986 : }
1987 687007 : }
1988 :
1989 :
1990 : /*
1991 : * transformWhereClause -
1992 : * Transform the qualification and make sure it is of type boolean.
1993 : * Used for WHERE and allied clauses.
1994 : *
1995 : * constructName does not affect the semantics, but is used in error messages
1996 : */
1997 : Node *
1998 681736 : transformWhereClause(ParseState *pstate, Node *clause,
1999 : ParseExprKind exprKind, const char *constructName)
2000 : {
2001 : Node *qual;
2002 :
2003 681736 : if (clause == NULL)
2004 471699 : return NULL;
2005 :
2006 210037 : qual = transformExpr(pstate, clause, exprKind);
2007 :
2008 209898 : qual = coerce_to_boolean(pstate, qual, constructName);
2009 :
2010 209894 : return qual;
2011 : }
2012 :
2013 :
2014 : /*
2015 : * transformLimitClause -
2016 : * Transform the expression and make sure it is of type bigint.
2017 : * Used for LIMIT and allied clauses.
2018 : *
2019 : * Note: as of Postgres 8.2, LIMIT expressions are expected to yield int8,
2020 : * rather than int4 as before.
2021 : *
2022 : * constructName does not affect the semantics, but is used in error messages
2023 : */
2024 : Node *
2025 635816 : transformLimitClause(ParseState *pstate, Node *clause,
2026 : ParseExprKind exprKind, const char *constructName,
2027 : LimitOption limitOption)
2028 : {
2029 : Node *qual;
2030 :
2031 635816 : if (clause == NULL)
2032 632345 : return NULL;
2033 :
2034 3471 : qual = transformExpr(pstate, clause, exprKind);
2035 :
2036 3467 : qual = coerce_to_specific_type(pstate, qual, INT8OID, constructName);
2037 :
2038 : /* LIMIT can't refer to any variables of the current query */
2039 3467 : checkExprIsVarFree(pstate, qual, constructName);
2040 :
2041 : /*
2042 : * Don't allow NULLs in FETCH FIRST .. WITH TIES. This test is ugly and
2043 : * extremely simplistic, in that you can pass a NULL anyway by hiding it
2044 : * inside an expression -- but this protects ruleutils against emitting an
2045 : * unadorned NULL that's not accepted back by the grammar.
2046 : */
2047 3467 : if (exprKind == EXPR_KIND_LIMIT && limitOption == LIMIT_OPTION_WITH_TIES &&
2048 38 : IsA(clause, A_Const) && castNode(A_Const, clause)->isnull)
2049 4 : ereport(ERROR,
2050 : (errcode(ERRCODE_INVALID_ROW_COUNT_IN_LIMIT_CLAUSE),
2051 : errmsg("row count cannot be null in FETCH FIRST ... WITH TIES clause")));
2052 :
2053 3463 : return qual;
2054 : }
2055 :
2056 : /*
2057 : * checkExprIsVarFree
2058 : * Check that given expr has no Vars of the current query level
2059 : * (aggregates and window functions should have been rejected already).
2060 : *
2061 : * This is used to check expressions that have to have a consistent value
2062 : * across all rows of the query, such as a LIMIT. Arguably it should reject
2063 : * volatile functions, too, but we don't do that --- whatever value the
2064 : * function gives on first execution is what you get.
2065 : *
2066 : * constructName does not affect the semantics, but is used in error messages
2067 : */
2068 : static void
2069 4765 : checkExprIsVarFree(ParseState *pstate, Node *n, const char *constructName)
2070 : {
2071 4765 : if (contain_vars_of_level(n, 0))
2072 : {
2073 4 : ereport(ERROR,
2074 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
2075 : /* translator: %s is name of a SQL construct, eg LIMIT */
2076 : errmsg("argument of %s must not contain variables",
2077 : constructName),
2078 : parser_errposition(pstate,
2079 : locate_var_of_level(n, 0))));
2080 : }
2081 4761 : }
2082 :
2083 :
2084 : /*
2085 : * checkTargetlistEntrySQL92 -
2086 : * Validate a targetlist entry found by findTargetlistEntrySQL92
2087 : *
2088 : * When we select a pre-existing tlist entry as a result of syntax such
2089 : * as "GROUP BY 1", we have to make sure it is acceptable for use in the
2090 : * indicated clause type; transformExpr() will have treated it as a regular
2091 : * targetlist item.
2092 : */
2093 : static void
2094 51811 : checkTargetlistEntrySQL92(ParseState *pstate, TargetEntry *tle,
2095 : ParseExprKind exprKind)
2096 : {
2097 51811 : switch (exprKind)
2098 : {
2099 518 : case EXPR_KIND_GROUP_BY:
2100 : /* reject aggregates and window functions */
2101 928 : if (pstate->p_hasAggs &&
2102 410 : contain_aggs_of_level((Node *) tle->expr, 0))
2103 0 : ereport(ERROR,
2104 : (errcode(ERRCODE_GROUPING_ERROR),
2105 : /* translator: %s is name of a SQL construct, eg GROUP BY */
2106 : errmsg("aggregate functions are not allowed in %s",
2107 : ParseExprKindName(exprKind)),
2108 : parser_errposition(pstate,
2109 : locate_agg_of_level((Node *) tle->expr, 0))));
2110 526 : if (pstate->p_hasWindowFuncs &&
2111 8 : contain_windowfuncs((Node *) tle->expr))
2112 4 : ereport(ERROR,
2113 : (errcode(ERRCODE_WINDOWING_ERROR),
2114 : /* translator: %s is name of a SQL construct, eg GROUP BY */
2115 : errmsg("window functions are not allowed in %s",
2116 : ParseExprKindName(exprKind)),
2117 : parser_errposition(pstate,
2118 : locate_windowfunc((Node *) tle->expr))));
2119 514 : break;
2120 51125 : case EXPR_KIND_ORDER_BY:
2121 : /* no extra checks needed */
2122 51125 : break;
2123 168 : case EXPR_KIND_DISTINCT_ON:
2124 : /* no extra checks needed */
2125 168 : break;
2126 0 : default:
2127 0 : elog(ERROR, "unexpected exprKind in checkTargetlistEntrySQL92");
2128 : break;
2129 : }
2130 51807 : }
2131 :
2132 : /*
2133 : * findTargetlistEntrySQL92 -
2134 : * Returns the targetlist entry matching the given (untransformed) node.
2135 : * If no matching entry exists, one is created and appended to the target
2136 : * list as a "resjunk" node.
2137 : *
2138 : * This function supports the old SQL92 ORDER BY interpretation, where the
2139 : * expression is an output column name or number. If we fail to find a
2140 : * match of that sort, we fall through to the SQL99 rules. For historical
2141 : * reasons, Postgres also allows this interpretation for GROUP BY, though
2142 : * the standard never did. However, for GROUP BY we prefer a SQL99 match.
2143 : * This function is *not* used for WINDOW definitions.
2144 : *
2145 : * node the ORDER BY, GROUP BY, or DISTINCT ON expression to be matched
2146 : * tlist the target list (passed by reference so we can append to it)
2147 : * exprKind identifies clause type being processed
2148 : */
2149 : static TargetEntry *
2150 78779 : findTargetlistEntrySQL92(ParseState *pstate, Node *node, List **tlist,
2151 : ParseExprKind exprKind)
2152 : {
2153 : ListCell *tl;
2154 :
2155 : /*----------
2156 : * Handle two special cases as mandated by the SQL92 spec:
2157 : *
2158 : * 1. Bare ColumnName (no qualifier or subscripts)
2159 : * For a bare identifier, we search for a matching column name
2160 : * in the existing target list. Multiple matches are an error
2161 : * unless they refer to identical values; for example,
2162 : * we allow SELECT a, a FROM table ORDER BY a
2163 : * but not SELECT a AS b, b FROM table ORDER BY b
2164 : * If no match is found, we fall through and treat the identifier
2165 : * as an expression.
2166 : * For GROUP BY, it is incorrect to match the grouping item against
2167 : * targetlist entries: according to SQL92, an identifier in GROUP BY
2168 : * is a reference to a column name exposed by FROM, not to a target
2169 : * list column. However, many implementations (including pre-7.0
2170 : * PostgreSQL) accept this anyway. So for GROUP BY, we look first
2171 : * to see if the identifier matches any FROM column name, and only
2172 : * try for a targetlist name if it doesn't. This ensures that we
2173 : * adhere to the spec in the case where the name could be both.
2174 : * DISTINCT ON isn't in the standard, so we can do what we like there;
2175 : * we choose to make it work like ORDER BY, on the rather flimsy
2176 : * grounds that ordinary DISTINCT works on targetlist entries.
2177 : *
2178 : * 2. IntegerConstant
2179 : * This means to use the n'th item in the existing target list.
2180 : * Note that it would make no sense to order/group/distinct by an
2181 : * actual constant, so this does not create a conflict with SQL99.
2182 : * GROUP BY column-number is not allowed by SQL92, but since
2183 : * the standard has no other behavior defined for this syntax,
2184 : * we may as well accept this common extension.
2185 : *
2186 : * Note that pre-existing resjunk targets must not be used in either case,
2187 : * since the user didn't write them in his SELECT list.
2188 : *
2189 : * If neither special case applies, fall through to treat the item as
2190 : * an expression per SQL99.
2191 : *----------
2192 : */
2193 123971 : if (IsA(node, ColumnRef) &&
2194 45192 : list_length(((ColumnRef *) node)->fields) == 1 &&
2195 32599 : IsA(linitial(((ColumnRef *) node)->fields), String))
2196 : {
2197 32599 : char *name = strVal(linitial(((ColumnRef *) node)->fields));
2198 32599 : int location = ((ColumnRef *) node)->location;
2199 :
2200 32599 : if (exprKind == EXPR_KIND_GROUP_BY)
2201 : {
2202 : /*
2203 : * In GROUP BY, we must prefer a match against a FROM-clause
2204 : * column to one against the targetlist. Look to see if there is
2205 : * a matching column. If so, fall through to use SQL99 rules.
2206 : * NOTE: if name could refer ambiguously to more than one column
2207 : * name exposed by FROM, colNameToVar will ereport(ERROR). That's
2208 : * just what we want here.
2209 : *
2210 : * Small tweak for 7.4.3: ignore matches in upper query levels.
2211 : * This effectively changes the search order for bare names to (1)
2212 : * local FROM variables, (2) local targetlist aliases, (3) outer
2213 : * FROM variables, whereas before it was (1) (3) (2). SQL92 and
2214 : * SQL99 do not allow GROUPing BY an outer reference, so this
2215 : * breaks no cases that are legal per spec, and it seems a more
2216 : * self-consistent behavior.
2217 : */
2218 3582 : if (colNameToVar(pstate, name, true, location) != NULL)
2219 3490 : name = NULL;
2220 : }
2221 :
2222 32599 : if (name != NULL)
2223 : {
2224 29109 : TargetEntry *target_result = NULL;
2225 :
2226 159963 : foreach(tl, *tlist)
2227 : {
2228 130854 : TargetEntry *tle = (TargetEntry *) lfirst(tl);
2229 :
2230 130854 : if (!tle->resjunk &&
2231 130193 : strcmp(tle->resname, name) == 0)
2232 : {
2233 25281 : if (target_result != NULL)
2234 : {
2235 6 : if (!equal(target_result->expr, tle->expr))
2236 0 : ereport(ERROR,
2237 : (errcode(ERRCODE_AMBIGUOUS_COLUMN),
2238 :
2239 : /*------
2240 : translator: first %s is name of a SQL construct, eg ORDER BY */
2241 : errmsg("%s \"%s\" is ambiguous",
2242 : ParseExprKindName(exprKind),
2243 : name),
2244 : parser_errposition(pstate, location)));
2245 : }
2246 : else
2247 25275 : target_result = tle;
2248 : /* Stay in loop to check for ambiguity */
2249 : }
2250 : }
2251 29109 : if (target_result != NULL)
2252 : {
2253 : /* return the first match, after suitable validation */
2254 25275 : checkTargetlistEntrySQL92(pstate, target_result, exprKind);
2255 25275 : return target_result;
2256 : }
2257 : }
2258 : }
2259 53504 : if (IsA(node, A_Const))
2260 : {
2261 26540 : A_Const *aconst = castNode(A_Const, node);
2262 26540 : int targetlist_pos = 0;
2263 : int target_pos;
2264 :
2265 26540 : if (!IsA(&aconst->val, Integer))
2266 0 : ereport(ERROR,
2267 : (errcode(ERRCODE_SYNTAX_ERROR),
2268 : /* translator: %s is name of a SQL construct, eg ORDER BY */
2269 : errmsg("non-integer constant in %s",
2270 : ParseExprKindName(exprKind)),
2271 : parser_errposition(pstate, aconst->location)));
2272 :
2273 26540 : target_pos = intVal(&aconst->val);
2274 45518 : foreach(tl, *tlist)
2275 : {
2276 45514 : TargetEntry *tle = (TargetEntry *) lfirst(tl);
2277 :
2278 45514 : if (!tle->resjunk)
2279 : {
2280 45514 : if (++targetlist_pos == target_pos)
2281 : {
2282 : /* return the unique match, after suitable validation */
2283 26536 : checkTargetlistEntrySQL92(pstate, tle, exprKind);
2284 26532 : return tle;
2285 : }
2286 : }
2287 : }
2288 4 : ereport(ERROR,
2289 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
2290 : /* translator: %s is name of a SQL construct, eg ORDER BY */
2291 : errmsg("%s position %d is not in select list",
2292 : ParseExprKindName(exprKind), target_pos),
2293 : parser_errposition(pstate, aconst->location)));
2294 : }
2295 :
2296 : /*
2297 : * Otherwise, we have an expression, so process it per SQL99 rules.
2298 : */
2299 26964 : return findTargetlistEntrySQL99(pstate, node, tlist, exprKind);
2300 : }
2301 :
2302 : /*
2303 : * findTargetlistEntrySQL99 -
2304 : * Returns the targetlist entry matching the given (untransformed) node.
2305 : * If no matching entry exists, one is created and appended to the target
2306 : * list as a "resjunk" node.
2307 : *
2308 : * This function supports the SQL99 interpretation, wherein the expression
2309 : * is just an ordinary expression referencing input column names.
2310 : *
2311 : * node the ORDER BY, GROUP BY, etc expression to be matched
2312 : * tlist the target list (passed by reference so we can append to it)
2313 : * exprKind identifies clause type being processed
2314 : */
2315 : static TargetEntry *
2316 31054 : findTargetlistEntrySQL99(ParseState *pstate, Node *node, List **tlist,
2317 : ParseExprKind exprKind)
2318 : {
2319 : TargetEntry *target_result;
2320 : ListCell *tl;
2321 : Node *expr;
2322 :
2323 : /*
2324 : * Convert the untransformed node to a transformed expression, and search
2325 : * for a match in the tlist. NOTE: it doesn't really matter whether there
2326 : * is more than one match. Also, we are willing to match an existing
2327 : * resjunk target here, though the SQL92 cases above must ignore resjunk
2328 : * targets.
2329 : */
2330 31054 : expr = transformExpr(pstate, node, exprKind);
2331 :
2332 115497 : foreach(tl, *tlist)
2333 : {
2334 97036 : TargetEntry *tle = (TargetEntry *) lfirst(tl);
2335 : Node *texpr;
2336 :
2337 : /*
2338 : * Ignore any implicit cast on the existing tlist expression.
2339 : *
2340 : * This essentially allows the ORDER/GROUP/etc item to adopt the same
2341 : * datatype previously selected for a textually-equivalent tlist item.
2342 : * There can't be any implicit cast at top level in an ordinary SELECT
2343 : * tlist at this stage, but the case does arise with ORDER BY in an
2344 : * aggregate function.
2345 : */
2346 97036 : texpr = strip_implicit_coercions((Node *) tle->expr);
2347 :
2348 97036 : if (equal(expr, texpr))
2349 12557 : return tle;
2350 : }
2351 :
2352 : /*
2353 : * If no matches, construct a new target entry which is appended to the
2354 : * end of the target list. This target is given resjunk = true so that it
2355 : * will not be projected into the final tuple.
2356 : */
2357 18461 : target_result = transformTargetEntry(pstate, node, expr, exprKind,
2358 : NULL, true);
2359 :
2360 18461 : *tlist = lappend(*tlist, target_result);
2361 :
2362 18461 : return target_result;
2363 : }
2364 :
2365 : /*-------------------------------------------------------------------------
2366 : * Flatten out parenthesized sublists in grouping lists, and some cases
2367 : * of nested grouping sets.
2368 : *
2369 : * Inside a grouping set (ROLLUP, CUBE, or GROUPING SETS), we expect the
2370 : * content to be nested no more than 2 deep: i.e. ROLLUP((a,b),(c,d)) is
2371 : * ok, but ROLLUP((a,(b,c)),d) is flattened to ((a,b,c),d), which we then
2372 : * (later) normalize to ((a,b,c),(d)).
2373 : *
2374 : * CUBE or ROLLUP can be nested inside GROUPING SETS (but not the reverse),
2375 : * and we leave that alone if we find it. But if we see GROUPING SETS inside
2376 : * GROUPING SETS, we can flatten and normalize as follows:
2377 : * GROUPING SETS (a, (b,c), GROUPING SETS ((c,d),(e)), (f,g))
2378 : * becomes
2379 : * GROUPING SETS ((a), (b,c), (c,d), (e), (f,g))
2380 : *
2381 : * This is per the spec's syntax transformations, but these are the only such
2382 : * transformations we do in parse analysis, so that queries retain the
2383 : * originally specified grouping set syntax for CUBE and ROLLUP as much as
2384 : * possible when deparsed. (Full expansion of the result into a list of
2385 : * grouping sets is left to the planner.)
2386 : *
2387 : * When we're done, the resulting list should contain only these possible
2388 : * elements:
2389 : * - an expression
2390 : * - a CUBE or ROLLUP with a list of expressions nested 2 deep
2391 : * - a GROUPING SET containing any of:
2392 : * - expression lists
2393 : * - empty grouping sets
2394 : * - CUBE or ROLLUP nodes with lists nested 2 deep
2395 : * The return is a new list, but doesn't deep-copy the old nodes except for
2396 : * GroupingSet nodes.
2397 : *
2398 : * As a side effect, flag whether the list has any GroupingSet nodes.
2399 : *-------------------------------------------------------------------------
2400 : */
2401 : static Node *
2402 313891 : flatten_grouping_sets(Node *expr, bool toplevel, bool *hasGroupingSets)
2403 : {
2404 : /* just in case of pathological input */
2405 313891 : check_stack_depth();
2406 :
2407 313891 : if (expr == (Node *) NIL)
2408 301603 : return (Node *) NIL;
2409 :
2410 12288 : switch (expr->type)
2411 : {
2412 242 : case T_RowExpr:
2413 : {
2414 242 : RowExpr *r = (RowExpr *) expr;
2415 :
2416 242 : if (r->row_format == COERCE_IMPLICIT_CAST)
2417 242 : return flatten_grouping_sets((Node *) r->args,
2418 : false, NULL);
2419 : }
2420 0 : break;
2421 1076 : case T_GroupingSet:
2422 : {
2423 1076 : GroupingSet *gset = (GroupingSet *) expr;
2424 : ListCell *l2;
2425 1076 : List *result_set = NIL;
2426 :
2427 1076 : if (hasGroupingSets)
2428 794 : *hasGroupingSets = true;
2429 :
2430 : /*
2431 : * at the top level, we skip over all empty grouping sets; the
2432 : * caller can supply the canonical GROUP BY () if nothing is
2433 : * left.
2434 : */
2435 :
2436 1076 : if (toplevel && gset->kind == GROUPING_SET_EMPTY)
2437 24 : return (Node *) NIL;
2438 :
2439 2756 : foreach(l2, gset->content)
2440 : {
2441 1704 : Node *n1 = lfirst(l2);
2442 1704 : Node *n2 = flatten_grouping_sets(n1, false, NULL);
2443 :
2444 1704 : if (IsA(n1, GroupingSet) &&
2445 282 : ((GroupingSet *) n1)->kind == GROUPING_SET_SETS)
2446 68 : result_set = list_concat(result_set, (List *) n2);
2447 : else
2448 1636 : result_set = lappend(result_set, n2);
2449 : }
2450 :
2451 : /*
2452 : * At top level, keep the grouping set node; but if we're in a
2453 : * nested grouping set, then we need to concat the flattened
2454 : * result into the outer list if it's simply nested.
2455 : */
2456 :
2457 1052 : if (toplevel || (gset->kind != GROUPING_SET_SETS))
2458 : {
2459 984 : return (Node *) makeGroupingSet(gset->kind, result_set, gset->location);
2460 : }
2461 : else
2462 68 : return (Node *) result_set;
2463 : }
2464 4288 : case T_List:
2465 : {
2466 4288 : List *result = NIL;
2467 : ListCell *l;
2468 :
2469 10584 : foreach(l, (List *) expr)
2470 : {
2471 6296 : Node *n = flatten_grouping_sets(lfirst(l), toplevel, hasGroupingSets);
2472 :
2473 6296 : if (n != (Node *) NIL)
2474 : {
2475 6272 : if (IsA(n, List))
2476 30 : result = list_concat(result, (List *) n);
2477 : else
2478 6242 : result = lappend(result, n);
2479 : }
2480 : }
2481 :
2482 4288 : return (Node *) result;
2483 : }
2484 6682 : default:
2485 6682 : break;
2486 : }
2487 :
2488 6682 : return expr;
2489 : }
2490 :
2491 : /*
2492 : * Transform a single expression within a GROUP BY clause or grouping set.
2493 : *
2494 : * The expression is added to the targetlist if not already present, and to the
2495 : * flatresult list (which will become the groupClause) if not already present
2496 : * there. The sortClause is consulted for operator and sort order hints.
2497 : *
2498 : * Returns the ressortgroupref of the expression.
2499 : *
2500 : * flatresult reference to flat list of SortGroupClause nodes
2501 : * seen_local bitmapset of sortgrouprefs already seen at the local level
2502 : * pstate ParseState
2503 : * gexpr node to transform
2504 : * targetlist reference to TargetEntry list
2505 : * sortClause ORDER BY clause (SortGroupClause nodes)
2506 : * exprKind expression kind
2507 : * useSQL99 SQL99 rather than SQL92 syntax
2508 : * toplevel false if within any grouping set
2509 : */
2510 : static Index
2511 6682 : transformGroupClauseExpr(List **flatresult, Bitmapset *seen_local,
2512 : ParseState *pstate, Node *gexpr,
2513 : List **targetlist, List *sortClause,
2514 : ParseExprKind exprKind, bool useSQL99, bool toplevel)
2515 : {
2516 : TargetEntry *tle;
2517 6682 : bool found = false;
2518 :
2519 6682 : if (useSQL99)
2520 786 : tle = findTargetlistEntrySQL99(pstate, gexpr,
2521 : targetlist, exprKind);
2522 : else
2523 5896 : tle = findTargetlistEntrySQL92(pstate, gexpr,
2524 : targetlist, exprKind);
2525 :
2526 6666 : if (tle->ressortgroupref > 0)
2527 : {
2528 : ListCell *sl;
2529 :
2530 : /*
2531 : * Eliminate duplicates (GROUP BY x, x) but only at local level.
2532 : * (Duplicates in grouping sets can affect the number of returned
2533 : * rows, so can't be dropped indiscriminately.)
2534 : *
2535 : * Since we don't care about anything except the sortgroupref, we can
2536 : * use a bitmapset rather than scanning lists.
2537 : */
2538 2037 : if (bms_is_member(tle->ressortgroupref, seen_local))
2539 16 : return 0;
2540 :
2541 : /*
2542 : * If we're already in the flat clause list, we don't need to consider
2543 : * adding ourselves again.
2544 : */
2545 2021 : found = targetIsInSortList(tle, InvalidOid, *flatresult);
2546 2021 : if (found)
2547 174 : return tle->ressortgroupref;
2548 :
2549 : /*
2550 : * If the GROUP BY tlist entry also appears in ORDER BY, copy operator
2551 : * info from the (first) matching ORDER BY item. This means that if
2552 : * you write something like "GROUP BY foo ORDER BY foo USING <<<", the
2553 : * GROUP BY operation silently takes on the equality semantics implied
2554 : * by the ORDER BY. There are two reasons to do this: it improves the
2555 : * odds that we can implement both GROUP BY and ORDER BY with a single
2556 : * sort step, and it allows the user to choose the equality semantics
2557 : * used by GROUP BY, should she be working with a datatype that has
2558 : * more than one equality operator.
2559 : *
2560 : * If we're in a grouping set, though, we force our requested ordering
2561 : * to be NULLS LAST, because if we have any hope of using a sorted agg
2562 : * for the job, we're going to be tacking on generated NULL values
2563 : * after the corresponding groups. If the user demands nulls first,
2564 : * another sort step is going to be inevitable, but that's the
2565 : * planner's problem.
2566 : */
2567 :
2568 2510 : foreach(sl, sortClause)
2569 : {
2570 2381 : SortGroupClause *sc = (SortGroupClause *) lfirst(sl);
2571 :
2572 2381 : if (sc->tleSortGroupRef == tle->ressortgroupref)
2573 : {
2574 1718 : SortGroupClause *grpc = copyObject(sc);
2575 :
2576 1718 : if (!toplevel)
2577 466 : grpc->nulls_first = false;
2578 1718 : *flatresult = lappend(*flatresult, grpc);
2579 1718 : found = true;
2580 1718 : break;
2581 : }
2582 : }
2583 : }
2584 :
2585 : /*
2586 : * If no match in ORDER BY, just add it to the result using default
2587 : * sort/group semantics.
2588 : */
2589 6476 : if (!found)
2590 4758 : *flatresult = addTargetToGroupList(pstate, tle,
2591 : *flatresult, *targetlist,
2592 : exprLocation(gexpr));
2593 :
2594 : /*
2595 : * _something_ must have assigned us a sortgroupref by now...
2596 : */
2597 :
2598 6476 : return tle->ressortgroupref;
2599 : }
2600 :
2601 : /*
2602 : * Transform a list of expressions within a GROUP BY clause or grouping set.
2603 : *
2604 : * The list of expressions belongs to a single clause within which duplicates
2605 : * can be safely eliminated.
2606 : *
2607 : * Returns an integer list of ressortgroupref values.
2608 : *
2609 : * flatresult reference to flat list of SortGroupClause nodes
2610 : * pstate ParseState
2611 : * list nodes to transform
2612 : * targetlist reference to TargetEntry list
2613 : * sortClause ORDER BY clause (SortGroupClause nodes)
2614 : * exprKind expression kind
2615 : * useSQL99 SQL99 rather than SQL92 syntax
2616 : * toplevel false if within any grouping set
2617 : */
2618 : static List *
2619 212 : transformGroupClauseList(List **flatresult,
2620 : ParseState *pstate, List *list,
2621 : List **targetlist, List *sortClause,
2622 : ParseExprKind exprKind, bool useSQL99, bool toplevel)
2623 : {
2624 212 : Bitmapset *seen_local = NULL;
2625 212 : List *result = NIL;
2626 : ListCell *gl;
2627 :
2628 652 : foreach(gl, list)
2629 : {
2630 440 : Node *gexpr = (Node *) lfirst(gl);
2631 :
2632 440 : Index ref = transformGroupClauseExpr(flatresult,
2633 : seen_local,
2634 : pstate,
2635 : gexpr,
2636 : targetlist,
2637 : sortClause,
2638 : exprKind,
2639 : useSQL99,
2640 : toplevel);
2641 :
2642 440 : if (ref > 0)
2643 : {
2644 432 : seen_local = bms_add_member(seen_local, ref);
2645 432 : result = lappend_int(result, ref);
2646 : }
2647 : }
2648 :
2649 212 : return result;
2650 : }
2651 :
2652 : /*
2653 : * Transform a grouping set and (recursively) its content.
2654 : *
2655 : * The grouping set might be a GROUPING SETS node with other grouping sets
2656 : * inside it, but SETS within SETS have already been flattened out before
2657 : * reaching here.
2658 : *
2659 : * Returns the transformed node, which now contains SIMPLE nodes with lists
2660 : * of ressortgrouprefs rather than expressions.
2661 : *
2662 : * flatresult reference to flat list of SortGroupClause nodes
2663 : * pstate ParseState
2664 : * gset grouping set to transform
2665 : * targetlist reference to TargetEntry list
2666 : * sortClause ORDER BY clause (SortGroupClause nodes)
2667 : * exprKind expression kind
2668 : * useSQL99 SQL99 rather than SQL92 syntax
2669 : * toplevel false if within any grouping set
2670 : */
2671 : static Node *
2672 984 : transformGroupingSet(List **flatresult,
2673 : ParseState *pstate, GroupingSet *gset,
2674 : List **targetlist, List *sortClause,
2675 : ParseExprKind exprKind, bool useSQL99, bool toplevel)
2676 : {
2677 : ListCell *gl;
2678 984 : List *content = NIL;
2679 :
2680 : Assert(toplevel || gset->kind != GROUPING_SET_SETS);
2681 :
2682 2620 : foreach(gl, gset->content)
2683 : {
2684 1636 : Node *n = lfirst(gl);
2685 :
2686 1636 : if (IsA(n, List))
2687 : {
2688 212 : List *l = transformGroupClauseList(flatresult,
2689 : pstate, (List *) n,
2690 : targetlist, sortClause,
2691 : exprKind, useSQL99, false);
2692 :
2693 212 : content = lappend(content, makeGroupingSet(GROUPING_SET_SIMPLE,
2694 : l,
2695 : exprLocation(n)));
2696 : }
2697 1424 : else if (IsA(n, GroupingSet))
2698 : {
2699 214 : GroupingSet *gset2 = (GroupingSet *) lfirst(gl);
2700 :
2701 214 : content = lappend(content, transformGroupingSet(flatresult,
2702 : pstate, gset2,
2703 : targetlist, sortClause,
2704 : exprKind, useSQL99, false));
2705 : }
2706 : else
2707 : {
2708 1210 : Index ref = transformGroupClauseExpr(flatresult,
2709 : NULL,
2710 : pstate,
2711 : n,
2712 : targetlist,
2713 : sortClause,
2714 : exprKind,
2715 : useSQL99,
2716 : false);
2717 :
2718 1210 : content = lappend(content, makeGroupingSet(GROUPING_SET_SIMPLE,
2719 : list_make1_int(ref),
2720 : exprLocation(n)));
2721 : }
2722 : }
2723 :
2724 : /* Arbitrarily cap the size of CUBE, which has exponential growth */
2725 984 : if (gset->kind == GROUPING_SET_CUBE)
2726 : {
2727 122 : if (list_length(content) > 12)
2728 0 : ereport(ERROR,
2729 : (errcode(ERRCODE_TOO_MANY_COLUMNS),
2730 : errmsg("CUBE is limited to 12 elements"),
2731 : parser_errposition(pstate, gset->location)));
2732 : }
2733 :
2734 984 : return (Node *) makeGroupingSet(gset->kind, content, gset->location);
2735 : }
2736 :
2737 :
2738 : /*
2739 : * transformGroupClause -
2740 : * transform a GROUP BY clause
2741 : *
2742 : * GROUP BY items will be added to the targetlist (as resjunk columns)
2743 : * if not already present, so the targetlist must be passed by reference.
2744 : *
2745 : * If GROUP BY ALL is specified, the groupClause will be inferred to be all
2746 : * non-aggregate, non-window expressions in the targetlist.
2747 : *
2748 : * This is also used for window PARTITION BY clauses (which act almost the
2749 : * same, but are always interpreted per SQL99 rules).
2750 : *
2751 : * Grouping sets make this a lot more complex than it was. Our goal here is
2752 : * twofold: we make a flat list of SortGroupClause nodes referencing each
2753 : * distinct expression used for grouping, with those expressions added to the
2754 : * targetlist if needed. At the same time, we build the groupingSets tree,
2755 : * which stores only ressortgrouprefs as integer lists inside GroupingSet nodes
2756 : * (possibly nested, but limited in depth: a GROUPING_SET_SETS node can contain
2757 : * nested SIMPLE, CUBE or ROLLUP nodes, but not more sets - we flatten that
2758 : * out; while CUBE and ROLLUP can contain only SIMPLE nodes).
2759 : *
2760 : * We skip much of the hard work if there are no grouping sets.
2761 : *
2762 : * One subtlety is that the groupClause list can end up empty while the
2763 : * groupingSets list is not; this happens if there are only empty grouping
2764 : * sets, or an explicit GROUP BY (). This has the same effect as specifying
2765 : * aggregates or a HAVING clause with no GROUP BY; the output is one row per
2766 : * grouping set even if the input is empty.
2767 : *
2768 : * Returns the transformed (flat) groupClause.
2769 : *
2770 : * pstate ParseState
2771 : * grouplist clause to transform
2772 : * groupByAll is this a GROUP BY ALL statement?
2773 : * groupingSets reference to list to contain the grouping set tree
2774 : * targetlist reference to TargetEntry list
2775 : * sortClause ORDER BY clause (SortGroupClause nodes)
2776 : * exprKind expression kind
2777 : * useSQL99 SQL99 rather than SQL92 syntax
2778 : */
2779 : List *
2780 305685 : transformGroupClause(ParseState *pstate, List *grouplist, bool groupByAll,
2781 : List **groupingSets,
2782 : List **targetlist, List *sortClause,
2783 : ParseExprKind exprKind, bool useSQL99)
2784 : {
2785 305685 : List *result = NIL;
2786 : List *flat_grouplist;
2787 305685 : List *gsets = NIL;
2788 : ListCell *gl;
2789 305685 : bool hasGroupingSets = false;
2790 305685 : Bitmapset *seen_local = NULL;
2791 :
2792 : /* Handle GROUP BY ALL */
2793 305685 : if (groupByAll)
2794 : {
2795 : /* There cannot have been any explicit grouplist items */
2796 : Assert(grouplist == NIL);
2797 :
2798 : /* Iterate over targets, adding acceptable ones to the result list */
2799 176 : foreach_ptr(TargetEntry, tle, *targetlist)
2800 : {
2801 : /* Ignore junk TLEs */
2802 88 : if (tle->resjunk)
2803 0 : continue;
2804 :
2805 : /*
2806 : * TLEs containing aggregates are not okay to add to GROUP BY
2807 : * (compare checkTargetlistEntrySQL92). But the SQL standard
2808 : * directs us to skip them, so it's fine.
2809 : */
2810 164 : if (pstate->p_hasAggs &&
2811 76 : contain_aggs_of_level((Node *) tle->expr, 0))
2812 36 : continue;
2813 :
2814 : /*
2815 : * Likewise, TLEs containing window functions are not okay to add
2816 : * to GROUP BY. At this writing, the SQL standard is silent on
2817 : * what to do with them, but by analogy to aggregates we'll just
2818 : * skip them.
2819 : */
2820 60 : if (pstate->p_hasWindowFuncs &&
2821 8 : contain_windowfuncs((Node *) tle->expr))
2822 4 : continue;
2823 :
2824 : /*
2825 : * Otherwise, add the TLE to the result using default sort/group
2826 : * semantics. We specify the parse location as the TLE's
2827 : * location, despite the comment for addTargetToGroupList
2828 : * discouraging that. The only other thing we could point to is
2829 : * the ALL keyword, which seems unhelpful when there are multiple
2830 : * TLEs.
2831 : */
2832 48 : result = addTargetToGroupList(pstate, tle,
2833 : result, *targetlist,
2834 48 : exprLocation((Node *) tle->expr));
2835 : }
2836 :
2837 : /* If we found any acceptable targets, we're done */
2838 44 : if (result != NIL)
2839 36 : return result;
2840 :
2841 : /*
2842 : * Otherwise, the SQL standard says to treat it like "GROUP BY ()".
2843 : * Build a representation of that, and let the rest of this function
2844 : * handle it.
2845 : */
2846 8 : grouplist = list_make1(makeGroupingSet(GROUPING_SET_EMPTY, NIL, -1));
2847 : }
2848 :
2849 : /*
2850 : * Recursively flatten implicit RowExprs. (Technically this is only needed
2851 : * for GROUP BY, per the syntax rules for grouping sets, but we do it
2852 : * anyway.)
2853 : */
2854 305649 : flat_grouplist = (List *) flatten_grouping_sets((Node *) grouplist,
2855 : true,
2856 : &hasGroupingSets);
2857 :
2858 : /*
2859 : * If the list is now empty, but hasGroupingSets is true, it's because we
2860 : * elided redundant empty grouping sets. Restore a single empty grouping
2861 : * set to leave a canonical form: GROUP BY ()
2862 : */
2863 :
2864 305649 : if (flat_grouplist == NIL && hasGroupingSets)
2865 : {
2866 24 : flat_grouplist = list_make1(makeGroupingSet(GROUPING_SET_EMPTY,
2867 : NIL,
2868 : exprLocation((Node *) grouplist)));
2869 : }
2870 :
2871 311459 : foreach(gl, flat_grouplist)
2872 : {
2873 5826 : Node *gexpr = (Node *) lfirst(gl);
2874 :
2875 5826 : if (IsA(gexpr, GroupingSet))
2876 : {
2877 794 : GroupingSet *gset = (GroupingSet *) gexpr;
2878 :
2879 794 : switch (gset->kind)
2880 : {
2881 24 : case GROUPING_SET_EMPTY:
2882 24 : gsets = lappend(gsets, gset);
2883 24 : break;
2884 0 : case GROUPING_SET_SIMPLE:
2885 : /* can't happen */
2886 : Assert(false);
2887 0 : break;
2888 770 : case GROUPING_SET_SETS:
2889 : case GROUPING_SET_CUBE:
2890 : case GROUPING_SET_ROLLUP:
2891 770 : gsets = lappend(gsets,
2892 770 : transformGroupingSet(&result,
2893 : pstate, gset,
2894 : targetlist, sortClause,
2895 : exprKind, useSQL99, true));
2896 770 : break;
2897 : }
2898 : }
2899 : else
2900 : {
2901 5032 : Index ref = transformGroupClauseExpr(&result, seen_local,
2902 : pstate, gexpr,
2903 : targetlist, sortClause,
2904 : exprKind, useSQL99, true);
2905 :
2906 5016 : if (ref > 0)
2907 : {
2908 5008 : seen_local = bms_add_member(seen_local, ref);
2909 5008 : if (hasGroupingSets)
2910 32 : gsets = lappend(gsets,
2911 32 : makeGroupingSet(GROUPING_SET_SIMPLE,
2912 : list_make1_int(ref),
2913 : exprLocation(gexpr)));
2914 : }
2915 : }
2916 : }
2917 :
2918 : /* parser should prevent this */
2919 : Assert(gsets == NIL || groupingSets != NULL);
2920 :
2921 305633 : if (groupingSets)
2922 303563 : *groupingSets = gsets;
2923 :
2924 305633 : return result;
2925 : }
2926 :
2927 : /*
2928 : * transformSortClause -
2929 : * transform an ORDER BY clause
2930 : *
2931 : * ORDER BY items will be added to the targetlist (as resjunk columns)
2932 : * if not already present, so the targetlist must be passed by reference.
2933 : *
2934 : * This is also used for window and aggregate ORDER BY clauses (which act
2935 : * almost the same, but are always interpreted per SQL99 rules).
2936 : */
2937 : List *
2938 350622 : transformSortClause(ParseState *pstate,
2939 : List *orderlist,
2940 : List **targetlist,
2941 : ParseExprKind exprKind,
2942 : bool useSQL99)
2943 : {
2944 350622 : List *sortlist = NIL;
2945 : ListCell *olitem;
2946 :
2947 426528 : foreach(olitem, orderlist)
2948 : {
2949 75934 : SortBy *sortby = (SortBy *) lfirst(olitem);
2950 : TargetEntry *tle;
2951 :
2952 75934 : if (useSQL99)
2953 3304 : tle = findTargetlistEntrySQL99(pstate, sortby->node,
2954 : targetlist, exprKind);
2955 : else
2956 72630 : tle = findTargetlistEntrySQL92(pstate, sortby->node,
2957 : targetlist, exprKind);
2958 :
2959 75910 : sortlist = addTargetToSortList(pstate, tle,
2960 : sortlist, *targetlist, sortby);
2961 : }
2962 :
2963 350594 : return sortlist;
2964 : }
2965 :
2966 : /*
2967 : * transformWindowDefinitions -
2968 : * transform window definitions (WindowDef to WindowClause)
2969 : */
2970 : List *
2971 303583 : transformWindowDefinitions(ParseState *pstate,
2972 : List *windowdefs,
2973 : List **targetlist)
2974 : {
2975 303583 : List *result = NIL;
2976 303583 : Index winref = 0;
2977 : ListCell *lc;
2978 :
2979 305617 : foreach(lc, windowdefs)
2980 : {
2981 2078 : WindowDef *windef = (WindowDef *) lfirst(lc);
2982 2078 : WindowClause *refwc = NULL;
2983 : List *partitionClause;
2984 : List *orderClause;
2985 2078 : Oid rangeopfamily = InvalidOid;
2986 2078 : Oid rangeopcintype = InvalidOid;
2987 : WindowClause *wc;
2988 :
2989 2078 : winref++;
2990 :
2991 : /*
2992 : * Check for duplicate window names.
2993 : */
2994 2504 : if (windef->name &&
2995 426 : findWindowClause(result, windef->name) != NULL)
2996 4 : ereport(ERROR,
2997 : (errcode(ERRCODE_WINDOWING_ERROR),
2998 : errmsg("window \"%s\" is already defined", windef->name),
2999 : parser_errposition(pstate, windef->location)));
3000 :
3001 : /*
3002 : * If it references a previous window, look that up.
3003 : */
3004 2074 : if (windef->refname)
3005 : {
3006 28 : refwc = findWindowClause(result, windef->refname);
3007 28 : if (refwc == NULL)
3008 0 : ereport(ERROR,
3009 : (errcode(ERRCODE_UNDEFINED_OBJECT),
3010 : errmsg("window \"%s\" does not exist",
3011 : windef->refname),
3012 : parser_errposition(pstate, windef->location)));
3013 : }
3014 :
3015 : /*
3016 : * Transform PARTITION and ORDER specs, if any. These are treated
3017 : * almost exactly like top-level GROUP BY and ORDER BY clauses,
3018 : * including the special handling of nondefault operator semantics.
3019 : */
3020 2074 : orderClause = transformSortClause(pstate,
3021 : windef->orderClause,
3022 : targetlist,
3023 : EXPR_KIND_WINDOW_ORDER,
3024 : true /* force SQL99 rules */ );
3025 2070 : partitionClause = transformGroupClause(pstate,
3026 : windef->partitionClause,
3027 : false /* not GROUP BY ALL */ ,
3028 : NULL,
3029 : targetlist,
3030 : orderClause,
3031 : EXPR_KIND_WINDOW_PARTITION,
3032 : true /* force SQL99 rules */ );
3033 :
3034 : /*
3035 : * And prepare the new WindowClause.
3036 : */
3037 2070 : wc = makeNode(WindowClause);
3038 2070 : wc->name = windef->name;
3039 2070 : wc->refname = windef->refname;
3040 :
3041 : /*
3042 : * Per spec, a windowdef that references a previous one copies the
3043 : * previous partition clause (and mustn't specify its own). It can
3044 : * specify its own ordering clause, but only if the previous one had
3045 : * none. It always specifies its own frame clause, and the previous
3046 : * one must not have a frame clause. Yeah, it's bizarre that each of
3047 : * these cases works differently, but SQL:2008 says so; see 7.11
3048 : * <window clause> syntax rule 10 and general rule 1. The frame
3049 : * clause rule is especially bizarre because it makes "OVER foo"
3050 : * different from "OVER (foo)", and requires the latter to throw an
3051 : * error if foo has a nondefault frame clause. Well, ours not to
3052 : * reason why, but we do go out of our way to throw a useful error
3053 : * message for such cases.
3054 : */
3055 2070 : if (refwc)
3056 : {
3057 28 : if (partitionClause)
3058 0 : ereport(ERROR,
3059 : (errcode(ERRCODE_WINDOWING_ERROR),
3060 : errmsg("cannot override PARTITION BY clause of window \"%s\"",
3061 : windef->refname),
3062 : parser_errposition(pstate, windef->location)));
3063 28 : wc->partitionClause = copyObject(refwc->partitionClause);
3064 : }
3065 : else
3066 2042 : wc->partitionClause = partitionClause;
3067 2070 : if (refwc)
3068 : {
3069 28 : if (orderClause && refwc->orderClause)
3070 0 : ereport(ERROR,
3071 : (errcode(ERRCODE_WINDOWING_ERROR),
3072 : errmsg("cannot override ORDER BY clause of window \"%s\"",
3073 : windef->refname),
3074 : parser_errposition(pstate, windef->location)));
3075 28 : if (orderClause)
3076 : {
3077 12 : wc->orderClause = orderClause;
3078 12 : wc->copiedOrder = false;
3079 : }
3080 : else
3081 : {
3082 16 : wc->orderClause = copyObject(refwc->orderClause);
3083 16 : wc->copiedOrder = true;
3084 : }
3085 : }
3086 : else
3087 : {
3088 2042 : wc->orderClause = orderClause;
3089 2042 : wc->copiedOrder = false;
3090 : }
3091 2070 : if (refwc && refwc->frameOptions != FRAMEOPTION_DEFAULTS)
3092 : {
3093 : /*
3094 : * Use this message if this is a WINDOW clause, or if it's an OVER
3095 : * clause that includes ORDER BY or framing clauses. (We already
3096 : * rejected PARTITION BY above, so no need to check that.)
3097 : */
3098 0 : if (windef->name ||
3099 0 : orderClause || windef->frameOptions != FRAMEOPTION_DEFAULTS)
3100 0 : ereport(ERROR,
3101 : (errcode(ERRCODE_WINDOWING_ERROR),
3102 : errmsg("cannot copy window \"%s\" because it has a frame clause",
3103 : windef->refname),
3104 : parser_errposition(pstate, windef->location)));
3105 : /* Else this clause is just OVER (foo), so say this: */
3106 0 : ereport(ERROR,
3107 : (errcode(ERRCODE_WINDOWING_ERROR),
3108 : errmsg("cannot copy window \"%s\" because it has a frame clause",
3109 : windef->refname),
3110 : errhint("Omit the parentheses in this OVER clause."),
3111 : parser_errposition(pstate, windef->location)));
3112 : }
3113 2070 : wc->frameOptions = windef->frameOptions;
3114 :
3115 : /*
3116 : * RANGE offset PRECEDING/FOLLOWING requires exactly one ORDER BY
3117 : * column; check that and get its sort opfamily info.
3118 : */
3119 2070 : if ((wc->frameOptions & FRAMEOPTION_RANGE) &&
3120 1484 : (wc->frameOptions & (FRAMEOPTION_START_OFFSET |
3121 : FRAMEOPTION_END_OFFSET)))
3122 : {
3123 : SortGroupClause *sortcl;
3124 : Node *sortkey;
3125 : CompareType rangecmptype;
3126 :
3127 424 : if (list_length(wc->orderClause) != 1)
3128 12 : ereport(ERROR,
3129 : (errcode(ERRCODE_WINDOWING_ERROR),
3130 : errmsg("RANGE with offset PRECEDING/FOLLOWING requires exactly one ORDER BY column"),
3131 : parser_errposition(pstate, windef->location)));
3132 412 : sortcl = linitial_node(SortGroupClause, wc->orderClause);
3133 412 : sortkey = get_sortgroupclause_expr(sortcl, *targetlist);
3134 : /* Find the sort operator in pg_amop */
3135 412 : if (!get_ordering_op_properties(sortcl->sortop,
3136 : &rangeopfamily,
3137 : &rangeopcintype,
3138 : &rangecmptype))
3139 0 : elog(ERROR, "operator %u is not a valid ordering operator",
3140 : sortcl->sortop);
3141 : /* Record properties of sort ordering */
3142 412 : wc->inRangeColl = exprCollation(sortkey);
3143 412 : wc->inRangeAsc = !sortcl->reverse_sort;
3144 412 : wc->inRangeNullsFirst = sortcl->nulls_first;
3145 : }
3146 :
3147 : /* Per spec, GROUPS mode requires an ORDER BY clause */
3148 2058 : if (wc->frameOptions & FRAMEOPTION_GROUPS)
3149 : {
3150 124 : if (wc->orderClause == NIL)
3151 4 : ereport(ERROR,
3152 : (errcode(ERRCODE_WINDOWING_ERROR),
3153 : errmsg("GROUPS mode requires an ORDER BY clause"),
3154 : parser_errposition(pstate, windef->location)));
3155 : }
3156 :
3157 : /* Process frame offset expressions */
3158 2054 : wc->startOffset = transformFrameOffset(pstate, wc->frameOptions,
3159 : rangeopfamily, rangeopcintype,
3160 : &wc->startInRangeFunc,
3161 : windef->startOffset);
3162 2038 : wc->endOffset = transformFrameOffset(pstate, wc->frameOptions,
3163 : rangeopfamily, rangeopcintype,
3164 : &wc->endInRangeFunc,
3165 : windef->endOffset);
3166 2034 : wc->winref = winref;
3167 :
3168 2034 : result = lappend(result, wc);
3169 : }
3170 :
3171 303539 : return result;
3172 : }
3173 :
3174 : /*
3175 : * transformDistinctClause -
3176 : * transform a DISTINCT clause
3177 : *
3178 : * Since we may need to add items to the query's targetlist, that list
3179 : * is passed by reference.
3180 : *
3181 : * As with GROUP BY, we absorb the sorting semantics of ORDER BY as much as
3182 : * possible into the distinctClause. This avoids a possible need to re-sort,
3183 : * and allows the user to choose the equality semantics used by DISTINCT,
3184 : * should she be working with a datatype that has more than one equality
3185 : * operator.
3186 : *
3187 : * is_agg is true if we are transforming an aggregate(DISTINCT ...)
3188 : * function call. This does not affect any behavior, only the phrasing
3189 : * of error messages.
3190 : */
3191 : List *
3192 2711 : transformDistinctClause(ParseState *pstate,
3193 : List **targetlist, List *sortClause, bool is_agg)
3194 : {
3195 2711 : List *result = NIL;
3196 : ListCell *slitem;
3197 : ListCell *tlitem;
3198 :
3199 : /*
3200 : * The distinctClause should consist of all ORDER BY items followed by all
3201 : * other non-resjunk targetlist items. There must not be any resjunk
3202 : * ORDER BY items --- that would imply that we are sorting by a value that
3203 : * isn't necessarily unique within a DISTINCT group, so the results
3204 : * wouldn't be well-defined. This construction ensures we follow the rule
3205 : * that sortClause and distinctClause match; in fact the sortClause will
3206 : * always be a prefix of distinctClause.
3207 : *
3208 : * Note a corner case: the same TLE could be in the ORDER BY list multiple
3209 : * times with different sortops. We have to include it in the
3210 : * distinctClause the same way to preserve the prefix property. The net
3211 : * effect will be that the TLE value will be made unique according to both
3212 : * sortops.
3213 : */
3214 3126 : foreach(slitem, sortClause)
3215 : {
3216 439 : SortGroupClause *scl = (SortGroupClause *) lfirst(slitem);
3217 439 : TargetEntry *tle = get_sortgroupclause_tle(scl, *targetlist);
3218 :
3219 439 : if (tle->resjunk)
3220 24 : ereport(ERROR,
3221 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3222 : is_agg ?
3223 : errmsg("in an aggregate with DISTINCT, ORDER BY expressions must appear in argument list") :
3224 : errmsg("for SELECT DISTINCT, ORDER BY expressions must appear in select list"),
3225 : parser_errposition(pstate,
3226 : exprLocation((Node *) tle->expr))));
3227 415 : result = lappend(result, copyObject(scl));
3228 : }
3229 :
3230 : /*
3231 : * Now add any remaining non-resjunk tlist items, using default sort/group
3232 : * semantics for their data types.
3233 : */
3234 10916 : foreach(tlitem, *targetlist)
3235 : {
3236 8229 : TargetEntry *tle = (TargetEntry *) lfirst(tlitem);
3237 :
3238 8229 : if (tle->resjunk)
3239 2 : continue; /* ignore junk */
3240 8227 : result = addTargetToGroupList(pstate, tle,
3241 : result, *targetlist,
3242 8227 : exprLocation((Node *) tle->expr));
3243 : }
3244 :
3245 : /*
3246 : * Complain if we found nothing to make DISTINCT. Returning an empty list
3247 : * would cause the parsed Query to look like it didn't have DISTINCT, with
3248 : * results that would probably surprise the user. Note: this case is
3249 : * presently impossible for aggregates because of grammar restrictions,
3250 : * but we check anyway.
3251 : */
3252 2687 : if (result == NIL)
3253 0 : ereport(ERROR,
3254 : (errcode(ERRCODE_SYNTAX_ERROR),
3255 : is_agg ?
3256 : errmsg("an aggregate with DISTINCT must have at least one argument") :
3257 : errmsg("SELECT DISTINCT must have at least one column")));
3258 :
3259 2687 : return result;
3260 : }
3261 :
3262 : /*
3263 : * transformDistinctOnClause -
3264 : * transform a DISTINCT ON clause
3265 : *
3266 : * Since we may need to add items to the query's targetlist, that list
3267 : * is passed by reference.
3268 : *
3269 : * As with GROUP BY, we absorb the sorting semantics of ORDER BY as much as
3270 : * possible into the distinctClause. This avoids a possible need to re-sort,
3271 : * and allows the user to choose the equality semantics used by DISTINCT,
3272 : * should she be working with a datatype that has more than one equality
3273 : * operator.
3274 : */
3275 : List *
3276 168 : transformDistinctOnClause(ParseState *pstate, List *distinctlist,
3277 : List **targetlist, List *sortClause)
3278 : {
3279 168 : List *result = NIL;
3280 168 : List *sortgrouprefs = NIL;
3281 : bool skipped_sortitem;
3282 : ListCell *lc;
3283 : ListCell *lc2;
3284 :
3285 : /*
3286 : * Add all the DISTINCT ON expressions to the tlist (if not already
3287 : * present, they are added as resjunk items). Assign sortgroupref numbers
3288 : * to them, and make a list of these numbers. (NB: we rely below on the
3289 : * sortgrouprefs list being one-for-one with the original distinctlist.
3290 : * Also notice that we could have duplicate DISTINCT ON expressions and
3291 : * hence duplicate entries in sortgrouprefs.)
3292 : */
3293 417 : foreach(lc, distinctlist)
3294 : {
3295 253 : Node *dexpr = (Node *) lfirst(lc);
3296 : int sortgroupref;
3297 : TargetEntry *tle;
3298 :
3299 253 : tle = findTargetlistEntrySQL92(pstate, dexpr, targetlist,
3300 : EXPR_KIND_DISTINCT_ON);
3301 249 : sortgroupref = assignSortGroupRef(tle, *targetlist);
3302 249 : sortgrouprefs = lappend_int(sortgrouprefs, sortgroupref);
3303 : }
3304 :
3305 : /*
3306 : * If the user writes both DISTINCT ON and ORDER BY, adopt the sorting
3307 : * semantics from ORDER BY items that match DISTINCT ON items, and also
3308 : * adopt their column sort order. We insist that the distinctClause and
3309 : * sortClause match, so throw error if we find the need to add any more
3310 : * distinctClause items after we've skipped an ORDER BY item that wasn't
3311 : * in DISTINCT ON.
3312 : */
3313 164 : skipped_sortitem = false;
3314 384 : foreach(lc, sortClause)
3315 : {
3316 224 : SortGroupClause *scl = (SortGroupClause *) lfirst(lc);
3317 :
3318 224 : if (list_member_int(sortgrouprefs, scl->tleSortGroupRef))
3319 : {
3320 168 : if (skipped_sortitem)
3321 4 : ereport(ERROR,
3322 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3323 : errmsg("SELECT DISTINCT ON expressions must match initial ORDER BY expressions"),
3324 : parser_errposition(pstate,
3325 : get_matching_location(scl->tleSortGroupRef,
3326 : sortgrouprefs,
3327 : distinctlist))));
3328 : else
3329 164 : result = lappend(result, copyObject(scl));
3330 : }
3331 : else
3332 56 : skipped_sortitem = true;
3333 : }
3334 :
3335 : /*
3336 : * Now add any remaining DISTINCT ON items, using default sort/group
3337 : * semantics for their data types. (Note: this is pretty questionable; if
3338 : * the ORDER BY list doesn't include all the DISTINCT ON items and more
3339 : * besides, you certainly aren't using DISTINCT ON in the intended way,
3340 : * and you probably aren't going to get consistent results. It might be
3341 : * better to throw an error or warning here. But historically we've
3342 : * allowed it, so keep doing so.)
3343 : */
3344 401 : forboth(lc, distinctlist, lc2, sortgrouprefs)
3345 : {
3346 241 : Node *dexpr = (Node *) lfirst(lc);
3347 241 : int sortgroupref = lfirst_int(lc2);
3348 241 : TargetEntry *tle = get_sortgroupref_tle(sortgroupref, *targetlist);
3349 :
3350 241 : if (targetIsInSortList(tle, InvalidOid, result))
3351 160 : continue; /* already in list (with some semantics) */
3352 81 : if (skipped_sortitem)
3353 0 : ereport(ERROR,
3354 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3355 : errmsg("SELECT DISTINCT ON expressions must match initial ORDER BY expressions"),
3356 : parser_errposition(pstate, exprLocation(dexpr))));
3357 81 : result = addTargetToGroupList(pstate, tle,
3358 : result, *targetlist,
3359 : exprLocation(dexpr));
3360 : }
3361 :
3362 : /*
3363 : * An empty result list is impossible here because of grammar
3364 : * restrictions.
3365 : */
3366 : Assert(result != NIL);
3367 :
3368 160 : return result;
3369 : }
3370 :
3371 : /*
3372 : * get_matching_location
3373 : * Get the exprLocation of the exprs member corresponding to the
3374 : * (first) member of sortgrouprefs that equals sortgroupref.
3375 : *
3376 : * This is used so that we can point at a troublesome DISTINCT ON entry.
3377 : * (Note that we need to use the original untransformed DISTINCT ON list
3378 : * item, as whatever TLE it corresponds to will very possibly have a
3379 : * parse location pointing to some matching entry in the SELECT list
3380 : * or ORDER BY list.)
3381 : */
3382 : static int
3383 4 : get_matching_location(int sortgroupref, List *sortgrouprefs, List *exprs)
3384 : {
3385 : ListCell *lcs;
3386 : ListCell *lce;
3387 :
3388 8 : forboth(lcs, sortgrouprefs, lce, exprs)
3389 : {
3390 8 : if (lfirst_int(lcs) == sortgroupref)
3391 4 : return exprLocation((Node *) lfirst(lce));
3392 : }
3393 : /* if no match, caller blew it */
3394 0 : elog(ERROR, "get_matching_location: no matching sortgroupref");
3395 : return -1; /* keep compiler quiet */
3396 : }
3397 :
3398 : /*
3399 : * resolve_unique_index_expr
3400 : * Infer a unique index from a list of indexElems, for ON
3401 : * CONFLICT clause
3402 : *
3403 : * Perform parse analysis of expressions and columns appearing within ON
3404 : * CONFLICT clause. During planning, the returned list of expressions is used
3405 : * to infer which unique index to use.
3406 : */
3407 : static List *
3408 1266 : resolve_unique_index_expr(ParseState *pstate, InferClause *infer,
3409 : Relation heapRel)
3410 : {
3411 1266 : List *result = NIL;
3412 : ListCell *l;
3413 :
3414 2822 : foreach(l, infer->indexElems)
3415 : {
3416 1572 : IndexElem *ielem = (IndexElem *) lfirst(l);
3417 1572 : InferenceElem *pInfer = makeNode(InferenceElem);
3418 : Node *parse;
3419 :
3420 : /*
3421 : * Raw grammar re-uses CREATE INDEX infrastructure for unique index
3422 : * inference clause, and so will accept opclasses by name and so on.
3423 : *
3424 : * Make no attempt to match ASC or DESC ordering, NULLS FIRST/NULLS
3425 : * LAST ordering or opclass options, since those are not significant
3426 : * for inference purposes (any unique index matching the inference
3427 : * specification in other regards is accepted indifferently). Actively
3428 : * reject this as wrong-headed.
3429 : */
3430 1572 : if (ielem->ordering != SORTBY_DEFAULT)
3431 4 : ereport(ERROR,
3432 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3433 : errmsg("%s is not allowed in ON CONFLICT clause",
3434 : "ASC/DESC"),
3435 : parser_errposition(pstate, ielem->location)));
3436 1568 : if (ielem->nulls_ordering != SORTBY_NULLS_DEFAULT)
3437 4 : ereport(ERROR,
3438 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3439 : errmsg("%s is not allowed in ON CONFLICT clause",
3440 : "NULLS FIRST/LAST"),
3441 : parser_errposition(pstate, ielem->location)));
3442 1564 : if (ielem->opclassopts)
3443 4 : ereport(ERROR,
3444 : errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3445 : errmsg("operator class options are not allowed in ON CONFLICT clause"),
3446 : parser_errposition(pstate, ielem->location));
3447 :
3448 1560 : if (!ielem->expr)
3449 : {
3450 : /* Simple index attribute */
3451 : ColumnRef *n;
3452 :
3453 : /*
3454 : * Grammar won't have built raw expression for us in event of
3455 : * plain column reference. Create one directly, and perform
3456 : * expression transformation. Planner expects this, and performs
3457 : * its own normalization for the purposes of matching against
3458 : * pg_index.
3459 : */
3460 1446 : n = makeNode(ColumnRef);
3461 1446 : n->fields = list_make1(makeString(ielem->name));
3462 : /* Location is approximately that of inference specification */
3463 1446 : n->location = infer->location;
3464 1446 : parse = (Node *) n;
3465 : }
3466 : else
3467 : {
3468 : /* Do parse transformation of the raw expression */
3469 114 : parse = (Node *) ielem->expr;
3470 : }
3471 :
3472 : /*
3473 : * transformExpr() will reject subqueries, aggregates, window
3474 : * functions, and SRFs, based on being passed
3475 : * EXPR_KIND_INDEX_EXPRESSION. So we needn't worry about those
3476 : * further ... not that they would match any available index
3477 : * expression anyway.
3478 : */
3479 1560 : pInfer->expr = transformExpr(pstate, parse, EXPR_KIND_INDEX_EXPRESSION);
3480 :
3481 : /* Perform lookup of collation and operator class as required */
3482 1556 : if (!ielem->collation)
3483 1528 : pInfer->infercollid = InvalidOid;
3484 : else
3485 28 : pInfer->infercollid = LookupCollation(pstate, ielem->collation,
3486 : ielem->location);
3487 :
3488 1556 : if (!ielem->opclass)
3489 1528 : pInfer->inferopclass = InvalidOid;
3490 : else
3491 28 : pInfer->inferopclass = get_opclass_oid(BTREE_AM_OID,
3492 : ielem->opclass, false);
3493 :
3494 1556 : result = lappend(result, pInfer);
3495 : }
3496 :
3497 1250 : return result;
3498 : }
3499 :
3500 : /*
3501 : * transformOnConflictArbiter -
3502 : * transform arbiter expressions in an ON CONFLICT clause.
3503 : *
3504 : * Transformed expressions used to infer one unique index relation to serve as
3505 : * an ON CONFLICT arbiter. Partial unique indexes may be inferred using WHERE
3506 : * clause from inference specification clause.
3507 : */
3508 : void
3509 1556 : transformOnConflictArbiter(ParseState *pstate,
3510 : OnConflictClause *onConflictClause,
3511 : List **arbiterExpr, Node **arbiterWhere,
3512 : Oid *constraint)
3513 : {
3514 1556 : InferClause *infer = onConflictClause->infer;
3515 :
3516 1556 : *arbiterExpr = NIL;
3517 1556 : *arbiterWhere = NULL;
3518 1556 : *constraint = InvalidOid;
3519 :
3520 1556 : if ((onConflictClause->action == ONCONFLICT_UPDATE ||
3521 1556 : onConflictClause->action == ONCONFLICT_SELECT) && !infer)
3522 4 : ereport(ERROR,
3523 : errcode(ERRCODE_SYNTAX_ERROR),
3524 : errmsg("ON CONFLICT DO %s requires inference specification or constraint name",
3525 : onConflictClause->action == ONCONFLICT_UPDATE ? "UPDATE" : "SELECT"),
3526 : errhint("For example, ON CONFLICT (column_name)."),
3527 : parser_errposition(pstate,
3528 : exprLocation((Node *) onConflictClause)));
3529 :
3530 : /*
3531 : * To simplify certain aspects of its design, speculative insertion into
3532 : * system catalogs is disallowed
3533 : */
3534 1552 : if (IsCatalogRelation(pstate->p_target_relation))
3535 0 : ereport(ERROR,
3536 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3537 : errmsg("ON CONFLICT is not supported with system catalog tables"),
3538 : parser_errposition(pstate,
3539 : exprLocation((Node *) onConflictClause))));
3540 :
3541 : /* Same applies to table used by logical decoding as catalog table */
3542 1552 : if (RelationIsUsedAsCatalogTable(pstate->p_target_relation))
3543 0 : ereport(ERROR,
3544 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3545 : errmsg("ON CONFLICT is not supported on table \"%s\" used as a catalog table",
3546 : RelationGetRelationName(pstate->p_target_relation)),
3547 : parser_errposition(pstate,
3548 : exprLocation((Node *) onConflictClause))));
3549 :
3550 : /* ON CONFLICT DO NOTHING does not require an inference clause */
3551 1552 : if (infer)
3552 : {
3553 1404 : if (infer->indexElems)
3554 1266 : *arbiterExpr = resolve_unique_index_expr(pstate, infer,
3555 : pstate->p_target_relation);
3556 :
3557 : /*
3558 : * Handling inference WHERE clause (for partial unique index
3559 : * inference)
3560 : */
3561 1388 : if (infer->whereClause)
3562 34 : *arbiterWhere = transformExpr(pstate, infer->whereClause,
3563 : EXPR_KIND_INDEX_PREDICATE);
3564 :
3565 : /*
3566 : * If the arbiter is specified by constraint name, get the constraint
3567 : * OID and mark the constrained columns as requiring SELECT privilege,
3568 : * in the same way as would have happened if the arbiter had been
3569 : * specified by explicit reference to the constraint's index columns.
3570 : */
3571 1388 : if (infer->conname)
3572 : {
3573 138 : Oid relid = RelationGetRelid(pstate->p_target_relation);
3574 138 : RTEPermissionInfo *perminfo = pstate->p_target_nsitem->p_perminfo;
3575 : Bitmapset *conattnos;
3576 :
3577 138 : conattnos = get_relation_constraint_attnos(relid, infer->conname,
3578 : false, constraint);
3579 :
3580 : /* Make sure the rel as a whole is marked for SELECT access */
3581 138 : perminfo->requiredPerms |= ACL_SELECT;
3582 : /* Mark the constrained columns as requiring SELECT access */
3583 138 : perminfo->selectedCols = bms_add_members(perminfo->selectedCols,
3584 : conattnos);
3585 : }
3586 : }
3587 :
3588 : /*
3589 : * It's convenient to form a list of expressions based on the
3590 : * representation used by CREATE INDEX, since the same restrictions are
3591 : * appropriate (e.g. on subqueries). However, from here on, a dedicated
3592 : * primnode representation is used for inference elements, and so
3593 : * assign_query_collations() can be trusted to do the right thing with the
3594 : * post parse analysis query tree inference clause representation.
3595 : */
3596 1536 : }
3597 :
3598 : /*
3599 : * addTargetToSortList
3600 : * If the given targetlist entry isn't already in the SortGroupClause
3601 : * list, add it to the end of the list, using the given sort ordering
3602 : * info.
3603 : *
3604 : * Returns the updated SortGroupClause list.
3605 : */
3606 : List *
3607 76153 : addTargetToSortList(ParseState *pstate, TargetEntry *tle,
3608 : List *sortlist, List *targetlist, SortBy *sortby)
3609 : {
3610 76153 : Oid restype = exprType((Node *) tle->expr);
3611 : Oid sortop;
3612 : Oid eqop;
3613 : bool hashable;
3614 : bool reverse;
3615 : int location;
3616 : ParseCallbackState pcbstate;
3617 :
3618 : /* if tlist item is an UNKNOWN literal, change it to TEXT */
3619 76153 : if (restype == UNKNOWNOID)
3620 : {
3621 8 : tle->expr = (Expr *) coerce_type(pstate, (Node *) tle->expr,
3622 : restype, TEXTOID, -1,
3623 : COERCION_IMPLICIT,
3624 : COERCE_IMPLICIT_CAST,
3625 : -1);
3626 8 : restype = TEXTOID;
3627 : }
3628 :
3629 : /*
3630 : * Rather than clutter the API of get_sort_group_operators and the other
3631 : * functions we're about to use, make use of error context callback to
3632 : * mark any error reports with a parse position. We point to the operator
3633 : * location if present, else to the expression being sorted. (NB: use the
3634 : * original untransformed expression here; the TLE entry might well point
3635 : * at a duplicate expression in the regular SELECT list.)
3636 : */
3637 76153 : location = sortby->location;
3638 76153 : if (location < 0)
3639 76011 : location = exprLocation(sortby->node);
3640 76153 : setup_parser_errposition_callback(&pcbstate, pstate, location);
3641 :
3642 : /* determine the sortop, eqop, and directionality */
3643 76153 : switch (sortby->sortby_dir)
3644 : {
3645 73652 : case SORTBY_DEFAULT:
3646 : case SORTBY_ASC:
3647 73652 : get_sort_group_operators(restype,
3648 : true, true, false,
3649 : &sortop, &eqop, NULL,
3650 : &hashable);
3651 73648 : reverse = false;
3652 73648 : break;
3653 2359 : case SORTBY_DESC:
3654 2359 : get_sort_group_operators(restype,
3655 : false, true, true,
3656 : NULL, &eqop, &sortop,
3657 : &hashable);
3658 2359 : reverse = true;
3659 2359 : break;
3660 142 : case SORTBY_USING:
3661 : Assert(sortby->useOp != NIL);
3662 142 : sortop = compatible_oper_opid(sortby->useOp,
3663 : restype,
3664 : restype,
3665 : false);
3666 :
3667 : /*
3668 : * Verify it's a valid ordering operator, fetch the corresponding
3669 : * equality operator, and determine whether to consider it like
3670 : * ASC or DESC.
3671 : */
3672 142 : eqop = get_equality_op_for_ordering_op(sortop, &reverse);
3673 142 : if (!OidIsValid(eqop))
3674 0 : ereport(ERROR,
3675 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
3676 : errmsg("operator %s is not a valid ordering operator",
3677 : strVal(llast(sortby->useOp))),
3678 : errhint("Ordering operators must be \"<\" or \">\" members of btree operator families.")));
3679 :
3680 : /*
3681 : * Also see if the equality operator is hashable.
3682 : */
3683 142 : hashable = op_hashjoinable(eqop, restype);
3684 142 : break;
3685 0 : default:
3686 0 : elog(ERROR, "unrecognized sortby_dir: %d", sortby->sortby_dir);
3687 : sortop = InvalidOid; /* keep compiler quiet */
3688 : eqop = InvalidOid;
3689 : hashable = false;
3690 : reverse = false;
3691 : break;
3692 : }
3693 :
3694 76149 : cancel_parser_errposition_callback(&pcbstate);
3695 :
3696 : /* avoid making duplicate sortlist entries */
3697 76149 : if (!targetIsInSortList(tle, sortop, sortlist))
3698 : {
3699 76149 : SortGroupClause *sortcl = makeNode(SortGroupClause);
3700 :
3701 76149 : sortcl->tleSortGroupRef = assignSortGroupRef(tle, targetlist);
3702 :
3703 76149 : sortcl->eqop = eqop;
3704 76149 : sortcl->sortop = sortop;
3705 76149 : sortcl->hashable = hashable;
3706 76149 : sortcl->reverse_sort = reverse;
3707 :
3708 76149 : switch (sortby->sortby_nulls)
3709 : {
3710 75056 : case SORTBY_NULLS_DEFAULT:
3711 : /* NULLS FIRST is default for DESC; other way for ASC */
3712 75056 : sortcl->nulls_first = reverse;
3713 75056 : break;
3714 204 : case SORTBY_NULLS_FIRST:
3715 204 : sortcl->nulls_first = true;
3716 204 : break;
3717 889 : case SORTBY_NULLS_LAST:
3718 889 : sortcl->nulls_first = false;
3719 889 : break;
3720 0 : default:
3721 0 : elog(ERROR, "unrecognized sortby_nulls: %d",
3722 : sortby->sortby_nulls);
3723 : break;
3724 : }
3725 :
3726 76149 : sortlist = lappend(sortlist, sortcl);
3727 : }
3728 :
3729 76149 : return sortlist;
3730 : }
3731 :
3732 : /*
3733 : * addTargetToGroupList
3734 : * If the given targetlist entry isn't already in the SortGroupClause
3735 : * list, add it to the end of the list, using default sort/group
3736 : * semantics.
3737 : *
3738 : * This is very similar to addTargetToSortList, except that we allow the
3739 : * case where only a grouping (equality) operator can be found, and that
3740 : * the TLE is considered "already in the list" if it appears there with any
3741 : * sorting semantics.
3742 : *
3743 : * location is the parse location to be fingered in event of trouble. Note
3744 : * that we can't rely on exprLocation(tle->expr), because that might point
3745 : * to a SELECT item that matches the GROUP BY item; it'd be pretty confusing
3746 : * to report such a location.
3747 : *
3748 : * Returns the updated SortGroupClause list.
3749 : */
3750 : static List *
3751 13114 : addTargetToGroupList(ParseState *pstate, TargetEntry *tle,
3752 : List *grouplist, List *targetlist, int location)
3753 : {
3754 13114 : Oid restype = exprType((Node *) tle->expr);
3755 :
3756 : /* if tlist item is an UNKNOWN literal, change it to TEXT */
3757 13114 : if (restype == UNKNOWNOID)
3758 : {
3759 10 : tle->expr = (Expr *) coerce_type(pstate, (Node *) tle->expr,
3760 : restype, TEXTOID, -1,
3761 : COERCION_IMPLICIT,
3762 : COERCE_IMPLICIT_CAST,
3763 : -1);
3764 10 : restype = TEXTOID;
3765 : }
3766 :
3767 : /* avoid making duplicate grouplist entries */
3768 13114 : if (!targetIsInSortList(tle, InvalidOid, grouplist))
3769 : {
3770 12715 : SortGroupClause *grpcl = makeNode(SortGroupClause);
3771 : Oid sortop;
3772 : Oid eqop;
3773 : bool hashable;
3774 : ParseCallbackState pcbstate;
3775 :
3776 12715 : setup_parser_errposition_callback(&pcbstate, pstate, location);
3777 :
3778 : /* determine the eqop and optional sortop */
3779 12715 : get_sort_group_operators(restype,
3780 : false, true, false,
3781 : &sortop, &eqop, NULL,
3782 : &hashable);
3783 :
3784 12715 : cancel_parser_errposition_callback(&pcbstate);
3785 :
3786 12715 : grpcl->tleSortGroupRef = assignSortGroupRef(tle, targetlist);
3787 12715 : grpcl->eqop = eqop;
3788 12715 : grpcl->sortop = sortop;
3789 12715 : grpcl->reverse_sort = false; /* sortop is "less than", or
3790 : * InvalidOid */
3791 12715 : grpcl->nulls_first = false; /* OK with or without sortop */
3792 12715 : grpcl->hashable = hashable;
3793 :
3794 12715 : grouplist = lappend(grouplist, grpcl);
3795 : }
3796 :
3797 13114 : return grouplist;
3798 : }
3799 :
3800 : /*
3801 : * assignSortGroupRef
3802 : * Assign the targetentry an unused ressortgroupref, if it doesn't
3803 : * already have one. Return the assigned or pre-existing refnumber.
3804 : *
3805 : * 'tlist' is the targetlist containing (or to contain) the given targetentry.
3806 : */
3807 : Index
3808 129503 : assignSortGroupRef(TargetEntry *tle, List *tlist)
3809 : {
3810 : Index maxRef;
3811 : ListCell *l;
3812 :
3813 129503 : if (tle->ressortgroupref) /* already has one? */
3814 4394 : return tle->ressortgroupref;
3815 :
3816 : /* easiest way to pick an unused refnumber: max used + 1 */
3817 125109 : maxRef = 0;
3818 722357 : foreach(l, tlist)
3819 : {
3820 597248 : Index ref = ((TargetEntry *) lfirst(l))->ressortgroupref;
3821 :
3822 597248 : if (ref > maxRef)
3823 102226 : maxRef = ref;
3824 : }
3825 125109 : tle->ressortgroupref = maxRef + 1;
3826 125109 : return tle->ressortgroupref;
3827 : }
3828 :
3829 : /*
3830 : * targetIsInSortList
3831 : * Is the given target item already in the sortlist?
3832 : * If sortop is not InvalidOid, also test for a match to the sortop.
3833 : *
3834 : * It is not an oversight that this function ignores the nulls_first flag.
3835 : * We check sortop when determining if an ORDER BY item is redundant with
3836 : * earlier ORDER BY items, because it's conceivable that "ORDER BY
3837 : * foo USING <, foo USING <<<" is not redundant, if <<< distinguishes
3838 : * values that < considers equal. We need not check nulls_first
3839 : * however, because a lower-order column with the same sortop but
3840 : * opposite nulls direction is redundant. Also, we can consider
3841 : * ORDER BY foo ASC, foo DESC redundant, so check for a commutator match.
3842 : *
3843 : * Works for both ordering and grouping lists (sortop would normally be
3844 : * InvalidOid when considering grouping). Note that the main reason we need
3845 : * this routine (and not just a quick test for nonzeroness of ressortgroupref)
3846 : * is that a TLE might be in only one of the lists.
3847 : */
3848 : bool
3849 93297 : targetIsInSortList(TargetEntry *tle, Oid sortop, List *sortList)
3850 : {
3851 93297 : Index ref = tle->ressortgroupref;
3852 : ListCell *l;
3853 :
3854 : /* no need to scan list if tle has no marker */
3855 93297 : if (ref == 0)
3856 89312 : return false;
3857 :
3858 4905 : foreach(l, sortList)
3859 : {
3860 2373 : SortGroupClause *scl = (SortGroupClause *) lfirst(l);
3861 :
3862 2373 : if (scl->tleSortGroupRef == ref &&
3863 0 : (sortop == InvalidOid ||
3864 0 : sortop == scl->sortop ||
3865 0 : sortop == get_commutator(scl->sortop)))
3866 1453 : return true;
3867 : }
3868 2532 : return false;
3869 : }
3870 :
3871 : /*
3872 : * findWindowClause
3873 : * Find the named WindowClause in the list, or return NULL if not there
3874 : */
3875 : static WindowClause *
3876 454 : findWindowClause(List *wclist, const char *name)
3877 : {
3878 : ListCell *l;
3879 :
3880 470 : foreach(l, wclist)
3881 : {
3882 48 : WindowClause *wc = (WindowClause *) lfirst(l);
3883 :
3884 48 : if (wc->name && strcmp(wc->name, name) == 0)
3885 32 : return wc;
3886 : }
3887 :
3888 422 : return NULL;
3889 : }
3890 :
3891 : /*
3892 : * transformFrameOffset
3893 : * Process a window frame offset expression
3894 : *
3895 : * In RANGE mode, rangeopfamily is the sort opfamily for the input ORDER BY
3896 : * column, and rangeopcintype is the input data type the sort operator is
3897 : * registered with. We expect the in_range function to be registered with
3898 : * that same type. (In binary-compatible cases, it might be different from
3899 : * the input column's actual type, so we can't use that for the lookups.)
3900 : * We'll return the OID of the in_range function to *inRangeFunc.
3901 : */
3902 : static Node *
3903 4092 : transformFrameOffset(ParseState *pstate, int frameOptions,
3904 : Oid rangeopfamily, Oid rangeopcintype, Oid *inRangeFunc,
3905 : Node *clause)
3906 : {
3907 4092 : const char *constructName = NULL;
3908 : Node *node;
3909 :
3910 4092 : *inRangeFunc = InvalidOid; /* default result */
3911 :
3912 : /* Quick exit if no offset expression */
3913 4092 : if (clause == NULL)
3914 2778 : return NULL;
3915 :
3916 1314 : if (frameOptions & FRAMEOPTION_ROWS)
3917 : {
3918 : /* Transform the raw expression tree */
3919 350 : node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_ROWS);
3920 :
3921 : /*
3922 : * Like LIMIT clause, simply coerce to int8
3923 : */
3924 350 : constructName = "ROWS";
3925 350 : node = coerce_to_specific_type(pstate, node, INT8OID, constructName);
3926 : }
3927 964 : else if (frameOptions & FRAMEOPTION_RANGE)
3928 : {
3929 : /*
3930 : * We must look up the in_range support function that's to be used,
3931 : * possibly choosing one of several, and coerce the "offset" value to
3932 : * the appropriate input type.
3933 : */
3934 : Oid nodeType;
3935 : Oid preferredType;
3936 768 : int nfuncs = 0;
3937 768 : int nmatches = 0;
3938 768 : Oid selectedType = InvalidOid;
3939 768 : Oid selectedFunc = InvalidOid;
3940 : CatCList *proclist;
3941 : int i;
3942 :
3943 : /* Transform the raw expression tree */
3944 768 : node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_RANGE);
3945 768 : nodeType = exprType(node);
3946 :
3947 : /*
3948 : * If there are multiple candidates, we'll prefer the one that exactly
3949 : * matches nodeType; or if nodeType is as yet unknown, prefer the one
3950 : * that exactly matches the sort column type. (The second rule is
3951 : * like what we do for "known_type operator unknown".)
3952 : */
3953 768 : preferredType = (nodeType != UNKNOWNOID) ? nodeType : rangeopcintype;
3954 :
3955 : /* Find the in_range support functions applicable to this case */
3956 768 : proclist = SearchSysCacheList2(AMPROCNUM,
3957 : ObjectIdGetDatum(rangeopfamily),
3958 : ObjectIdGetDatum(rangeopcintype));
3959 5340 : for (i = 0; i < proclist->n_members; i++)
3960 : {
3961 4572 : HeapTuple proctup = &proclist->members[i]->tuple;
3962 4572 : Form_pg_amproc procform = (Form_pg_amproc) GETSTRUCT(proctup);
3963 :
3964 : /* The search will find all support proc types; ignore others */
3965 4572 : if (procform->amprocnum != BTINRANGE_PROC)
3966 3400 : continue;
3967 1172 : nfuncs++;
3968 :
3969 : /* Ignore function if given value can't be coerced to that type */
3970 1172 : if (!can_coerce_type(1, &nodeType, &procform->amprocrighttype,
3971 : COERCION_IMPLICIT))
3972 220 : continue;
3973 952 : nmatches++;
3974 :
3975 : /* Remember preferred match, or any match if didn't find that */
3976 952 : if (selectedType != preferredType)
3977 : {
3978 912 : selectedType = procform->amprocrighttype;
3979 912 : selectedFunc = procform->amproc;
3980 : }
3981 : }
3982 768 : ReleaseCatCacheList(proclist);
3983 :
3984 : /*
3985 : * Throw error if needed. It seems worth taking the trouble to
3986 : * distinguish "no support at all" from "you didn't match any
3987 : * available offset type".
3988 : */
3989 768 : if (nfuncs == 0)
3990 4 : ereport(ERROR,
3991 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3992 : errmsg("RANGE with offset PRECEDING/FOLLOWING is not supported for column type %s",
3993 : format_type_be(rangeopcintype)),
3994 : parser_errposition(pstate, exprLocation(node))));
3995 764 : if (nmatches == 0)
3996 12 : ereport(ERROR,
3997 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3998 : errmsg("RANGE with offset PRECEDING/FOLLOWING is not supported 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 an appropriate type."),
4002 : parser_errposition(pstate, exprLocation(node))));
4003 752 : if (nmatches != 1 && selectedType != preferredType)
4004 0 : ereport(ERROR,
4005 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
4006 : errmsg("RANGE with offset PRECEDING/FOLLOWING has multiple interpretations for column type %s and offset type %s",
4007 : format_type_be(rangeopcintype),
4008 : format_type_be(nodeType)),
4009 : errhint("Cast the offset value to the exact intended type."),
4010 : parser_errposition(pstate, exprLocation(node))));
4011 :
4012 : /* OK, coerce the offset to the right type */
4013 752 : constructName = "RANGE";
4014 752 : node = coerce_to_specific_type(pstate, node,
4015 : selectedType, constructName);
4016 752 : *inRangeFunc = selectedFunc;
4017 : }
4018 196 : else if (frameOptions & FRAMEOPTION_GROUPS)
4019 : {
4020 : /* Transform the raw expression tree */
4021 196 : node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_GROUPS);
4022 :
4023 : /*
4024 : * Like LIMIT clause, simply coerce to int8
4025 : */
4026 196 : constructName = "GROUPS";
4027 196 : node = coerce_to_specific_type(pstate, node, INT8OID, constructName);
4028 : }
4029 : else
4030 : {
4031 : Assert(false);
4032 0 : node = NULL;
4033 : }
4034 :
4035 : /* Disallow variables in frame offsets */
4036 1298 : checkExprIsVarFree(pstate, node, constructName);
4037 :
4038 1294 : return node;
4039 : }
|
