Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * parse_clause.c
4 : * handle clauses in parser
5 : *
6 : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/parser/parse_clause.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 :
16 : #include "postgres.h"
17 :
18 : #include "access/htup_details.h"
19 : #include "access/nbtree.h"
20 : #include "access/table.h"
21 : #include "access/tsmapi.h"
22 : #include "catalog/catalog.h"
23 : #include "catalog/pg_am.h"
24 : #include "catalog/pg_amproc.h"
25 : #include "catalog/pg_constraint.h"
26 : #include "catalog/pg_type.h"
27 : #include "commands/defrem.h"
28 : #include "miscadmin.h"
29 : #include "nodes/makefuncs.h"
30 : #include "nodes/nodeFuncs.h"
31 : #include "optimizer/optimizer.h"
32 : #include "parser/analyze.h"
33 : #include "parser/parse_clause.h"
34 : #include "parser/parse_coerce.h"
35 : #include "parser/parse_collate.h"
36 : #include "parser/parse_expr.h"
37 : #include "parser/parse_func.h"
38 : #include "parser/parse_oper.h"
39 : #include "parser/parse_relation.h"
40 : #include "parser/parse_target.h"
41 : #include "parser/parse_type.h"
42 : #include "parser/parser.h"
43 : #include "parser/parsetree.h"
44 : #include "rewrite/rewriteManip.h"
45 : #include "utils/builtins.h"
46 : #include "utils/catcache.h"
47 : #include "utils/guc.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 TableSampleClause *transformRangeTableSample(ParseState *pstate,
71 : RangeTableSample *rts);
72 : static ParseNamespaceItem *getNSItemForSpecialRelationTypes(ParseState *pstate,
73 : RangeVar *rv);
74 : static Node *transformFromClauseItem(ParseState *pstate, Node *n,
75 : ParseNamespaceItem **top_nsitem,
76 : List **namespace);
77 : static Var *buildVarFromNSColumn(ParseState *pstate,
78 : ParseNamespaceColumn *nscol);
79 : static Node *buildMergedJoinVar(ParseState *pstate, JoinType jointype,
80 : Var *l_colvar, Var *r_colvar);
81 : static void markRelsAsNulledBy(ParseState *pstate, Node *n, int jindex);
82 : static void setNamespaceColumnVisibility(List *namespace, bool cols_visible);
83 : static void setNamespaceLateralState(List *namespace,
84 : bool lateral_only, bool lateral_ok);
85 : static void checkExprIsVarFree(ParseState *pstate, Node *n,
86 : const char *constructName);
87 : static TargetEntry *findTargetlistEntrySQL92(ParseState *pstate, Node *node,
88 : List **tlist, ParseExprKind exprKind);
89 : static TargetEntry *findTargetlistEntrySQL99(ParseState *pstate, Node *node,
90 : List **tlist, ParseExprKind exprKind);
91 : static int get_matching_location(int sortgroupref,
92 : List *sortgrouprefs, List *exprs);
93 : static List *resolve_unique_index_expr(ParseState *pstate, InferClause *infer,
94 : Relation heapRel);
95 : static List *addTargetToGroupList(ParseState *pstate, TargetEntry *tle,
96 : List *grouplist, List *targetlist, int location);
97 : static WindowClause *findWindowClause(List *wclist, const char *name);
98 : static Node *transformFrameOffset(ParseState *pstate, int frameOptions,
99 : Oid rangeopfamily, Oid rangeopcintype, Oid *inRangeFunc,
100 : Node *clause);
101 :
102 :
103 : /*
104 : * transformFromClause -
105 : * Process the FROM clause and add items to the query's range table,
106 : * joinlist, and namespace.
107 : *
108 : * Note: we assume that the pstate's p_rtable, p_joinlist, and p_namespace
109 : * lists were initialized to NIL when the pstate was created.
110 : * We will add onto any entries already present --- this is needed for rule
111 : * processing, as well as for UPDATE and DELETE.
112 : */
113 : void
114 462036 : transformFromClause(ParseState *pstate, List *frmList)
115 : {
116 : ListCell *fl;
117 :
118 : /*
119 : * The grammar will have produced a list of RangeVars, RangeSubselects,
120 : * RangeFunctions, and/or JoinExprs. Transform each one (possibly adding
121 : * entries to the rtable), check for duplicate refnames, and then add it
122 : * to the joinlist and namespace.
123 : *
124 : * Note we must process the items left-to-right for proper handling of
125 : * LATERAL references.
126 : */
127 785652 : foreach(fl, frmList)
128 : {
129 324304 : Node *n = lfirst(fl);
130 : ParseNamespaceItem *nsitem;
131 : List *namespace;
132 :
133 324304 : n = transformFromClauseItem(pstate, n,
134 : &nsitem,
135 : &namespace);
136 :
137 323622 : checkNameSpaceConflicts(pstate, pstate->p_namespace, namespace);
138 :
139 : /* Mark the new namespace items as visible only to LATERAL */
140 323616 : setNamespaceLateralState(namespace, true, true);
141 :
142 323616 : pstate->p_joinlist = lappend(pstate->p_joinlist, n);
143 323616 : pstate->p_namespace = list_concat(pstate->p_namespace, namespace);
144 : }
145 :
146 : /*
147 : * We're done parsing the FROM list, so make all namespace items
148 : * unconditionally visible. Note that this will also reset lateral_only
149 : * for any namespace items that were already present when we were called;
150 : * but those should have been that way already.
151 : */
152 461348 : setNamespaceLateralState(pstate->p_namespace, false, true);
153 461348 : }
154 :
155 : /*
156 : * setTargetTable
157 : * Add the target relation of INSERT/UPDATE/DELETE/MERGE to the range table,
158 : * and make the special links to it in the ParseState.
159 : *
160 : * We also open the target relation and acquire a write lock on it.
161 : * This must be done before processing the FROM list, in case the target
162 : * is also mentioned as a source relation --- we want to be sure to grab
163 : * the write lock before any read lock.
164 : *
165 : * If alsoSource is true, add the target to the query's joinlist and
166 : * namespace. For INSERT, we don't want the target to be joined to;
167 : * it's a destination of tuples, not a source. MERGE is actually
168 : * both, but we'll add it separately to joinlist and namespace, so
169 : * doing nothing (like INSERT) is correct here. For UPDATE/DELETE,
170 : * we do need to scan or join the target. (NOTE: we do not bother
171 : * to check for namespace conflict; we assume that the namespace was
172 : * initially empty in these cases.)
173 : *
174 : * Finally, we mark the relation as requiring the permissions specified
175 : * by requiredPerms.
176 : *
177 : * Returns the rangetable index of the target relation.
178 : */
179 : int
180 91620 : setTargetTable(ParseState *pstate, RangeVar *relation,
181 : bool inh, bool alsoSource, AclMode requiredPerms)
182 : {
183 : ParseNamespaceItem *nsitem;
184 :
185 : /*
186 : * ENRs hide tables of the same name, so we need to check for them first.
187 : * In contrast, CTEs don't hide tables (for this purpose).
188 : */
189 172036 : if (relation->schemaname == NULL &&
190 80416 : scanNameSpaceForENR(pstate, relation->relname))
191 6 : ereport(ERROR,
192 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
193 : errmsg("relation \"%s\" cannot be the target of a modifying statement",
194 : relation->relname)));
195 :
196 : /* Close old target; this could only happen for multi-action rules */
197 91614 : if (pstate->p_target_relation != NULL)
198 0 : table_close(pstate->p_target_relation, NoLock);
199 :
200 : /*
201 : * Open target rel and grab suitable lock (which we will hold till end of
202 : * transaction).
203 : *
204 : * free_parsestate() will eventually do the corresponding table_close(),
205 : * but *not* release the lock.
206 : */
207 91614 : pstate->p_target_relation = parserOpenTable(pstate, relation,
208 : RowExclusiveLock);
209 :
210 : /*
211 : * Now build an RTE and a ParseNamespaceItem.
212 : */
213 91594 : nsitem = addRangeTableEntryForRelation(pstate, pstate->p_target_relation,
214 : RowExclusiveLock,
215 : relation->alias, inh, false);
216 :
217 : /* remember the RTE/nsitem as being the query target */
218 91594 : pstate->p_target_nsitem = nsitem;
219 :
220 : /*
221 : * Override addRangeTableEntry's default ACL_SELECT permissions check, and
222 : * instead mark target table as requiring exactly the specified
223 : * permissions.
224 : *
225 : * If we find an explicit reference to the rel later during parse
226 : * analysis, we will add the ACL_SELECT bit back again; see
227 : * markVarForSelectPriv and its callers.
228 : */
229 91594 : nsitem->p_perminfo->requiredPerms = requiredPerms;
230 :
231 : /*
232 : * If UPDATE/DELETE, add table to joinlist and namespace.
233 : */
234 91594 : if (alsoSource)
235 17568 : addNSItemToQuery(pstate, nsitem, true, true, true);
236 :
237 91594 : return nsitem->p_rtindex;
238 : }
239 :
240 : /*
241 : * Extract all not-in-common columns from column lists of a source table
242 : *
243 : * src_nscolumns and src_colnames describe the source table.
244 : *
245 : * *src_colnos initially contains the column numbers of the already-merged
246 : * columns. We add to it the column number of each additional column.
247 : * Also append to *res_colnames the name of each additional column,
248 : * append to *res_colvars a Var for each additional column, and copy the
249 : * columns' nscolumns data into res_nscolumns[] (which is caller-allocated
250 : * space that had better be big enough).
251 : *
252 : * Returns the number of columns added.
253 : */
254 : static int
255 147500 : extractRemainingColumns(ParseState *pstate,
256 : ParseNamespaceColumn *src_nscolumns,
257 : List *src_colnames,
258 : List **src_colnos,
259 : List **res_colnames, List **res_colvars,
260 : ParseNamespaceColumn *res_nscolumns)
261 : {
262 147500 : int colcount = 0;
263 : Bitmapset *prevcols;
264 : int attnum;
265 : ListCell *lc;
266 :
267 : /*
268 : * While we could just test "list_member_int(*src_colnos, attnum)" to
269 : * detect already-merged columns in the loop below, that would be O(N^2)
270 : * for a wide input table. Instead build a bitmapset of just the merged
271 : * USING columns, which we won't add to within the main loop.
272 : */
273 147500 : prevcols = NULL;
274 150816 : foreach(lc, *src_colnos)
275 : {
276 3316 : prevcols = bms_add_member(prevcols, lfirst_int(lc));
277 : }
278 :
279 147500 : attnum = 0;
280 3333302 : foreach(lc, src_colnames)
281 : {
282 3185802 : char *colname = strVal(lfirst(lc));
283 :
284 3185802 : attnum++;
285 : /* Non-dropped and not already merged? */
286 3185802 : if (colname[0] != '\0' && !bms_is_member(attnum, prevcols))
287 : {
288 : /* Yes, so emit it as next output column */
289 3182020 : *src_colnos = lappend_int(*src_colnos, attnum);
290 3182020 : *res_colnames = lappend(*res_colnames, lfirst(lc));
291 3182020 : *res_colvars = lappend(*res_colvars,
292 3182020 : buildVarFromNSColumn(pstate,
293 3182020 : src_nscolumns + attnum - 1));
294 : /* Copy the input relation's nscolumn data for this column */
295 3182020 : res_nscolumns[colcount] = src_nscolumns[attnum - 1];
296 3182020 : colcount++;
297 : }
298 : }
299 147500 : return colcount;
300 : }
301 :
302 : /* transformJoinUsingClause()
303 : * Build a complete ON clause from a partially-transformed USING list.
304 : * We are given lists of nodes representing left and right match columns.
305 : * Result is a transformed qualification expression.
306 : */
307 : static Node *
308 1458 : transformJoinUsingClause(ParseState *pstate,
309 : List *leftVars, List *rightVars)
310 : {
311 : Node *result;
312 1458 : List *andargs = NIL;
313 : ListCell *lvars,
314 : *rvars;
315 :
316 : /*
317 : * We cheat a little bit here by building an untransformed operator tree
318 : * whose leaves are the already-transformed Vars. This requires collusion
319 : * from transformExpr(), which normally could be expected to complain
320 : * about already-transformed subnodes. However, this does mean that we
321 : * have to mark the columns as requiring SELECT privilege for ourselves;
322 : * transformExpr() won't do it.
323 : */
324 3116 : forboth(lvars, leftVars, rvars, rightVars)
325 : {
326 1658 : Var *lvar = (Var *) lfirst(lvars);
327 1658 : Var *rvar = (Var *) lfirst(rvars);
328 : A_Expr *e;
329 :
330 : /* Require read access to the join variables */
331 1658 : markVarForSelectPriv(pstate, lvar);
332 1658 : markVarForSelectPriv(pstate, rvar);
333 :
334 : /* Now create the lvar = rvar join condition */
335 1658 : e = makeSimpleA_Expr(AEXPR_OP, "=",
336 1658 : (Node *) copyObject(lvar), (Node *) copyObject(rvar),
337 : -1);
338 :
339 : /* Prepare to combine into an AND clause, if multiple join columns */
340 1658 : andargs = lappend(andargs, e);
341 : }
342 :
343 : /* Only need an AND if there's more than one join column */
344 1458 : if (list_length(andargs) == 1)
345 1280 : result = (Node *) linitial(andargs);
346 : else
347 178 : result = (Node *) makeBoolExpr(AND_EXPR, andargs, -1);
348 :
349 : /*
350 : * Since the references are already Vars, and are certainly from the input
351 : * relations, we don't have to go through the same pushups that
352 : * transformJoinOnClause() does. Just invoke transformExpr() to fix up
353 : * the operators, and we're done.
354 : */
355 1458 : result = transformExpr(pstate, result, EXPR_KIND_JOIN_USING);
356 :
357 1458 : result = coerce_to_boolean(pstate, result, "JOIN/USING");
358 :
359 1458 : return result;
360 : }
361 :
362 : /* transformJoinOnClause()
363 : * Transform the qual conditions for JOIN/ON.
364 : * Result is a transformed qualification expression.
365 : */
366 : static Node *
367 72024 : transformJoinOnClause(ParseState *pstate, JoinExpr *j, List *namespace)
368 : {
369 : Node *result;
370 : List *save_namespace;
371 :
372 : /*
373 : * The namespace that the join expression should see is just the two
374 : * subtrees of the JOIN plus any outer references from upper pstate
375 : * levels. Temporarily set this pstate's namespace accordingly. (We need
376 : * not check for refname conflicts, because transformFromClauseItem()
377 : * already did.) All namespace items are marked visible regardless of
378 : * LATERAL state.
379 : */
380 72024 : setNamespaceLateralState(namespace, false, true);
381 :
382 72024 : save_namespace = pstate->p_namespace;
383 72024 : pstate->p_namespace = namespace;
384 :
385 72024 : result = transformWhereClause(pstate, j->quals,
386 : EXPR_KIND_JOIN_ON, "JOIN/ON");
387 :
388 72006 : pstate->p_namespace = save_namespace;
389 :
390 72006 : return result;
391 : }
392 :
393 : /*
394 : * transformTableEntry --- transform a RangeVar (simple relation reference)
395 : */
396 : static ParseNamespaceItem *
397 337988 : transformTableEntry(ParseState *pstate, RangeVar *r)
398 : {
399 : /* addRangeTableEntry does all the work */
400 337988 : return addRangeTableEntry(pstate, r, r->alias, r->inh, true);
401 : }
402 :
403 : /*
404 : * transformRangeSubselect --- transform a sub-SELECT appearing in FROM
405 : */
406 : static ParseNamespaceItem *
407 13832 : transformRangeSubselect(ParseState *pstate, RangeSubselect *r)
408 : {
409 : Query *query;
410 :
411 : /*
412 : * Set p_expr_kind to show this parse level is recursing to a subselect.
413 : * We can't be nested within any expression, so don't need save-restore
414 : * logic here.
415 : */
416 : Assert(pstate->p_expr_kind == EXPR_KIND_NONE);
417 13832 : pstate->p_expr_kind = EXPR_KIND_FROM_SUBSELECT;
418 :
419 : /*
420 : * If the subselect is LATERAL, make lateral_only names of this level
421 : * visible to it. (LATERAL can't nest within a single pstate level, so we
422 : * don't need save/restore logic here.)
423 : */
424 : Assert(!pstate->p_lateral_active);
425 13832 : pstate->p_lateral_active = r->lateral;
426 :
427 : /*
428 : * Analyze and transform the subquery. Note that if the subquery doesn't
429 : * have an alias, it can't be explicitly selected for locking, but locking
430 : * might still be required (if there is an all-tables locking clause).
431 : */
432 13832 : query = parse_sub_analyze(r->subquery, pstate, NULL,
433 13832 : isLockedRefname(pstate,
434 13832 : r->alias == NULL ? NULL :
435 13718 : r->alias->aliasname),
436 : true);
437 :
438 : /* Restore state */
439 13724 : pstate->p_lateral_active = false;
440 13724 : pstate->p_expr_kind = EXPR_KIND_NONE;
441 :
442 : /*
443 : * Check that we got a SELECT. Anything else should be impossible given
444 : * restrictions of the grammar, but check anyway.
445 : */
446 13724 : if (!IsA(query, Query) ||
447 13724 : query->commandType != CMD_SELECT)
448 0 : elog(ERROR, "unexpected non-SELECT command in subquery in FROM");
449 :
450 : /*
451 : * OK, build an RTE and nsitem for the subquery.
452 : */
453 27442 : return addRangeTableEntryForSubquery(pstate,
454 : query,
455 : r->alias,
456 13724 : r->lateral,
457 : true);
458 : }
459 :
460 :
461 : /*
462 : * transformRangeFunction --- transform a function call appearing in FROM
463 : */
464 : static ParseNamespaceItem *
465 39516 : transformRangeFunction(ParseState *pstate, RangeFunction *r)
466 : {
467 39516 : List *funcexprs = NIL;
468 39516 : List *funcnames = NIL;
469 39516 : List *coldeflists = NIL;
470 : bool is_lateral;
471 : ListCell *lc;
472 :
473 : /*
474 : * We make lateral_only names of this level visible, whether or not the
475 : * RangeFunction is explicitly marked LATERAL. This is needed for SQL
476 : * spec compliance in the case of UNNEST(), and seems useful on
477 : * convenience grounds for all functions in FROM.
478 : *
479 : * (LATERAL can't nest within a single pstate level, so we don't need
480 : * save/restore logic here.)
481 : */
482 : Assert(!pstate->p_lateral_active);
483 39516 : pstate->p_lateral_active = true;
484 :
485 : /*
486 : * Transform the raw expressions.
487 : *
488 : * While transforming, also save function names for possible use as alias
489 : * and column names. We use the same transformation rules as for a SELECT
490 : * output expression. For a FuncCall node, the result will be the
491 : * function name, but it is possible for the grammar to hand back other
492 : * node types.
493 : *
494 : * We have to get this info now, because FigureColname only works on raw
495 : * parsetrees. Actually deciding what to do with the names is left up to
496 : * addRangeTableEntryForFunction.
497 : *
498 : * Likewise, collect column definition lists if there were any. But
499 : * complain if we find one here and the RangeFunction has one too.
500 : */
501 79050 : foreach(lc, r->functions)
502 : {
503 39702 : List *pair = (List *) lfirst(lc);
504 : Node *fexpr;
505 : List *coldeflist;
506 : Node *newfexpr;
507 : Node *last_srf;
508 :
509 : /* Disassemble the function-call/column-def-list pairs */
510 : Assert(list_length(pair) == 2);
511 39702 : fexpr = (Node *) linitial(pair);
512 39702 : coldeflist = (List *) lsecond(pair);
513 :
514 : /*
515 : * If we find a function call unnest() with more than one argument and
516 : * no special decoration, transform it into separate unnest() calls on
517 : * each argument. This is a kluge, for sure, but it's less nasty than
518 : * other ways of implementing the SQL-standard UNNEST() syntax.
519 : *
520 : * If there is any decoration (including a coldeflist), we don't
521 : * transform, which probably means a no-such-function error later. We
522 : * could alternatively throw an error right now, but that doesn't seem
523 : * tremendously helpful. If someone is using any such decoration,
524 : * then they're not using the SQL-standard syntax, and they're more
525 : * likely expecting an un-tweaked function call.
526 : *
527 : * Note: the transformation changes a non-schema-qualified unnest()
528 : * function name into schema-qualified pg_catalog.unnest(). This
529 : * choice is also a bit debatable, but it seems reasonable to force
530 : * use of built-in unnest() when we make this transformation.
531 : */
532 39702 : if (IsA(fexpr, FuncCall))
533 : {
534 39558 : FuncCall *fc = (FuncCall *) fexpr;
535 :
536 39558 : if (list_length(fc->funcname) == 1 &&
537 28652 : strcmp(strVal(linitial(fc->funcname)), "unnest") == 0 &&
538 2436 : list_length(fc->args) > 1 &&
539 66 : fc->agg_order == NIL &&
540 66 : fc->agg_filter == NULL &&
541 66 : fc->over == NULL &&
542 66 : !fc->agg_star &&
543 66 : !fc->agg_distinct &&
544 66 : !fc->func_variadic &&
545 : coldeflist == NIL)
546 : {
547 : ListCell *lc2;
548 :
549 240 : foreach(lc2, fc->args)
550 : {
551 174 : Node *arg = (Node *) lfirst(lc2);
552 : FuncCall *newfc;
553 :
554 174 : last_srf = pstate->p_last_srf;
555 :
556 174 : newfc = makeFuncCall(SystemFuncName("unnest"),
557 174 : list_make1(arg),
558 : COERCE_EXPLICIT_CALL,
559 : fc->location);
560 :
561 174 : newfexpr = transformExpr(pstate, (Node *) newfc,
562 : EXPR_KIND_FROM_FUNCTION);
563 :
564 : /* nodeFunctionscan.c requires SRFs to be at top level */
565 174 : if (pstate->p_last_srf != last_srf &&
566 174 : pstate->p_last_srf != newfexpr)
567 0 : ereport(ERROR,
568 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
569 : errmsg("set-returning functions must appear at top level of FROM"),
570 : parser_errposition(pstate,
571 : exprLocation(pstate->p_last_srf))));
572 :
573 174 : funcexprs = lappend(funcexprs, newfexpr);
574 :
575 174 : funcnames = lappend(funcnames,
576 174 : FigureColname((Node *) newfc));
577 :
578 : /* coldeflist is empty, so no error is possible */
579 :
580 174 : coldeflists = lappend(coldeflists, coldeflist);
581 : }
582 66 : continue; /* done with this function item */
583 : }
584 : }
585 :
586 : /* normal case ... */
587 39636 : last_srf = pstate->p_last_srf;
588 :
589 39636 : newfexpr = transformExpr(pstate, fexpr,
590 : EXPR_KIND_FROM_FUNCTION);
591 :
592 : /* nodeFunctionscan.c requires SRFs to be at top level */
593 39474 : if (pstate->p_last_srf != last_srf &&
594 34476 : pstate->p_last_srf != newfexpr)
595 6 : ereport(ERROR,
596 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
597 : errmsg("set-returning functions must appear at top level of FROM"),
598 : parser_errposition(pstate,
599 : exprLocation(pstate->p_last_srf))));
600 :
601 39468 : funcexprs = lappend(funcexprs, newfexpr);
602 :
603 39468 : funcnames = lappend(funcnames,
604 39468 : FigureColname(fexpr));
605 :
606 39468 : if (coldeflist && r->coldeflist)
607 0 : ereport(ERROR,
608 : (errcode(ERRCODE_SYNTAX_ERROR),
609 : errmsg("multiple column definition lists are not allowed for the same function"),
610 : parser_errposition(pstate,
611 : exprLocation((Node *) r->coldeflist))));
612 :
613 39468 : coldeflists = lappend(coldeflists, coldeflist);
614 : }
615 :
616 39348 : pstate->p_lateral_active = false;
617 :
618 : /*
619 : * We must assign collations now so that the RTE exposes correct collation
620 : * info for Vars created from it.
621 : */
622 39348 : assign_list_collations(pstate, funcexprs);
623 :
624 : /*
625 : * Install the top-level coldeflist if there was one (we already checked
626 : * that there was no conflicting per-function coldeflist).
627 : *
628 : * We only allow this when there's a single function (even after UNNEST
629 : * expansion) and no WITH ORDINALITY. The reason for the latter
630 : * restriction is that it's not real clear whether the ordinality column
631 : * should be in the coldeflist, and users are too likely to make mistakes
632 : * in one direction or the other. Putting the coldeflist inside ROWS
633 : * FROM() is much clearer in this case.
634 : */
635 39348 : if (r->coldeflist)
636 : {
637 744 : if (list_length(funcexprs) != 1)
638 : {
639 0 : if (r->is_rowsfrom)
640 0 : ereport(ERROR,
641 : (errcode(ERRCODE_SYNTAX_ERROR),
642 : errmsg("ROWS FROM() with multiple functions cannot have a column definition list"),
643 : errhint("Put a separate column definition list for each function inside ROWS FROM()."),
644 : parser_errposition(pstate,
645 : exprLocation((Node *) r->coldeflist))));
646 : else
647 0 : ereport(ERROR,
648 : (errcode(ERRCODE_SYNTAX_ERROR),
649 : errmsg("UNNEST() with multiple arguments cannot have a column definition list"),
650 : errhint("Use separate UNNEST() calls inside ROWS FROM(), and attach a column definition list to each one."),
651 : parser_errposition(pstate,
652 : exprLocation((Node *) r->coldeflist))));
653 : }
654 744 : if (r->ordinality)
655 0 : ereport(ERROR,
656 : (errcode(ERRCODE_SYNTAX_ERROR),
657 : errmsg("WITH ORDINALITY cannot be used with a column definition list"),
658 : errhint("Put the column definition list inside ROWS FROM()."),
659 : parser_errposition(pstate,
660 : exprLocation((Node *) r->coldeflist))));
661 :
662 744 : coldeflists = list_make1(r->coldeflist);
663 : }
664 :
665 : /*
666 : * Mark the RTE as LATERAL if the user said LATERAL explicitly, or if
667 : * there are any lateral cross-references in it.
668 : */
669 39348 : is_lateral = r->lateral || contain_vars_of_level((Node *) funcexprs, 0);
670 :
671 : /*
672 : * OK, build an RTE and nsitem for the function.
673 : */
674 39348 : return addRangeTableEntryForFunction(pstate,
675 : funcnames, funcexprs, coldeflists,
676 : r, is_lateral, true);
677 : }
678 :
679 : /*
680 : * transformRangeTableFunc -
681 : * Transform a raw RangeTableFunc into TableFunc.
682 : *
683 : * Transform the namespace clauses, the document-generating expression, the
684 : * row-generating expression, the column-generating expressions, and the
685 : * default value expressions.
686 : */
687 : static ParseNamespaceItem *
688 220 : transformRangeTableFunc(ParseState *pstate, RangeTableFunc *rtf)
689 : {
690 220 : TableFunc *tf = makeNode(TableFunc);
691 : const char *constructName;
692 : Oid docType;
693 : bool is_lateral;
694 : ListCell *col;
695 : char **names;
696 : int colno;
697 :
698 : /*
699 : * Currently we only support XMLTABLE here. See transformJsonTable() for
700 : * JSON_TABLE support.
701 : */
702 220 : tf->functype = TFT_XMLTABLE;
703 220 : constructName = "XMLTABLE";
704 220 : docType = XMLOID;
705 :
706 : /*
707 : * We make lateral_only names of this level visible, whether or not the
708 : * RangeTableFunc is explicitly marked LATERAL. This is needed for SQL
709 : * spec compliance and seems useful on convenience grounds for all
710 : * functions in FROM.
711 : *
712 : * (LATERAL can't nest within a single pstate level, so we don't need
713 : * save/restore logic here.)
714 : */
715 : Assert(!pstate->p_lateral_active);
716 220 : pstate->p_lateral_active = true;
717 :
718 : /* Transform and apply typecast to the row-generating expression ... */
719 : Assert(rtf->rowexpr != NULL);
720 220 : tf->rowexpr = coerce_to_specific_type(pstate,
721 : transformExpr(pstate, rtf->rowexpr, EXPR_KIND_FROM_FUNCTION),
722 : TEXTOID,
723 : constructName);
724 220 : assign_expr_collations(pstate, tf->rowexpr);
725 :
726 : /* ... and to the document itself */
727 : Assert(rtf->docexpr != NULL);
728 220 : tf->docexpr = coerce_to_specific_type(pstate,
729 : transformExpr(pstate, rtf->docexpr, EXPR_KIND_FROM_FUNCTION),
730 : docType,
731 : constructName);
732 220 : assign_expr_collations(pstate, tf->docexpr);
733 :
734 : /* undef ordinality column number */
735 220 : tf->ordinalitycol = -1;
736 :
737 : /* Process column specs */
738 220 : names = palloc(sizeof(char *) * list_length(rtf->columns));
739 :
740 220 : colno = 0;
741 970 : foreach(col, rtf->columns)
742 : {
743 750 : RangeTableFuncCol *rawc = (RangeTableFuncCol *) lfirst(col);
744 : Oid typid;
745 : int32 typmod;
746 : Node *colexpr;
747 : Node *coldefexpr;
748 : int j;
749 :
750 750 : tf->colnames = lappend(tf->colnames,
751 750 : makeString(pstrdup(rawc->colname)));
752 :
753 : /*
754 : * Determine the type and typmod for the new column. FOR ORDINALITY
755 : * columns are INTEGER per spec; the others are user-specified.
756 : */
757 750 : if (rawc->for_ordinality)
758 : {
759 62 : if (tf->ordinalitycol != -1)
760 0 : ereport(ERROR,
761 : (errcode(ERRCODE_SYNTAX_ERROR),
762 : errmsg("only one FOR ORDINALITY column is allowed"),
763 : parser_errposition(pstate, rawc->location)));
764 :
765 62 : typid = INT4OID;
766 62 : typmod = -1;
767 62 : tf->ordinalitycol = colno;
768 : }
769 : else
770 : {
771 688 : if (rawc->typeName->setof)
772 0 : ereport(ERROR,
773 : (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
774 : errmsg("column \"%s\" cannot be declared SETOF",
775 : rawc->colname),
776 : parser_errposition(pstate, rawc->location)));
777 :
778 688 : typenameTypeIdAndMod(pstate, rawc->typeName,
779 : &typid, &typmod);
780 : }
781 :
782 750 : tf->coltypes = lappend_oid(tf->coltypes, typid);
783 750 : tf->coltypmods = lappend_int(tf->coltypmods, typmod);
784 750 : tf->colcollations = lappend_oid(tf->colcollations,
785 : get_typcollation(typid));
786 :
787 : /* Transform the PATH and DEFAULT expressions */
788 750 : if (rawc->colexpr)
789 : {
790 490 : colexpr = coerce_to_specific_type(pstate,
791 : transformExpr(pstate, rawc->colexpr,
792 : EXPR_KIND_FROM_FUNCTION),
793 : TEXTOID,
794 : constructName);
795 490 : assign_expr_collations(pstate, colexpr);
796 : }
797 : else
798 260 : colexpr = NULL;
799 :
800 750 : if (rawc->coldefexpr)
801 : {
802 56 : coldefexpr = coerce_to_specific_type_typmod(pstate,
803 : transformExpr(pstate, rawc->coldefexpr,
804 : EXPR_KIND_FROM_FUNCTION),
805 : typid, typmod,
806 : constructName);
807 56 : assign_expr_collations(pstate, coldefexpr);
808 : }
809 : else
810 694 : coldefexpr = NULL;
811 :
812 750 : tf->colexprs = lappend(tf->colexprs, colexpr);
813 750 : tf->coldefexprs = lappend(tf->coldefexprs, coldefexpr);
814 :
815 750 : if (rawc->is_not_null)
816 56 : tf->notnulls = bms_add_member(tf->notnulls, colno);
817 :
818 : /* make sure column names are unique */
819 2534 : for (j = 0; j < colno; j++)
820 1784 : if (strcmp(names[j], rawc->colname) == 0)
821 0 : ereport(ERROR,
822 : (errcode(ERRCODE_SYNTAX_ERROR),
823 : errmsg("column name \"%s\" is not unique",
824 : rawc->colname),
825 : parser_errposition(pstate, rawc->location)));
826 750 : names[colno] = rawc->colname;
827 :
828 750 : colno++;
829 : }
830 220 : pfree(names);
831 :
832 : /* Namespaces, if any, also need to be transformed */
833 220 : if (rtf->namespaces != NIL)
834 : {
835 : ListCell *ns;
836 : ListCell *lc2;
837 20 : List *ns_uris = NIL;
838 20 : List *ns_names = NIL;
839 20 : bool default_ns_seen = false;
840 :
841 40 : foreach(ns, rtf->namespaces)
842 : {
843 20 : ResTarget *r = (ResTarget *) lfirst(ns);
844 : Node *ns_uri;
845 :
846 : Assert(IsA(r, ResTarget));
847 20 : ns_uri = transformExpr(pstate, r->val, EXPR_KIND_FROM_FUNCTION);
848 20 : ns_uri = coerce_to_specific_type(pstate, ns_uri,
849 : TEXTOID, constructName);
850 20 : assign_expr_collations(pstate, ns_uri);
851 20 : ns_uris = lappend(ns_uris, ns_uri);
852 :
853 : /* Verify consistency of name list: no dupes, only one DEFAULT */
854 20 : if (r->name != NULL)
855 : {
856 14 : foreach(lc2, ns_names)
857 : {
858 0 : String *ns_node = lfirst_node(String, lc2);
859 :
860 0 : if (ns_node == NULL)
861 0 : continue;
862 0 : if (strcmp(strVal(ns_node), r->name) == 0)
863 0 : ereport(ERROR,
864 : (errcode(ERRCODE_SYNTAX_ERROR),
865 : errmsg("namespace name \"%s\" is not unique",
866 : r->name),
867 : parser_errposition(pstate, r->location)));
868 : }
869 : }
870 : else
871 : {
872 6 : if (default_ns_seen)
873 0 : ereport(ERROR,
874 : (errcode(ERRCODE_SYNTAX_ERROR),
875 : errmsg("only one default namespace is allowed"),
876 : parser_errposition(pstate, r->location)));
877 6 : default_ns_seen = true;
878 : }
879 :
880 : /* We represent DEFAULT by a null pointer */
881 20 : ns_names = lappend(ns_names,
882 20 : r->name ? makeString(r->name) : NULL);
883 : }
884 :
885 20 : tf->ns_uris = ns_uris;
886 20 : tf->ns_names = ns_names;
887 : }
888 :
889 220 : tf->location = rtf->location;
890 :
891 220 : pstate->p_lateral_active = false;
892 :
893 : /*
894 : * Mark the RTE as LATERAL if the user said LATERAL explicitly, or if
895 : * there are any lateral cross-references in it.
896 : */
897 220 : is_lateral = rtf->lateral || contain_vars_of_level((Node *) tf, 0);
898 :
899 220 : return addRangeTableEntryForTableFunc(pstate,
900 : tf, rtf->alias, is_lateral, true);
901 : }
902 :
903 : /*
904 : * transformRangeTableSample --- transform a TABLESAMPLE clause
905 : *
906 : * Caller has already transformed rts->relation, we just have to validate
907 : * the remaining fields and create a TableSampleClause node.
908 : */
909 : static TableSampleClause *
910 242 : transformRangeTableSample(ParseState *pstate, RangeTableSample *rts)
911 : {
912 : TableSampleClause *tablesample;
913 : Oid handlerOid;
914 : Oid funcargtypes[1];
915 : TsmRoutine *tsm;
916 : List *fargs;
917 : ListCell *larg,
918 : *ltyp;
919 :
920 : /*
921 : * To validate the sample method name, look up the handler function, which
922 : * has the same name, one dummy INTERNAL argument, and a result type of
923 : * tsm_handler. (Note: tablesample method names are not schema-qualified
924 : * in the SQL standard; but since they are just functions to us, we allow
925 : * schema qualification to resolve any potential ambiguity.)
926 : */
927 242 : funcargtypes[0] = INTERNALOID;
928 :
929 242 : handlerOid = LookupFuncName(rts->method, 1, funcargtypes, true);
930 :
931 : /* we want error to complain about no-such-method, not no-such-function */
932 242 : if (!OidIsValid(handlerOid))
933 6 : ereport(ERROR,
934 : (errcode(ERRCODE_UNDEFINED_OBJECT),
935 : errmsg("tablesample method %s does not exist",
936 : NameListToString(rts->method)),
937 : parser_errposition(pstate, rts->location)));
938 :
939 : /* check that handler has correct return type */
940 236 : if (get_func_rettype(handlerOid) != TSM_HANDLEROID)
941 0 : ereport(ERROR,
942 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
943 : errmsg("function %s must return type %s",
944 : NameListToString(rts->method), "tsm_handler"),
945 : parser_errposition(pstate, rts->location)));
946 :
947 : /* OK, run the handler to get TsmRoutine, for argument type info */
948 236 : tsm = GetTsmRoutine(handlerOid);
949 :
950 236 : tablesample = makeNode(TableSampleClause);
951 236 : tablesample->tsmhandler = handlerOid;
952 :
953 : /* check user provided the expected number of arguments */
954 236 : if (list_length(rts->args) != list_length(tsm->parameterTypes))
955 0 : ereport(ERROR,
956 : (errcode(ERRCODE_INVALID_TABLESAMPLE_ARGUMENT),
957 : errmsg_plural("tablesample method %s requires %d argument, not %d",
958 : "tablesample method %s requires %d arguments, not %d",
959 : list_length(tsm->parameterTypes),
960 : NameListToString(rts->method),
961 : list_length(tsm->parameterTypes),
962 : list_length(rts->args)),
963 : parser_errposition(pstate, rts->location)));
964 :
965 : /*
966 : * Transform the arguments, typecasting them as needed. Note we must also
967 : * assign collations now, because assign_query_collations() doesn't
968 : * examine any substructure of RTEs.
969 : */
970 236 : fargs = NIL;
971 472 : forboth(larg, rts->args, ltyp, tsm->parameterTypes)
972 : {
973 236 : Node *arg = (Node *) lfirst(larg);
974 236 : Oid argtype = lfirst_oid(ltyp);
975 :
976 236 : arg = transformExpr(pstate, arg, EXPR_KIND_FROM_FUNCTION);
977 236 : arg = coerce_to_specific_type(pstate, arg, argtype, "TABLESAMPLE");
978 236 : assign_expr_collations(pstate, arg);
979 236 : fargs = lappend(fargs, arg);
980 : }
981 236 : tablesample->args = fargs;
982 :
983 : /* Process REPEATABLE (seed) */
984 236 : if (rts->repeatable != NULL)
985 : {
986 : Node *arg;
987 :
988 102 : if (!tsm->repeatable_across_queries)
989 4 : ereport(ERROR,
990 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
991 : errmsg("tablesample method %s does not support REPEATABLE",
992 : NameListToString(rts->method)),
993 : parser_errposition(pstate, rts->location)));
994 :
995 98 : arg = transformExpr(pstate, rts->repeatable, EXPR_KIND_FROM_FUNCTION);
996 98 : arg = coerce_to_specific_type(pstate, arg, FLOAT8OID, "REPEATABLE");
997 98 : assign_expr_collations(pstate, arg);
998 98 : tablesample->repeatable = (Expr *) arg;
999 : }
1000 : else
1001 134 : tablesample->repeatable = NULL;
1002 :
1003 232 : return tablesample;
1004 : }
1005 :
1006 : /*
1007 : * getNSItemForSpecialRelationTypes
1008 : *
1009 : * If given RangeVar refers to a CTE or an EphemeralNamedRelation,
1010 : * build and return an appropriate ParseNamespaceItem, otherwise return NULL
1011 : */
1012 : static ParseNamespaceItem *
1013 344112 : getNSItemForSpecialRelationTypes(ParseState *pstate, RangeVar *rv)
1014 : {
1015 : ParseNamespaceItem *nsitem;
1016 : CommonTableExpr *cte;
1017 : Index levelsup;
1018 :
1019 : /*
1020 : * if it is a qualified name, it can't be a CTE or tuplestore reference
1021 : */
1022 344112 : if (rv->schemaname)
1023 171642 : return NULL;
1024 :
1025 172470 : cte = scanNameSpaceForCTE(pstate, rv->relname, &levelsup);
1026 172470 : if (cte)
1027 5680 : nsitem = addRangeTableEntryForCTE(pstate, cte, levelsup, rv, true);
1028 166790 : else if (scanNameSpaceForENR(pstate, rv->relname))
1029 444 : nsitem = addRangeTableEntryForENR(pstate, rv, true);
1030 : else
1031 166346 : nsitem = NULL;
1032 :
1033 172458 : return nsitem;
1034 : }
1035 :
1036 : /*
1037 : * transformFromClauseItem -
1038 : * Transform a FROM-clause item, adding any required entries to the
1039 : * range table list being built in the ParseState, and return the
1040 : * transformed item ready to include in the joinlist. Also build a
1041 : * ParseNamespaceItem list describing the names exposed by this item.
1042 : * This routine can recurse to handle SQL92 JOIN expressions.
1043 : *
1044 : * The function return value is the node to add to the jointree (a
1045 : * RangeTblRef or JoinExpr). Additional output parameters are:
1046 : *
1047 : * *top_nsitem: receives the ParseNamespaceItem directly corresponding to the
1048 : * jointree item. (This is only used during internal recursion, not by
1049 : * outside callers.)
1050 : *
1051 : * *namespace: receives a List of ParseNamespaceItems for the RTEs exposed
1052 : * as table/column names by this item. (The lateral_only flags in these items
1053 : * are indeterminate and should be explicitly set by the caller before use.)
1054 : */
1055 : static Node *
1056 472166 : transformFromClauseItem(ParseState *pstate, Node *n,
1057 : ParseNamespaceItem **top_nsitem,
1058 : List **namespace)
1059 : {
1060 : /* Guard against stack overflow due to overly deep subtree */
1061 472166 : check_stack_depth();
1062 :
1063 472166 : if (IsA(n, RangeVar))
1064 : {
1065 : /* Plain relation reference, or perhaps a CTE reference */
1066 344112 : RangeVar *rv = (RangeVar *) n;
1067 : RangeTblRef *rtr;
1068 : ParseNamespaceItem *nsitem;
1069 :
1070 : /* Check if it's a CTE or tuplestore reference */
1071 344112 : nsitem = getNSItemForSpecialRelationTypes(pstate, rv);
1072 :
1073 : /* if not found above, must be a table reference */
1074 344100 : if (!nsitem)
1075 337988 : nsitem = transformTableEntry(pstate, rv);
1076 :
1077 343914 : *top_nsitem = nsitem;
1078 343914 : *namespace = list_make1(nsitem);
1079 343914 : rtr = makeNode(RangeTblRef);
1080 343914 : rtr->rtindex = nsitem->p_rtindex;
1081 343914 : return (Node *) rtr;
1082 : }
1083 128054 : else if (IsA(n, RangeSubselect))
1084 : {
1085 : /* sub-SELECT is like a plain relation */
1086 : RangeTblRef *rtr;
1087 : ParseNamespaceItem *nsitem;
1088 :
1089 13832 : nsitem = transformRangeSubselect(pstate, (RangeSubselect *) n);
1090 13718 : *top_nsitem = nsitem;
1091 13718 : *namespace = list_make1(nsitem);
1092 13718 : rtr = makeNode(RangeTblRef);
1093 13718 : rtr->rtindex = nsitem->p_rtindex;
1094 13718 : return (Node *) rtr;
1095 : }
1096 114222 : else if (IsA(n, RangeFunction))
1097 : {
1098 : /* function is like a plain relation */
1099 : RangeTblRef *rtr;
1100 : ParseNamespaceItem *nsitem;
1101 :
1102 39516 : nsitem = transformRangeFunction(pstate, (RangeFunction *) n);
1103 39294 : *top_nsitem = nsitem;
1104 39294 : *namespace = list_make1(nsitem);
1105 39294 : rtr = makeNode(RangeTblRef);
1106 39294 : rtr->rtindex = nsitem->p_rtindex;
1107 39294 : return (Node *) rtr;
1108 : }
1109 74706 : else if (IsA(n, RangeTableFunc) || IsA(n, JsonTable))
1110 : {
1111 : /* table function is like a plain relation */
1112 : RangeTblRef *rtr;
1113 : ParseNamespaceItem *nsitem;
1114 :
1115 648 : if (IsA(n, JsonTable))
1116 428 : nsitem = transformJsonTable(pstate, (JsonTable *) n);
1117 : else
1118 220 : nsitem = transformRangeTableFunc(pstate, (RangeTableFunc *) n);
1119 :
1120 552 : *top_nsitem = nsitem;
1121 552 : *namespace = list_make1(nsitem);
1122 552 : rtr = makeNode(RangeTblRef);
1123 552 : rtr->rtindex = nsitem->p_rtindex;
1124 552 : return (Node *) rtr;
1125 : }
1126 74058 : else if (IsA(n, RangeTableSample))
1127 : {
1128 : /* TABLESAMPLE clause (wrapping some other valid FROM node) */
1129 254 : RangeTableSample *rts = (RangeTableSample *) n;
1130 : Node *rel;
1131 : RangeTblEntry *rte;
1132 :
1133 : /* Recursively transform the contained relation */
1134 254 : rel = transformFromClauseItem(pstate, rts->relation,
1135 : top_nsitem, namespace);
1136 254 : rte = (*top_nsitem)->p_rte;
1137 : /* We only support this on plain relations and matviews */
1138 254 : if (rte->rtekind != RTE_RELATION ||
1139 248 : (rte->relkind != RELKIND_RELATION &&
1140 24 : rte->relkind != RELKIND_MATVIEW &&
1141 24 : rte->relkind != RELKIND_PARTITIONED_TABLE))
1142 12 : ereport(ERROR,
1143 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1144 : errmsg("TABLESAMPLE clause can only be applied to tables and materialized views"),
1145 : parser_errposition(pstate, exprLocation(rts->relation))));
1146 :
1147 : /* Transform TABLESAMPLE details and attach to the RTE */
1148 242 : rte->tablesample = transformRangeTableSample(pstate, rts);
1149 232 : return rel;
1150 : }
1151 73804 : else if (IsA(n, JoinExpr))
1152 : {
1153 : /* A newfangled join expression */
1154 73804 : JoinExpr *j = (JoinExpr *) n;
1155 : ParseNamespaceItem *nsitem;
1156 : ParseNamespaceItem *l_nsitem;
1157 : ParseNamespaceItem *r_nsitem;
1158 : List *l_namespace,
1159 : *r_namespace,
1160 : *my_namespace,
1161 : *l_colnames,
1162 : *r_colnames,
1163 : *res_colnames,
1164 : *l_colnos,
1165 : *r_colnos,
1166 : *res_colvars;
1167 : ParseNamespaceColumn *l_nscolumns,
1168 : *r_nscolumns,
1169 : *res_nscolumns;
1170 : int res_colindex;
1171 : bool lateral_ok;
1172 : int sv_namespace_length;
1173 : int k;
1174 :
1175 : /*
1176 : * Recursively process the left subtree, then the right. We must do
1177 : * it in this order for correct visibility of LATERAL references.
1178 : */
1179 73804 : j->larg = transformFromClauseItem(pstate, j->larg,
1180 : &l_nsitem,
1181 : &l_namespace);
1182 :
1183 : /*
1184 : * Make the left-side RTEs available for LATERAL access within the
1185 : * right side, by temporarily adding them to the pstate's namespace
1186 : * list. Per SQL:2008, if the join type is not INNER or LEFT then the
1187 : * left-side names must still be exposed, but it's an error to
1188 : * reference them. (Stupid design, but that's what it says.) Hence,
1189 : * we always push them into the namespace, but mark them as not
1190 : * lateral_ok if the jointype is wrong.
1191 : *
1192 : * Notice that we don't require the merged namespace list to be
1193 : * conflict-free. See the comments for scanNameSpaceForRefname().
1194 : */
1195 73804 : lateral_ok = (j->jointype == JOIN_INNER || j->jointype == JOIN_LEFT);
1196 73804 : setNamespaceLateralState(l_namespace, true, lateral_ok);
1197 :
1198 73804 : sv_namespace_length = list_length(pstate->p_namespace);
1199 73804 : pstate->p_namespace = list_concat(pstate->p_namespace, l_namespace);
1200 :
1201 : /* And now we can process the RHS */
1202 73804 : j->rarg = transformFromClauseItem(pstate, j->rarg,
1203 : &r_nsitem,
1204 : &r_namespace);
1205 :
1206 : /* Remove the left-side RTEs from the namespace list again */
1207 73768 : pstate->p_namespace = list_truncate(pstate->p_namespace,
1208 : sv_namespace_length);
1209 :
1210 : /*
1211 : * Check for conflicting refnames in left and right subtrees. Must do
1212 : * this because higher levels will assume I hand back a self-
1213 : * consistent namespace list.
1214 : */
1215 73768 : checkNameSpaceConflicts(pstate, l_namespace, r_namespace);
1216 :
1217 : /*
1218 : * Generate combined namespace info for possible use below.
1219 : */
1220 73768 : my_namespace = list_concat(l_namespace, r_namespace);
1221 :
1222 : /*
1223 : * We'll work from the nscolumns data and eref alias column names for
1224 : * each of the input nsitems. Note that these include dropped
1225 : * columns, which is helpful because we can keep track of physical
1226 : * input column numbers more easily.
1227 : */
1228 73768 : l_nscolumns = l_nsitem->p_nscolumns;
1229 73768 : l_colnames = l_nsitem->p_names->colnames;
1230 73768 : r_nscolumns = r_nsitem->p_nscolumns;
1231 73768 : r_colnames = r_nsitem->p_names->colnames;
1232 :
1233 : /*
1234 : * Natural join does not explicitly specify columns; must generate
1235 : * columns to join. Need to run through the list of columns from each
1236 : * table or join result and match up the column names. Use the first
1237 : * table, and check every column in the second table for a match.
1238 : * (We'll check that the matches were unique later on.) The result of
1239 : * this step is a list of column names just like an explicitly-written
1240 : * USING list.
1241 : */
1242 73768 : if (j->isNatural)
1243 : {
1244 258 : List *rlist = NIL;
1245 : ListCell *lx,
1246 : *rx;
1247 :
1248 : Assert(j->usingClause == NIL); /* shouldn't have USING() too */
1249 :
1250 1140 : foreach(lx, l_colnames)
1251 : {
1252 882 : char *l_colname = strVal(lfirst(lx));
1253 882 : String *m_name = NULL;
1254 :
1255 882 : if (l_colname[0] == '\0')
1256 12 : continue; /* ignore dropped columns */
1257 :
1258 2412 : foreach(rx, r_colnames)
1259 : {
1260 1848 : char *r_colname = strVal(lfirst(rx));
1261 :
1262 1848 : if (strcmp(l_colname, r_colname) == 0)
1263 : {
1264 306 : m_name = makeString(l_colname);
1265 306 : break;
1266 : }
1267 : }
1268 :
1269 : /* matched a right column? then keep as join column... */
1270 870 : if (m_name != NULL)
1271 306 : rlist = lappend(rlist, m_name);
1272 : }
1273 :
1274 258 : j->usingClause = rlist;
1275 : }
1276 :
1277 : /*
1278 : * If a USING clause alias was specified, save the USING columns as
1279 : * its column list.
1280 : */
1281 73768 : if (j->join_using_alias)
1282 84 : j->join_using_alias->colnames = j->usingClause;
1283 :
1284 : /*
1285 : * Now transform the join qualifications, if any.
1286 : */
1287 73768 : l_colnos = NIL;
1288 73768 : r_colnos = NIL;
1289 73768 : res_colnames = NIL;
1290 73768 : res_colvars = NIL;
1291 :
1292 : /* this may be larger than needed, but it's not worth being exact */
1293 : res_nscolumns = (ParseNamespaceColumn *)
1294 73768 : palloc0((list_length(l_colnames) + list_length(r_colnames)) *
1295 : sizeof(ParseNamespaceColumn));
1296 73768 : res_colindex = 0;
1297 :
1298 73768 : if (j->usingClause)
1299 : {
1300 : /*
1301 : * JOIN/USING (or NATURAL JOIN, as transformed above). Transform
1302 : * the list into an explicit ON-condition.
1303 : */
1304 1458 : List *ucols = j->usingClause;
1305 1458 : List *l_usingvars = NIL;
1306 1458 : List *r_usingvars = NIL;
1307 : ListCell *ucol;
1308 :
1309 : Assert(j->quals == NULL); /* shouldn't have ON() too */
1310 :
1311 3116 : foreach(ucol, ucols)
1312 : {
1313 1658 : char *u_colname = strVal(lfirst(ucol));
1314 : ListCell *col;
1315 : int ndx;
1316 1658 : int l_index = -1;
1317 1658 : int r_index = -1;
1318 : Var *l_colvar,
1319 : *r_colvar;
1320 :
1321 : Assert(u_colname[0] != '\0');
1322 :
1323 : /* Check for USING(foo,foo) */
1324 1886 : foreach(col, res_colnames)
1325 : {
1326 228 : char *res_colname = strVal(lfirst(col));
1327 :
1328 228 : if (strcmp(res_colname, u_colname) == 0)
1329 0 : ereport(ERROR,
1330 : (errcode(ERRCODE_DUPLICATE_COLUMN),
1331 : errmsg("column name \"%s\" appears more than once in USING clause",
1332 : u_colname)));
1333 : }
1334 :
1335 : /* Find it in left input */
1336 1658 : ndx = 0;
1337 8130 : foreach(col, l_colnames)
1338 : {
1339 6472 : char *l_colname = strVal(lfirst(col));
1340 :
1341 6472 : if (strcmp(l_colname, u_colname) == 0)
1342 : {
1343 1658 : if (l_index >= 0)
1344 0 : ereport(ERROR,
1345 : (errcode(ERRCODE_AMBIGUOUS_COLUMN),
1346 : errmsg("common column name \"%s\" appears more than once in left table",
1347 : u_colname)));
1348 1658 : l_index = ndx;
1349 : }
1350 6472 : ndx++;
1351 : }
1352 1658 : if (l_index < 0)
1353 0 : ereport(ERROR,
1354 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1355 : errmsg("column \"%s\" specified in USING clause does not exist in left table",
1356 : u_colname)));
1357 1658 : l_colnos = lappend_int(l_colnos, l_index + 1);
1358 :
1359 : /* Find it in right input */
1360 1658 : ndx = 0;
1361 8030 : foreach(col, r_colnames)
1362 : {
1363 6372 : char *r_colname = strVal(lfirst(col));
1364 :
1365 6372 : if (strcmp(r_colname, u_colname) == 0)
1366 : {
1367 1658 : if (r_index >= 0)
1368 0 : ereport(ERROR,
1369 : (errcode(ERRCODE_AMBIGUOUS_COLUMN),
1370 : errmsg("common column name \"%s\" appears more than once in right table",
1371 : u_colname)));
1372 1658 : r_index = ndx;
1373 : }
1374 6372 : ndx++;
1375 : }
1376 1658 : if (r_index < 0)
1377 0 : ereport(ERROR,
1378 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1379 : errmsg("column \"%s\" specified in USING clause does not exist in right table",
1380 : u_colname)));
1381 1658 : r_colnos = lappend_int(r_colnos, r_index + 1);
1382 :
1383 : /* Build Vars to use in the generated JOIN ON clause */
1384 1658 : l_colvar = buildVarFromNSColumn(pstate, l_nscolumns + l_index);
1385 1658 : l_usingvars = lappend(l_usingvars, l_colvar);
1386 1658 : r_colvar = buildVarFromNSColumn(pstate, r_nscolumns + r_index);
1387 1658 : r_usingvars = lappend(r_usingvars, r_colvar);
1388 :
1389 : /*
1390 : * While we're here, add column names to the res_colnames
1391 : * list. It's a bit ugly to do this here while the
1392 : * corresponding res_colvars entries are not made till later,
1393 : * but doing this later would require an additional traversal
1394 : * of the usingClause list.
1395 : */
1396 1658 : res_colnames = lappend(res_colnames, lfirst(ucol));
1397 : }
1398 :
1399 : /* Construct the generated JOIN ON clause */
1400 1458 : j->quals = transformJoinUsingClause(pstate,
1401 : l_usingvars,
1402 : r_usingvars);
1403 : }
1404 72310 : else if (j->quals)
1405 : {
1406 : /* User-written ON-condition; transform it */
1407 72024 : j->quals = transformJoinOnClause(pstate, j, my_namespace);
1408 : }
1409 : else
1410 : {
1411 : /* CROSS JOIN: no quals */
1412 : }
1413 :
1414 : /*
1415 : * If this is an outer join, now mark the appropriate child RTEs as
1416 : * being nulled by this join. We have finished processing the child
1417 : * join expressions as well as the current join's quals, which deal in
1418 : * non-nulled input columns. All future references to those RTEs will
1419 : * see possibly-nulled values, and we should mark generated Vars to
1420 : * account for that. In particular, the join alias Vars that we're
1421 : * about to build should reflect the nulling effects of this join.
1422 : *
1423 : * A difficulty with doing this is that we need the join's RT index,
1424 : * which we don't officially have yet. However, no other RTE can get
1425 : * made between here and the addRangeTableEntryForJoin call, so we can
1426 : * predict what the assignment will be. (Alternatively, we could call
1427 : * addRangeTableEntryForJoin before we have all the data computed, but
1428 : * this seems less ugly.)
1429 : */
1430 73750 : j->rtindex = list_length(pstate->p_rtable) + 1;
1431 :
1432 73750 : switch (j->jointype)
1433 : {
1434 32974 : case JOIN_INNER:
1435 32974 : break;
1436 39462 : case JOIN_LEFT:
1437 39462 : markRelsAsNulledBy(pstate, j->rarg, j->rtindex);
1438 39462 : break;
1439 994 : case JOIN_FULL:
1440 994 : markRelsAsNulledBy(pstate, j->larg, j->rtindex);
1441 994 : markRelsAsNulledBy(pstate, j->rarg, j->rtindex);
1442 994 : break;
1443 320 : case JOIN_RIGHT:
1444 320 : markRelsAsNulledBy(pstate, j->larg, j->rtindex);
1445 320 : break;
1446 0 : default:
1447 : /* shouldn't see any other types here */
1448 0 : elog(ERROR, "unrecognized join type: %d",
1449 : (int) j->jointype);
1450 : break;
1451 : }
1452 :
1453 : /*
1454 : * Now we can construct join alias expressions for the USING columns.
1455 : */
1456 73750 : if (j->usingClause)
1457 : {
1458 : ListCell *lc1,
1459 : *lc2;
1460 :
1461 : /* Scan the colnos lists to recover info from the previous loop */
1462 3116 : forboth(lc1, l_colnos, lc2, r_colnos)
1463 : {
1464 1658 : int l_index = lfirst_int(lc1) - 1;
1465 1658 : int r_index = lfirst_int(lc2) - 1;
1466 : Var *l_colvar,
1467 : *r_colvar;
1468 : Node *u_colvar;
1469 : ParseNamespaceColumn *res_nscolumn;
1470 :
1471 : /*
1472 : * Note we re-build these Vars: they might have different
1473 : * varnullingrels than the ones made in the previous loop.
1474 : */
1475 1658 : l_colvar = buildVarFromNSColumn(pstate, l_nscolumns + l_index);
1476 1658 : r_colvar = buildVarFromNSColumn(pstate, r_nscolumns + r_index);
1477 :
1478 : /* Construct the join alias Var for this column */
1479 1658 : u_colvar = buildMergedJoinVar(pstate,
1480 : j->jointype,
1481 : l_colvar,
1482 : r_colvar);
1483 1658 : res_colvars = lappend(res_colvars, u_colvar);
1484 :
1485 : /* Construct column's res_nscolumns[] entry */
1486 1658 : res_nscolumn = res_nscolumns + res_colindex;
1487 1658 : res_colindex++;
1488 1658 : if (u_colvar == (Node *) l_colvar)
1489 : {
1490 : /* Merged column is equivalent to left input */
1491 1214 : *res_nscolumn = l_nscolumns[l_index];
1492 : }
1493 444 : else if (u_colvar == (Node *) r_colvar)
1494 : {
1495 : /* Merged column is equivalent to right input */
1496 42 : *res_nscolumn = r_nscolumns[r_index];
1497 : }
1498 : else
1499 : {
1500 : /*
1501 : * Merged column is not semantically equivalent to either
1502 : * input, so it needs to be referenced as the join output
1503 : * column.
1504 : */
1505 402 : res_nscolumn->p_varno = j->rtindex;
1506 402 : res_nscolumn->p_varattno = res_colindex;
1507 402 : res_nscolumn->p_vartype = exprType(u_colvar);
1508 402 : res_nscolumn->p_vartypmod = exprTypmod(u_colvar);
1509 402 : res_nscolumn->p_varcollid = exprCollation(u_colvar);
1510 402 : res_nscolumn->p_varnosyn = j->rtindex;
1511 402 : res_nscolumn->p_varattnosyn = res_colindex;
1512 : }
1513 : }
1514 : }
1515 :
1516 : /* Add remaining columns from each side to the output columns */
1517 73750 : res_colindex +=
1518 73750 : extractRemainingColumns(pstate,
1519 : l_nscolumns, l_colnames, &l_colnos,
1520 : &res_colnames, &res_colvars,
1521 73750 : res_nscolumns + res_colindex);
1522 73750 : res_colindex +=
1523 73750 : extractRemainingColumns(pstate,
1524 : r_nscolumns, r_colnames, &r_colnos,
1525 : &res_colnames, &res_colvars,
1526 73750 : res_nscolumns + res_colindex);
1527 :
1528 : /* If join has an alias, it syntactically hides all inputs */
1529 73750 : if (j->alias)
1530 : {
1531 978 : for (k = 0; k < res_colindex; k++)
1532 : {
1533 804 : ParseNamespaceColumn *nscol = res_nscolumns + k;
1534 :
1535 804 : nscol->p_varnosyn = j->rtindex;
1536 804 : nscol->p_varattnosyn = k + 1;
1537 : }
1538 : }
1539 :
1540 : /*
1541 : * Now build an RTE and nsitem for the result of the join.
1542 : */
1543 73750 : nsitem = addRangeTableEntryForJoin(pstate,
1544 : res_colnames,
1545 : res_nscolumns,
1546 : j->jointype,
1547 73750 : list_length(j->usingClause),
1548 : res_colvars,
1549 : l_colnos,
1550 : r_colnos,
1551 : j->join_using_alias,
1552 : j->alias,
1553 : true);
1554 :
1555 : /* Verify that we correctly predicted the join's RT index */
1556 : Assert(j->rtindex == nsitem->p_rtindex);
1557 : /* Cross-check number of columns, too */
1558 : Assert(res_colindex == list_length(nsitem->p_names->colnames));
1559 :
1560 : /*
1561 : * Save a link to the JoinExpr in the proper element of p_joinexprs.
1562 : * Since we maintain that list lazily, it may be necessary to fill in
1563 : * empty entries before we can add the JoinExpr in the right place.
1564 : */
1565 192046 : for (k = list_length(pstate->p_joinexprs) + 1; k < j->rtindex; k++)
1566 118302 : pstate->p_joinexprs = lappend(pstate->p_joinexprs, NULL);
1567 73744 : pstate->p_joinexprs = lappend(pstate->p_joinexprs, j);
1568 : Assert(list_length(pstate->p_joinexprs) == j->rtindex);
1569 :
1570 : /*
1571 : * If the join has a USING alias, build a ParseNamespaceItem for that
1572 : * and add it to the list of nsitems in the join's input.
1573 : */
1574 73744 : if (j->join_using_alias)
1575 : {
1576 : ParseNamespaceItem *jnsitem;
1577 :
1578 84 : jnsitem = (ParseNamespaceItem *) palloc(sizeof(ParseNamespaceItem));
1579 84 : jnsitem->p_names = j->join_using_alias;
1580 84 : jnsitem->p_rte = nsitem->p_rte;
1581 84 : jnsitem->p_rtindex = nsitem->p_rtindex;
1582 84 : jnsitem->p_perminfo = NULL;
1583 : /* no need to copy the first N columns, just use res_nscolumns */
1584 84 : jnsitem->p_nscolumns = res_nscolumns;
1585 : /* set default visibility flags; might get changed later */
1586 84 : jnsitem->p_rel_visible = true;
1587 84 : jnsitem->p_cols_visible = true;
1588 84 : jnsitem->p_lateral_only = false;
1589 84 : jnsitem->p_lateral_ok = true;
1590 : /* Per SQL, we must check for alias conflicts */
1591 84 : checkNameSpaceConflicts(pstate, list_make1(jnsitem), my_namespace);
1592 78 : my_namespace = lappend(my_namespace, jnsitem);
1593 : }
1594 :
1595 : /*
1596 : * Prepare returned namespace list. If the JOIN has an alias then it
1597 : * hides the contained RTEs completely; otherwise, the contained RTEs
1598 : * are still visible as table names, but are not visible for
1599 : * unqualified column-name access.
1600 : *
1601 : * Note: if there are nested alias-less JOINs, the lower-level ones
1602 : * will remain in the list although they have neither p_rel_visible
1603 : * nor p_cols_visible set. We could delete such list items, but it's
1604 : * unclear that it's worth expending cycles to do so.
1605 : */
1606 73738 : if (j->alias != NULL)
1607 168 : my_namespace = NIL;
1608 : else
1609 73570 : setNamespaceColumnVisibility(my_namespace, false);
1610 :
1611 : /*
1612 : * The join RTE itself is always made visible for unqualified column
1613 : * names. It's visible as a relation name only if it has an alias.
1614 : */
1615 73738 : nsitem->p_rel_visible = (j->alias != NULL);
1616 73738 : nsitem->p_cols_visible = true;
1617 73738 : nsitem->p_lateral_only = false;
1618 73738 : nsitem->p_lateral_ok = true;
1619 :
1620 73738 : *top_nsitem = nsitem;
1621 73738 : *namespace = lappend(my_namespace, nsitem);
1622 :
1623 73738 : return (Node *) j;
1624 : }
1625 : else
1626 0 : elog(ERROR, "unrecognized node type: %d", (int) nodeTag(n));
1627 : return NULL; /* can't get here, keep compiler quiet */
1628 : }
1629 :
1630 : /*
1631 : * buildVarFromNSColumn -
1632 : * build a Var node using ParseNamespaceColumn data
1633 : *
1634 : * This is used to construct joinaliasvars entries.
1635 : * We can assume varlevelsup should be 0, and no location is specified.
1636 : * Note also that no column SELECT privilege is requested here; that would
1637 : * happen only if the column is actually referenced in the query.
1638 : */
1639 : static Var *
1640 3188652 : buildVarFromNSColumn(ParseState *pstate, ParseNamespaceColumn *nscol)
1641 : {
1642 : Var *var;
1643 :
1644 : Assert(nscol->p_varno > 0); /* i.e., not deleted column */
1645 3188652 : var = makeVar(nscol->p_varno,
1646 3188652 : nscol->p_varattno,
1647 : nscol->p_vartype,
1648 : nscol->p_vartypmod,
1649 : nscol->p_varcollid,
1650 : 0);
1651 : /* makeVar doesn't offer parameters for these, so set by hand: */
1652 3188652 : var->varnosyn = nscol->p_varnosyn;
1653 3188652 : var->varattnosyn = nscol->p_varattnosyn;
1654 :
1655 : /* ... and update varnullingrels */
1656 3188652 : markNullableIfNeeded(pstate, var);
1657 :
1658 3188652 : return var;
1659 : }
1660 :
1661 : /*
1662 : * buildMergedJoinVar -
1663 : * generate a suitable replacement expression for a merged join column
1664 : */
1665 : static Node *
1666 1658 : buildMergedJoinVar(ParseState *pstate, JoinType jointype,
1667 : Var *l_colvar, Var *r_colvar)
1668 : {
1669 : Oid outcoltype;
1670 : int32 outcoltypmod;
1671 : Node *l_node,
1672 : *r_node,
1673 : *res_node;
1674 :
1675 1658 : outcoltype = select_common_type(pstate,
1676 1658 : list_make2(l_colvar, r_colvar),
1677 : "JOIN/USING",
1678 : NULL);
1679 1658 : outcoltypmod = select_common_typmod(pstate,
1680 1658 : list_make2(l_colvar, r_colvar),
1681 : outcoltype);
1682 :
1683 : /*
1684 : * Insert coercion functions if needed. Note that a difference in typmod
1685 : * can only happen if input has typmod but outcoltypmod is -1. In that
1686 : * case we insert a RelabelType to clearly mark that result's typmod is
1687 : * not same as input. We never need coerce_type_typmod.
1688 : */
1689 1658 : if (l_colvar->vartype != outcoltype)
1690 84 : l_node = coerce_type(pstate, (Node *) l_colvar, l_colvar->vartype,
1691 : outcoltype, outcoltypmod,
1692 : COERCION_IMPLICIT, COERCE_IMPLICIT_CAST, -1);
1693 1574 : else if (l_colvar->vartypmod != outcoltypmod)
1694 0 : l_node = (Node *) makeRelabelType((Expr *) l_colvar,
1695 : outcoltype, outcoltypmod,
1696 : InvalidOid, /* fixed below */
1697 : COERCE_IMPLICIT_CAST);
1698 : else
1699 1574 : l_node = (Node *) l_colvar;
1700 :
1701 1658 : if (r_colvar->vartype != outcoltype)
1702 30 : r_node = coerce_type(pstate, (Node *) r_colvar, r_colvar->vartype,
1703 : outcoltype, outcoltypmod,
1704 : COERCION_IMPLICIT, COERCE_IMPLICIT_CAST, -1);
1705 1628 : else if (r_colvar->vartypmod != outcoltypmod)
1706 0 : r_node = (Node *) makeRelabelType((Expr *) r_colvar,
1707 : outcoltype, outcoltypmod,
1708 : InvalidOid, /* fixed below */
1709 : COERCE_IMPLICIT_CAST);
1710 : else
1711 1628 : r_node = (Node *) r_colvar;
1712 :
1713 : /*
1714 : * Choose what to emit
1715 : */
1716 1658 : switch (jointype)
1717 : {
1718 1088 : case JOIN_INNER:
1719 :
1720 : /*
1721 : * We can use either var; prefer non-coerced one if available.
1722 : */
1723 1088 : if (IsA(l_node, Var))
1724 1058 : res_node = l_node;
1725 30 : else if (IsA(r_node, Var))
1726 30 : res_node = r_node;
1727 : else
1728 0 : res_node = l_node;
1729 1088 : break;
1730 210 : case JOIN_LEFT:
1731 : /* Always use left var */
1732 210 : res_node = l_node;
1733 210 : break;
1734 12 : case JOIN_RIGHT:
1735 : /* Always use right var */
1736 12 : res_node = r_node;
1737 12 : break;
1738 348 : case JOIN_FULL:
1739 : {
1740 : /*
1741 : * Here we must build a COALESCE expression to ensure that the
1742 : * join output is non-null if either input is.
1743 : */
1744 348 : CoalesceExpr *c = makeNode(CoalesceExpr);
1745 :
1746 348 : c->coalescetype = outcoltype;
1747 : /* coalescecollid will get set below */
1748 348 : c->args = list_make2(l_node, r_node);
1749 348 : c->location = -1;
1750 348 : res_node = (Node *) c;
1751 348 : break;
1752 : }
1753 0 : default:
1754 0 : elog(ERROR, "unrecognized join type: %d", (int) jointype);
1755 : res_node = NULL; /* keep compiler quiet */
1756 : break;
1757 : }
1758 :
1759 : /*
1760 : * Apply assign_expr_collations to fix up the collation info in the
1761 : * coercion and CoalesceExpr nodes, if we made any. This must be done now
1762 : * so that the join node's alias vars show correct collation info.
1763 : */
1764 1658 : assign_expr_collations(pstate, res_node);
1765 :
1766 1658 : return res_node;
1767 : }
1768 :
1769 : /*
1770 : * markRelsAsNulledBy -
1771 : * Mark the given jointree node and its children as nulled by join jindex
1772 : */
1773 : static void
1774 44622 : markRelsAsNulledBy(ParseState *pstate, Node *n, int jindex)
1775 : {
1776 : int varno;
1777 : ListCell *lc;
1778 :
1779 : /* Note: we can't see FromExpr here */
1780 44622 : if (IsA(n, RangeTblRef))
1781 : {
1782 43196 : varno = ((RangeTblRef *) n)->rtindex;
1783 : }
1784 1426 : else if (IsA(n, JoinExpr))
1785 : {
1786 1426 : JoinExpr *j = (JoinExpr *) n;
1787 :
1788 : /* recurse to children */
1789 1426 : markRelsAsNulledBy(pstate, j->larg, jindex);
1790 1426 : markRelsAsNulledBy(pstate, j->rarg, jindex);
1791 1426 : varno = j->rtindex;
1792 : }
1793 : else
1794 : {
1795 0 : elog(ERROR, "unrecognized node type: %d", (int) nodeTag(n));
1796 : varno = 0; /* keep compiler quiet */
1797 : }
1798 :
1799 : /*
1800 : * Now add jindex to the p_nullingrels set for relation varno. Since we
1801 : * maintain the p_nullingrels list lazily, we might need to extend it to
1802 : * make the varno'th entry exist.
1803 : */
1804 142166 : while (list_length(pstate->p_nullingrels) < varno)
1805 97544 : pstate->p_nullingrels = lappend(pstate->p_nullingrels, NULL);
1806 44622 : lc = list_nth_cell(pstate->p_nullingrels, varno - 1);
1807 44622 : lfirst(lc) = bms_add_member((Bitmapset *) lfirst(lc), jindex);
1808 44622 : }
1809 :
1810 : /*
1811 : * setNamespaceColumnVisibility -
1812 : * Convenience subroutine to update cols_visible flags in a namespace list.
1813 : */
1814 : static void
1815 73570 : setNamespaceColumnVisibility(List *namespace, bool cols_visible)
1816 : {
1817 : ListCell *lc;
1818 :
1819 315728 : foreach(lc, namespace)
1820 : {
1821 242158 : ParseNamespaceItem *nsitem = (ParseNamespaceItem *) lfirst(lc);
1822 :
1823 242158 : nsitem->p_cols_visible = cols_visible;
1824 : }
1825 73570 : }
1826 :
1827 : /*
1828 : * setNamespaceLateralState -
1829 : * Convenience subroutine to update LATERAL flags in a namespace list.
1830 : */
1831 : static void
1832 930792 : setNamespaceLateralState(List *namespace, bool lateral_only, bool lateral_ok)
1833 : {
1834 : ListCell *lc;
1835 :
1836 2295142 : foreach(lc, namespace)
1837 : {
1838 1364350 : ParseNamespaceItem *nsitem = (ParseNamespaceItem *) lfirst(lc);
1839 :
1840 1364350 : nsitem->p_lateral_only = lateral_only;
1841 1364350 : nsitem->p_lateral_ok = lateral_ok;
1842 : }
1843 930792 : }
1844 :
1845 :
1846 : /*
1847 : * transformWhereClause -
1848 : * Transform the qualification and make sure it is of type boolean.
1849 : * Used for WHERE and allied clauses.
1850 : *
1851 : * constructName does not affect the semantics, but is used in error messages
1852 : */
1853 : Node *
1854 972044 : transformWhereClause(ParseState *pstate, Node *clause,
1855 : ParseExprKind exprKind, const char *constructName)
1856 : {
1857 : Node *qual;
1858 :
1859 972044 : if (clause == NULL)
1860 703292 : return NULL;
1861 :
1862 268752 : qual = transformExpr(pstate, clause, exprKind);
1863 :
1864 268542 : qual = coerce_to_boolean(pstate, qual, constructName);
1865 :
1866 268536 : return qual;
1867 : }
1868 :
1869 :
1870 : /*
1871 : * transformLimitClause -
1872 : * Transform the expression and make sure it is of type bigint.
1873 : * Used for LIMIT and allied clauses.
1874 : *
1875 : * Note: as of Postgres 8.2, LIMIT expressions are expected to yield int8,
1876 : * rather than int4 as before.
1877 : *
1878 : * constructName does not affect the semantics, but is used in error messages
1879 : */
1880 : Node *
1881 901956 : transformLimitClause(ParseState *pstate, Node *clause,
1882 : ParseExprKind exprKind, const char *constructName,
1883 : LimitOption limitOption)
1884 : {
1885 : Node *qual;
1886 :
1887 901956 : if (clause == NULL)
1888 896778 : return NULL;
1889 :
1890 5178 : qual = transformExpr(pstate, clause, exprKind);
1891 :
1892 5172 : qual = coerce_to_specific_type(pstate, qual, INT8OID, constructName);
1893 :
1894 : /* LIMIT can't refer to any variables of the current query */
1895 5172 : checkExprIsVarFree(pstate, qual, constructName);
1896 :
1897 : /*
1898 : * Don't allow NULLs in FETCH FIRST .. WITH TIES. This test is ugly and
1899 : * extremely simplistic, in that you can pass a NULL anyway by hiding it
1900 : * inside an expression -- but this protects ruleutils against emitting an
1901 : * unadorned NULL that's not accepted back by the grammar.
1902 : */
1903 5172 : if (exprKind == EXPR_KIND_LIMIT && limitOption == LIMIT_OPTION_WITH_TIES &&
1904 48 : IsA(clause, A_Const) && castNode(A_Const, clause)->isnull)
1905 6 : ereport(ERROR,
1906 : (errcode(ERRCODE_INVALID_ROW_COUNT_IN_LIMIT_CLAUSE),
1907 : errmsg("row count cannot be null in FETCH FIRST ... WITH TIES clause")));
1908 :
1909 5166 : return qual;
1910 : }
1911 :
1912 : /*
1913 : * checkExprIsVarFree
1914 : * Check that given expr has no Vars of the current query level
1915 : * (aggregates and window functions should have been rejected already).
1916 : *
1917 : * This is used to check expressions that have to have a consistent value
1918 : * across all rows of the query, such as a LIMIT. Arguably it should reject
1919 : * volatile functions, too, but we don't do that --- whatever value the
1920 : * function gives on first execution is what you get.
1921 : *
1922 : * constructName does not affect the semantics, but is used in error messages
1923 : */
1924 : static void
1925 6978 : checkExprIsVarFree(ParseState *pstate, Node *n, const char *constructName)
1926 : {
1927 6978 : if (contain_vars_of_level(n, 0))
1928 : {
1929 6 : ereport(ERROR,
1930 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
1931 : /* translator: %s is name of a SQL construct, eg LIMIT */
1932 : errmsg("argument of %s must not contain variables",
1933 : constructName),
1934 : parser_errposition(pstate,
1935 : locate_var_of_level(n, 0))));
1936 : }
1937 6972 : }
1938 :
1939 :
1940 : /*
1941 : * checkTargetlistEntrySQL92 -
1942 : * Validate a targetlist entry found by findTargetlistEntrySQL92
1943 : *
1944 : * When we select a pre-existing tlist entry as a result of syntax such
1945 : * as "GROUP BY 1", we have to make sure it is acceptable for use in the
1946 : * indicated clause type; transformExpr() will have treated it as a regular
1947 : * targetlist item.
1948 : */
1949 : static void
1950 55658 : checkTargetlistEntrySQL92(ParseState *pstate, TargetEntry *tle,
1951 : ParseExprKind exprKind)
1952 : {
1953 55658 : switch (exprKind)
1954 : {
1955 698 : case EXPR_KIND_GROUP_BY:
1956 : /* reject aggregates and window functions */
1957 1274 : if (pstate->p_hasAggs &&
1958 576 : contain_aggs_of_level((Node *) tle->expr, 0))
1959 0 : ereport(ERROR,
1960 : (errcode(ERRCODE_GROUPING_ERROR),
1961 : /* translator: %s is name of a SQL construct, eg GROUP BY */
1962 : errmsg("aggregate functions are not allowed in %s",
1963 : ParseExprKindName(exprKind)),
1964 : parser_errposition(pstate,
1965 : locate_agg_of_level((Node *) tle->expr, 0))));
1966 710 : if (pstate->p_hasWindowFuncs &&
1967 12 : contain_windowfuncs((Node *) tle->expr))
1968 6 : ereport(ERROR,
1969 : (errcode(ERRCODE_WINDOWING_ERROR),
1970 : /* translator: %s is name of a SQL construct, eg GROUP BY */
1971 : errmsg("window functions are not allowed in %s",
1972 : ParseExprKindName(exprKind)),
1973 : parser_errposition(pstate,
1974 : locate_windowfunc((Node *) tle->expr))));
1975 692 : break;
1976 54852 : case EXPR_KIND_ORDER_BY:
1977 : /* no extra checks needed */
1978 54852 : break;
1979 108 : case EXPR_KIND_DISTINCT_ON:
1980 : /* no extra checks needed */
1981 108 : break;
1982 0 : default:
1983 0 : elog(ERROR, "unexpected exprKind in checkTargetlistEntrySQL92");
1984 : break;
1985 : }
1986 55652 : }
1987 :
1988 : /*
1989 : * findTargetlistEntrySQL92 -
1990 : * Returns the targetlist entry matching the given (untransformed) node.
1991 : * If no matching entry exists, one is created and appended to the target
1992 : * list as a "resjunk" node.
1993 : *
1994 : * This function supports the old SQL92 ORDER BY interpretation, where the
1995 : * expression is an output column name or number. If we fail to find a
1996 : * match of that sort, we fall through to the SQL99 rules. For historical
1997 : * reasons, Postgres also allows this interpretation for GROUP BY, though
1998 : * the standard never did. However, for GROUP BY we prefer a SQL99 match.
1999 : * This function is *not* used for WINDOW definitions.
2000 : *
2001 : * node the ORDER BY, GROUP BY, or DISTINCT ON expression to be matched
2002 : * tlist the target list (passed by reference so we can append to it)
2003 : * exprKind identifies clause type being processed
2004 : */
2005 : static TargetEntry *
2006 89776 : findTargetlistEntrySQL92(ParseState *pstate, Node *node, List **tlist,
2007 : ParseExprKind exprKind)
2008 : {
2009 : ListCell *tl;
2010 :
2011 : /*----------
2012 : * Handle two special cases as mandated by the SQL92 spec:
2013 : *
2014 : * 1. Bare ColumnName (no qualifier or subscripts)
2015 : * For a bare identifier, we search for a matching column name
2016 : * in the existing target list. Multiple matches are an error
2017 : * unless they refer to identical values; for example,
2018 : * we allow SELECT a, a FROM table ORDER BY a
2019 : * but not SELECT a AS b, b FROM table ORDER BY b
2020 : * If no match is found, we fall through and treat the identifier
2021 : * as an expression.
2022 : * For GROUP BY, it is incorrect to match the grouping item against
2023 : * targetlist entries: according to SQL92, an identifier in GROUP BY
2024 : * is a reference to a column name exposed by FROM, not to a target
2025 : * list column. However, many implementations (including pre-7.0
2026 : * PostgreSQL) accept this anyway. So for GROUP BY, we look first
2027 : * to see if the identifier matches any FROM column name, and only
2028 : * try for a targetlist name if it doesn't. This ensures that we
2029 : * adhere to the spec in the case where the name could be both.
2030 : * DISTINCT ON isn't in the standard, so we can do what we like there;
2031 : * we choose to make it work like ORDER BY, on the rather flimsy
2032 : * grounds that ordinary DISTINCT works on targetlist entries.
2033 : *
2034 : * 2. IntegerConstant
2035 : * This means to use the n'th item in the existing target list.
2036 : * Note that it would make no sense to order/group/distinct by an
2037 : * actual constant, so this does not create a conflict with SQL99.
2038 : * GROUP BY column-number is not allowed by SQL92, but since
2039 : * the standard has no other behavior defined for this syntax,
2040 : * we may as well accept this common extension.
2041 : *
2042 : * Note that pre-existing resjunk targets must not be used in either case,
2043 : * since the user didn't write them in his SELECT list.
2044 : *
2045 : * If neither special case applies, fall through to treat the item as
2046 : * an expression per SQL99.
2047 : *----------
2048 : */
2049 139772 : if (IsA(node, ColumnRef) &&
2050 49996 : list_length(((ColumnRef *) node)->fields) == 1 &&
2051 34012 : IsA(linitial(((ColumnRef *) node)->fields), String))
2052 : {
2053 34012 : char *name = strVal(linitial(((ColumnRef *) node)->fields));
2054 34012 : int location = ((ColumnRef *) node)->location;
2055 :
2056 34012 : if (exprKind == EXPR_KIND_GROUP_BY)
2057 : {
2058 : /*
2059 : * In GROUP BY, we must prefer a match against a FROM-clause
2060 : * column to one against the targetlist. Look to see if there is
2061 : * a matching column. If so, fall through to use SQL99 rules.
2062 : * NOTE: if name could refer ambiguously to more than one column
2063 : * name exposed by FROM, colNameToVar will ereport(ERROR). That's
2064 : * just what we want here.
2065 : *
2066 : * Small tweak for 7.4.3: ignore matches in upper query levels.
2067 : * This effectively changes the search order for bare names to (1)
2068 : * local FROM variables, (2) local targetlist aliases, (3) outer
2069 : * FROM variables, whereas before it was (1) (3) (2). SQL92 and
2070 : * SQL99 do not allow GROUPing BY an outer reference, so this
2071 : * breaks no cases that are legal per spec, and it seems a more
2072 : * self-consistent behavior.
2073 : */
2074 4346 : if (colNameToVar(pstate, name, true, location) != NULL)
2075 4272 : name = NULL;
2076 : }
2077 :
2078 34012 : if (name != NULL)
2079 : {
2080 29740 : TargetEntry *target_result = NULL;
2081 :
2082 165898 : foreach(tl, *tlist)
2083 : {
2084 136158 : TargetEntry *tle = (TargetEntry *) lfirst(tl);
2085 :
2086 136158 : if (!tle->resjunk &&
2087 135604 : strcmp(tle->resname, name) == 0)
2088 : {
2089 24810 : if (target_result != NULL)
2090 : {
2091 6 : if (!equal(target_result->expr, tle->expr))
2092 0 : ereport(ERROR,
2093 : (errcode(ERRCODE_AMBIGUOUS_COLUMN),
2094 :
2095 : /*------
2096 : translator: first %s is name of a SQL construct, eg ORDER BY */
2097 : errmsg("%s \"%s\" is ambiguous",
2098 : ParseExprKindName(exprKind),
2099 : name),
2100 : parser_errposition(pstate, location)));
2101 : }
2102 : else
2103 24804 : target_result = tle;
2104 : /* Stay in loop to check for ambiguity */
2105 : }
2106 : }
2107 29740 : if (target_result != NULL)
2108 : {
2109 : /* return the first match, after suitable validation */
2110 24804 : checkTargetlistEntrySQL92(pstate, target_result, exprKind);
2111 24804 : return target_result;
2112 : }
2113 : }
2114 : }
2115 64972 : if (IsA(node, A_Const))
2116 : {
2117 30860 : A_Const *aconst = castNode(A_Const, node);
2118 30860 : int targetlist_pos = 0;
2119 : int target_pos;
2120 :
2121 30860 : if (!IsA(&aconst->val, Integer))
2122 0 : ereport(ERROR,
2123 : (errcode(ERRCODE_SYNTAX_ERROR),
2124 : /* translator: %s is name of a SQL construct, eg ORDER BY */
2125 : errmsg("non-integer constant in %s",
2126 : ParseExprKindName(exprKind)),
2127 : parser_errposition(pstate, aconst->location)));
2128 :
2129 30860 : target_pos = intVal(&aconst->val);
2130 54648 : foreach(tl, *tlist)
2131 : {
2132 54642 : TargetEntry *tle = (TargetEntry *) lfirst(tl);
2133 :
2134 54642 : if (!tle->resjunk)
2135 : {
2136 54642 : if (++targetlist_pos == target_pos)
2137 : {
2138 : /* return the unique match, after suitable validation */
2139 30854 : checkTargetlistEntrySQL92(pstate, tle, exprKind);
2140 30848 : return tle;
2141 : }
2142 : }
2143 : }
2144 6 : ereport(ERROR,
2145 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
2146 : /* translator: %s is name of a SQL construct, eg ORDER BY */
2147 : errmsg("%s position %d is not in select list",
2148 : ParseExprKindName(exprKind), target_pos),
2149 : parser_errposition(pstate, aconst->location)));
2150 : }
2151 :
2152 : /*
2153 : * Otherwise, we have an expression, so process it per SQL99 rules.
2154 : */
2155 34112 : return findTargetlistEntrySQL99(pstate, node, tlist, exprKind);
2156 : }
2157 :
2158 : /*
2159 : * findTargetlistEntrySQL99 -
2160 : * Returns the targetlist entry matching the given (untransformed) node.
2161 : * If no matching entry exists, one is created and appended to the target
2162 : * list as a "resjunk" node.
2163 : *
2164 : * This function supports the SQL99 interpretation, wherein the expression
2165 : * is just an ordinary expression referencing input column names.
2166 : *
2167 : * node the ORDER BY, GROUP BY, etc expression to be matched
2168 : * tlist the target list (passed by reference so we can append to it)
2169 : * exprKind identifies clause type being processed
2170 : */
2171 : static TargetEntry *
2172 39332 : findTargetlistEntrySQL99(ParseState *pstate, Node *node, List **tlist,
2173 : ParseExprKind exprKind)
2174 : {
2175 : TargetEntry *target_result;
2176 : ListCell *tl;
2177 : Node *expr;
2178 :
2179 : /*
2180 : * Convert the untransformed node to a transformed expression, and search
2181 : * for a match in the tlist. NOTE: it doesn't really matter whether there
2182 : * is more than one match. Also, we are willing to match an existing
2183 : * resjunk target here, though the SQL92 cases above must ignore resjunk
2184 : * targets.
2185 : */
2186 39332 : expr = transformExpr(pstate, node, exprKind);
2187 :
2188 145682 : foreach(tl, *tlist)
2189 : {
2190 122248 : TargetEntry *tle = (TargetEntry *) lfirst(tl);
2191 : Node *texpr;
2192 :
2193 : /*
2194 : * Ignore any implicit cast on the existing tlist expression.
2195 : *
2196 : * This essentially allows the ORDER/GROUP/etc item to adopt the same
2197 : * datatype previously selected for a textually-equivalent tlist item.
2198 : * There can't be any implicit cast at top level in an ordinary SELECT
2199 : * tlist at this stage, but the case does arise with ORDER BY in an
2200 : * aggregate function.
2201 : */
2202 122248 : texpr = strip_implicit_coercions((Node *) tle->expr);
2203 :
2204 122248 : if (equal(expr, texpr))
2205 15844 : return tle;
2206 : }
2207 :
2208 : /*
2209 : * If no matches, construct a new target entry which is appended to the
2210 : * end of the target list. This target is given resjunk = true so that it
2211 : * will not be projected into the final tuple.
2212 : */
2213 23434 : target_result = transformTargetEntry(pstate, node, expr, exprKind,
2214 : NULL, true);
2215 :
2216 23434 : *tlist = lappend(*tlist, target_result);
2217 :
2218 23434 : return target_result;
2219 : }
2220 :
2221 : /*-------------------------------------------------------------------------
2222 : * Flatten out parenthesized sublists in grouping lists, and some cases
2223 : * of nested grouping sets.
2224 : *
2225 : * Inside a grouping set (ROLLUP, CUBE, or GROUPING SETS), we expect the
2226 : * content to be nested no more than 2 deep: i.e. ROLLUP((a,b),(c,d)) is
2227 : * ok, but ROLLUP((a,(b,c)),d) is flattened to ((a,b,c),d), which we then
2228 : * (later) normalize to ((a,b,c),(d)).
2229 : *
2230 : * CUBE or ROLLUP can be nested inside GROUPING SETS (but not the reverse),
2231 : * and we leave that alone if we find it. But if we see GROUPING SETS inside
2232 : * GROUPING SETS, we can flatten and normalize as follows:
2233 : * GROUPING SETS (a, (b,c), GROUPING SETS ((c,d),(e)), (f,g))
2234 : * becomes
2235 : * GROUPING SETS ((a), (b,c), (c,d), (e), (f,g))
2236 : *
2237 : * This is per the spec's syntax transformations, but these are the only such
2238 : * transformations we do in parse analysis, so that queries retain the
2239 : * originally specified grouping set syntax for CUBE and ROLLUP as much as
2240 : * possible when deparsed. (Full expansion of the result into a list of
2241 : * grouping sets is left to the planner.)
2242 : *
2243 : * When we're done, the resulting list should contain only these possible
2244 : * elements:
2245 : * - an expression
2246 : * - a CUBE or ROLLUP with a list of expressions nested 2 deep
2247 : * - a GROUPING SET containing any of:
2248 : * - expression lists
2249 : * - empty grouping sets
2250 : * - CUBE or ROLLUP nodes with lists nested 2 deep
2251 : * The return is a new list, but doesn't deep-copy the old nodes except for
2252 : * GroupingSet nodes.
2253 : *
2254 : * As a side effect, flag whether the list has any GroupingSet nodes.
2255 : *-------------------------------------------------------------------------
2256 : */
2257 : static Node *
2258 449832 : flatten_grouping_sets(Node *expr, bool toplevel, bool *hasGroupingSets)
2259 : {
2260 : /* just in case of pathological input */
2261 449832 : check_stack_depth();
2262 :
2263 449832 : if (expr == (Node *) NIL)
2264 434904 : return (Node *) NIL;
2265 :
2266 14928 : switch (expr->type)
2267 : {
2268 292 : case T_RowExpr:
2269 : {
2270 292 : RowExpr *r = (RowExpr *) expr;
2271 :
2272 292 : if (r->row_format == COERCE_IMPLICIT_CAST)
2273 292 : return flatten_grouping_sets((Node *) r->args,
2274 : false, NULL);
2275 : }
2276 0 : break;
2277 1128 : case T_GroupingSet:
2278 : {
2279 1128 : GroupingSet *gset = (GroupingSet *) expr;
2280 : ListCell *l2;
2281 1128 : List *result_set = NIL;
2282 :
2283 1128 : if (hasGroupingSets)
2284 804 : *hasGroupingSets = true;
2285 :
2286 : /*
2287 : * at the top level, we skip over all empty grouping sets; the
2288 : * caller can supply the canonical GROUP BY () if nothing is
2289 : * left.
2290 : */
2291 :
2292 1128 : if (toplevel && gset->kind == GROUPING_SET_EMPTY)
2293 18 : return (Node *) NIL;
2294 :
2295 2968 : foreach(l2, gset->content)
2296 : {
2297 1858 : Node *n1 = lfirst(l2);
2298 1858 : Node *n2 = flatten_grouping_sets(n1, false, NULL);
2299 :
2300 1858 : if (IsA(n1, GroupingSet) &&
2301 324 : ((GroupingSet *) n1)->kind == GROUPING_SET_SETS)
2302 96 : result_set = list_concat(result_set, (List *) n2);
2303 : else
2304 1762 : result_set = lappend(result_set, n2);
2305 : }
2306 :
2307 : /*
2308 : * At top level, keep the grouping set node; but if we're in a
2309 : * nested grouping set, then we need to concat the flattened
2310 : * result into the outer list if it's simply nested.
2311 : */
2312 :
2313 1110 : if (toplevel || (gset->kind != GROUPING_SET_SETS))
2314 : {
2315 1014 : return (Node *) makeGroupingSet(gset->kind, result_set, gset->location);
2316 : }
2317 : else
2318 96 : return (Node *) result_set;
2319 : }
2320 5146 : case T_List:
2321 : {
2322 5146 : List *result = NIL;
2323 : ListCell *l;
2324 :
2325 13070 : foreach(l, (List *) expr)
2326 : {
2327 7924 : Node *n = flatten_grouping_sets(lfirst(l), toplevel, hasGroupingSets);
2328 :
2329 7924 : if (n != (Node *) NIL)
2330 : {
2331 7906 : if (IsA(n, List))
2332 46 : result = list_concat(result, (List *) n);
2333 : else
2334 7860 : result = lappend(result, n);
2335 : }
2336 : }
2337 :
2338 5146 : return (Node *) result;
2339 : }
2340 8362 : default:
2341 8362 : break;
2342 : }
2343 :
2344 8362 : return expr;
2345 : }
2346 :
2347 : /*
2348 : * Transform a single expression within a GROUP BY clause or grouping set.
2349 : *
2350 : * The expression is added to the targetlist if not already present, and to the
2351 : * flatresult list (which will become the groupClause) if not already present
2352 : * there. The sortClause is consulted for operator and sort order hints.
2353 : *
2354 : * Returns the ressortgroupref of the expression.
2355 : *
2356 : * flatresult reference to flat list of SortGroupClause nodes
2357 : * seen_local bitmapset of sortgrouprefs already seen at the local level
2358 : * pstate ParseState
2359 : * gexpr node to transform
2360 : * targetlist reference to TargetEntry list
2361 : * sortClause ORDER BY clause (SortGroupClause nodes)
2362 : * exprKind expression kind
2363 : * useSQL99 SQL99 rather than SQL92 syntax
2364 : * toplevel false if within any grouping set
2365 : */
2366 : static Index
2367 8362 : transformGroupClauseExpr(List **flatresult, Bitmapset *seen_local,
2368 : ParseState *pstate, Node *gexpr,
2369 : List **targetlist, List *sortClause,
2370 : ParseExprKind exprKind, bool useSQL99, bool toplevel)
2371 : {
2372 : TargetEntry *tle;
2373 8362 : bool found = false;
2374 :
2375 8362 : if (useSQL99)
2376 1092 : tle = findTargetlistEntrySQL99(pstate, gexpr,
2377 : targetlist, exprKind);
2378 : else
2379 7270 : tle = findTargetlistEntrySQL92(pstate, gexpr,
2380 : targetlist, exprKind);
2381 :
2382 8338 : if (tle->ressortgroupref > 0)
2383 : {
2384 : ListCell *sl;
2385 :
2386 : /*
2387 : * Eliminate duplicates (GROUP BY x, x) but only at local level.
2388 : * (Duplicates in grouping sets can affect the number of returned
2389 : * rows, so can't be dropped indiscriminately.)
2390 : *
2391 : * Since we don't care about anything except the sortgroupref, we can
2392 : * use a bitmapset rather than scanning lists.
2393 : */
2394 2446 : if (bms_is_member(tle->ressortgroupref, seen_local))
2395 24 : return 0;
2396 :
2397 : /*
2398 : * If we're already in the flat clause list, we don't need to consider
2399 : * adding ourselves again.
2400 : */
2401 2422 : found = targetIsInSortList(tle, InvalidOid, *flatresult);
2402 2422 : if (found)
2403 202 : return tle->ressortgroupref;
2404 :
2405 : /*
2406 : * If the GROUP BY tlist entry also appears in ORDER BY, copy operator
2407 : * info from the (first) matching ORDER BY item. This means that if
2408 : * you write something like "GROUP BY foo ORDER BY foo USING <<<", the
2409 : * GROUP BY operation silently takes on the equality semantics implied
2410 : * by the ORDER BY. There are two reasons to do this: it improves the
2411 : * odds that we can implement both GROUP BY and ORDER BY with a single
2412 : * sort step, and it allows the user to choose the equality semantics
2413 : * used by GROUP BY, should she be working with a datatype that has
2414 : * more than one equality operator.
2415 : *
2416 : * If we're in a grouping set, though, we force our requested ordering
2417 : * to be NULLS LAST, because if we have any hope of using a sorted agg
2418 : * for the job, we're going to be tacking on generated NULL values
2419 : * after the corresponding groups. If the user demands nulls first,
2420 : * another sort step is going to be inevitable, but that's the
2421 : * planner's problem.
2422 : */
2423 :
2424 3118 : foreach(sl, sortClause)
2425 : {
2426 2928 : SortGroupClause *sc = (SortGroupClause *) lfirst(sl);
2427 :
2428 2928 : if (sc->tleSortGroupRef == tle->ressortgroupref)
2429 : {
2430 2030 : SortGroupClause *grpc = copyObject(sc);
2431 :
2432 2030 : if (!toplevel)
2433 508 : grpc->nulls_first = false;
2434 2030 : *flatresult = lappend(*flatresult, grpc);
2435 2030 : found = true;
2436 2030 : break;
2437 : }
2438 : }
2439 : }
2440 :
2441 : /*
2442 : * If no match in ORDER BY, just add it to the result using default
2443 : * sort/group semantics.
2444 : */
2445 8112 : if (!found)
2446 6082 : *flatresult = addTargetToGroupList(pstate, tle,
2447 : *flatresult, *targetlist,
2448 : exprLocation(gexpr));
2449 :
2450 : /*
2451 : * _something_ must have assigned us a sortgroupref by now...
2452 : */
2453 :
2454 8112 : return tle->ressortgroupref;
2455 : }
2456 :
2457 : /*
2458 : * Transform a list of expressions within a GROUP BY clause or grouping set.
2459 : *
2460 : * The list of expressions belongs to a single clause within which duplicates
2461 : * can be safely eliminated.
2462 : *
2463 : * Returns an integer list of ressortgroupref values.
2464 : *
2465 : * flatresult reference to flat list of SortGroupClause nodes
2466 : * pstate ParseState
2467 : * list nodes to transform
2468 : * targetlist reference to TargetEntry list
2469 : * sortClause ORDER BY clause (SortGroupClause nodes)
2470 : * exprKind expression kind
2471 : * useSQL99 SQL99 rather than SQL92 syntax
2472 : * toplevel false if within any grouping set
2473 : */
2474 : static List *
2475 246 : transformGroupClauseList(List **flatresult,
2476 : ParseState *pstate, List *list,
2477 : List **targetlist, List *sortClause,
2478 : ParseExprKind exprKind, bool useSQL99, bool toplevel)
2479 : {
2480 246 : Bitmapset *seen_local = NULL;
2481 246 : List *result = NIL;
2482 : ListCell *gl;
2483 :
2484 762 : foreach(gl, list)
2485 : {
2486 516 : Node *gexpr = (Node *) lfirst(gl);
2487 :
2488 516 : Index ref = transformGroupClauseExpr(flatresult,
2489 : seen_local,
2490 : pstate,
2491 : gexpr,
2492 : targetlist,
2493 : sortClause,
2494 : exprKind,
2495 : useSQL99,
2496 : toplevel);
2497 :
2498 516 : if (ref > 0)
2499 : {
2500 504 : seen_local = bms_add_member(seen_local, ref);
2501 504 : result = lappend_int(result, ref);
2502 : }
2503 : }
2504 :
2505 246 : return result;
2506 : }
2507 :
2508 : /*
2509 : * Transform a grouping set and (recursively) its content.
2510 : *
2511 : * The grouping set might be a GROUPING SETS node with other grouping sets
2512 : * inside it, but SETS within SETS have already been flattened out before
2513 : * reaching here.
2514 : *
2515 : * Returns the transformed node, which now contains SIMPLE nodes with lists
2516 : * of ressortgrouprefs rather than expressions.
2517 : *
2518 : * flatresult reference to flat list of SortGroupClause nodes
2519 : * pstate ParseState
2520 : * gset grouping set to transform
2521 : * targetlist reference to TargetEntry list
2522 : * sortClause ORDER BY clause (SortGroupClause nodes)
2523 : * exprKind expression kind
2524 : * useSQL99 SQL99 rather than SQL92 syntax
2525 : * toplevel false if within any grouping set
2526 : */
2527 : static Node *
2528 1014 : transformGroupingSet(List **flatresult,
2529 : ParseState *pstate, GroupingSet *gset,
2530 : List **targetlist, List *sortClause,
2531 : ParseExprKind exprKind, bool useSQL99, bool toplevel)
2532 : {
2533 : ListCell *gl;
2534 1014 : List *content = NIL;
2535 :
2536 : Assert(toplevel || gset->kind != GROUPING_SET_SETS);
2537 :
2538 2776 : foreach(gl, gset->content)
2539 : {
2540 1762 : Node *n = lfirst(gl);
2541 :
2542 1762 : if (IsA(n, List))
2543 : {
2544 246 : List *l = transformGroupClauseList(flatresult,
2545 : pstate, (List *) n,
2546 : targetlist, sortClause,
2547 : exprKind, useSQL99, false);
2548 :
2549 246 : content = lappend(content, makeGroupingSet(GROUPING_SET_SIMPLE,
2550 : l,
2551 : exprLocation(n)));
2552 : }
2553 1516 : else if (IsA(n, GroupingSet))
2554 : {
2555 228 : GroupingSet *gset2 = (GroupingSet *) lfirst(gl);
2556 :
2557 228 : content = lappend(content, transformGroupingSet(flatresult,
2558 : pstate, gset2,
2559 : targetlist, sortClause,
2560 : exprKind, useSQL99, false));
2561 : }
2562 : else
2563 : {
2564 1288 : Index ref = transformGroupClauseExpr(flatresult,
2565 : NULL,
2566 : pstate,
2567 : n,
2568 : targetlist,
2569 : sortClause,
2570 : exprKind,
2571 : useSQL99,
2572 : false);
2573 :
2574 2576 : content = lappend(content, makeGroupingSet(GROUPING_SET_SIMPLE,
2575 1288 : list_make1_int(ref),
2576 : exprLocation(n)));
2577 : }
2578 : }
2579 :
2580 : /* Arbitrarily cap the size of CUBE, which has exponential growth */
2581 1014 : if (gset->kind == GROUPING_SET_CUBE)
2582 : {
2583 184 : if (list_length(content) > 12)
2584 0 : ereport(ERROR,
2585 : (errcode(ERRCODE_TOO_MANY_COLUMNS),
2586 : errmsg("CUBE is limited to 12 elements"),
2587 : parser_errposition(pstate, gset->location)));
2588 : }
2589 :
2590 1014 : return (Node *) makeGroupingSet(gset->kind, content, gset->location);
2591 : }
2592 :
2593 :
2594 : /*
2595 : * transformGroupClause -
2596 : * transform a GROUP BY clause
2597 : *
2598 : * GROUP BY items will be added to the targetlist (as resjunk columns)
2599 : * if not already present, so the targetlist must be passed by reference.
2600 : *
2601 : * This is also used for window PARTITION BY clauses (which act almost the
2602 : * same, but are always interpreted per SQL99 rules).
2603 : *
2604 : * Grouping sets make this a lot more complex than it was. Our goal here is
2605 : * twofold: we make a flat list of SortGroupClause nodes referencing each
2606 : * distinct expression used for grouping, with those expressions added to the
2607 : * targetlist if needed. At the same time, we build the groupingSets tree,
2608 : * which stores only ressortgrouprefs as integer lists inside GroupingSet nodes
2609 : * (possibly nested, but limited in depth: a GROUPING_SET_SETS node can contain
2610 : * nested SIMPLE, CUBE or ROLLUP nodes, but not more sets - we flatten that
2611 : * out; while CUBE and ROLLUP can contain only SIMPLE nodes).
2612 : *
2613 : * We skip much of the hard work if there are no grouping sets.
2614 : *
2615 : * One subtlety is that the groupClause list can end up empty while the
2616 : * groupingSets list is not; this happens if there are only empty grouping
2617 : * sets, or an explicit GROUP BY (). This has the same effect as specifying
2618 : * aggregates or a HAVING clause with no GROUP BY; the output is one row per
2619 : * grouping set even if the input is empty.
2620 : *
2621 : * Returns the transformed (flat) groupClause.
2622 : *
2623 : * pstate ParseState
2624 : * grouplist clause to transform
2625 : * groupingSets reference to list to contain the grouping set tree
2626 : * targetlist reference to TargetEntry list
2627 : * sortClause ORDER BY clause (SortGroupClause nodes)
2628 : * exprKind expression kind
2629 : * useSQL99 SQL99 rather than SQL92 syntax
2630 : */
2631 : List *
2632 439758 : transformGroupClause(ParseState *pstate, List *grouplist, List **groupingSets,
2633 : List **targetlist, List *sortClause,
2634 : ParseExprKind exprKind, bool useSQL99)
2635 : {
2636 439758 : List *result = NIL;
2637 : List *flat_grouplist;
2638 439758 : List *gsets = NIL;
2639 : ListCell *gl;
2640 439758 : bool hasGroupingSets = false;
2641 439758 : Bitmapset *seen_local = NULL;
2642 :
2643 : /*
2644 : * Recursively flatten implicit RowExprs. (Technically this is only needed
2645 : * for GROUP BY, per the syntax rules for grouping sets, but we do it
2646 : * anyway.)
2647 : */
2648 439758 : flat_grouplist = (List *) flatten_grouping_sets((Node *) grouplist,
2649 : true,
2650 : &hasGroupingSets);
2651 :
2652 : /*
2653 : * If the list is now empty, but hasGroupingSets is true, it's because we
2654 : * elided redundant empty grouping sets. Restore a single empty grouping
2655 : * set to leave a canonical form: GROUP BY ()
2656 : */
2657 :
2658 439758 : if (flat_grouplist == NIL && hasGroupingSets)
2659 : {
2660 18 : flat_grouplist = list_make1(makeGroupingSet(GROUPING_SET_EMPTY,
2661 : NIL,
2662 : exprLocation((Node *) grouplist)));
2663 : }
2664 :
2665 447096 : foreach(gl, flat_grouplist)
2666 : {
2667 7362 : Node *gexpr = (Node *) lfirst(gl);
2668 :
2669 7362 : if (IsA(gexpr, GroupingSet))
2670 : {
2671 804 : GroupingSet *gset = (GroupingSet *) gexpr;
2672 :
2673 804 : switch (gset->kind)
2674 : {
2675 18 : case GROUPING_SET_EMPTY:
2676 18 : gsets = lappend(gsets, gset);
2677 18 : break;
2678 0 : case GROUPING_SET_SIMPLE:
2679 : /* can't happen */
2680 : Assert(false);
2681 0 : break;
2682 786 : case GROUPING_SET_SETS:
2683 : case GROUPING_SET_CUBE:
2684 : case GROUPING_SET_ROLLUP:
2685 786 : gsets = lappend(gsets,
2686 786 : transformGroupingSet(&result,
2687 : pstate, gset,
2688 : targetlist, sortClause,
2689 : exprKind, useSQL99, true));
2690 786 : break;
2691 : }
2692 804 : }
2693 : else
2694 : {
2695 6558 : Index ref = transformGroupClauseExpr(&result, seen_local,
2696 : pstate, gexpr,
2697 : targetlist, sortClause,
2698 : exprKind, useSQL99, true);
2699 :
2700 6534 : if (ref > 0)
2701 : {
2702 6522 : seen_local = bms_add_member(seen_local, ref);
2703 6522 : if (hasGroupingSets)
2704 36 : gsets = lappend(gsets,
2705 72 : makeGroupingSet(GROUPING_SET_SIMPLE,
2706 36 : list_make1_int(ref),
2707 : exprLocation(gexpr)));
2708 : }
2709 : }
2710 : }
2711 :
2712 : /* parser should prevent this */
2713 : Assert(gsets == NIL || groupingSets != NULL);
2714 :
2715 439734 : if (groupingSets)
2716 437064 : *groupingSets = gsets;
2717 :
2718 439734 : return result;
2719 : }
2720 :
2721 : /*
2722 : * transformSortClause -
2723 : * transform an ORDER BY clause
2724 : *
2725 : * ORDER BY items will be added to the targetlist (as resjunk columns)
2726 : * if not already present, so the targetlist must be passed by reference.
2727 : *
2728 : * This is also used for window and aggregate ORDER BY clauses (which act
2729 : * almost the same, but are always interpreted per SQL99 rules).
2730 : */
2731 : List *
2732 492486 : transformSortClause(ParseState *pstate,
2733 : List *orderlist,
2734 : List **targetlist,
2735 : ParseExprKind exprKind,
2736 : bool useSQL99)
2737 : {
2738 492486 : List *sortlist = NIL;
2739 : ListCell *olitem;
2740 :
2741 578832 : foreach(olitem, orderlist)
2742 : {
2743 86388 : SortBy *sortby = (SortBy *) lfirst(olitem);
2744 : TargetEntry *tle;
2745 :
2746 86388 : if (useSQL99)
2747 4128 : tle = findTargetlistEntrySQL99(pstate, sortby->node,
2748 : targetlist, exprKind);
2749 : else
2750 82260 : tle = findTargetlistEntrySQL92(pstate, sortby->node,
2751 : targetlist, exprKind);
2752 :
2753 86352 : sortlist = addTargetToSortList(pstate, tle,
2754 : sortlist, *targetlist, sortby);
2755 : }
2756 :
2757 492444 : return sortlist;
2758 : }
2759 :
2760 : /*
2761 : * transformWindowDefinitions -
2762 : * transform window definitions (WindowDef to WindowClause)
2763 : */
2764 : List *
2765 437040 : transformWindowDefinitions(ParseState *pstate,
2766 : List *windowdefs,
2767 : List **targetlist)
2768 : {
2769 437040 : List *result = NIL;
2770 437040 : Index winref = 0;
2771 : ListCell *lc;
2772 :
2773 439656 : foreach(lc, windowdefs)
2774 : {
2775 2682 : WindowDef *windef = (WindowDef *) lfirst(lc);
2776 2682 : WindowClause *refwc = NULL;
2777 : List *partitionClause;
2778 : List *orderClause;
2779 2682 : Oid rangeopfamily = InvalidOid;
2780 2682 : Oid rangeopcintype = InvalidOid;
2781 : WindowClause *wc;
2782 :
2783 2682 : winref++;
2784 :
2785 : /*
2786 : * Check for duplicate window names.
2787 : */
2788 3222 : if (windef->name &&
2789 540 : findWindowClause(result, windef->name) != NULL)
2790 6 : ereport(ERROR,
2791 : (errcode(ERRCODE_WINDOWING_ERROR),
2792 : errmsg("window \"%s\" is already defined", windef->name),
2793 : parser_errposition(pstate, windef->location)));
2794 :
2795 : /*
2796 : * If it references a previous window, look that up.
2797 : */
2798 2676 : if (windef->refname)
2799 : {
2800 24 : refwc = findWindowClause(result, windef->refname);
2801 24 : if (refwc == NULL)
2802 0 : ereport(ERROR,
2803 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2804 : errmsg("window \"%s\" does not exist",
2805 : windef->refname),
2806 : parser_errposition(pstate, windef->location)));
2807 : }
2808 :
2809 : /*
2810 : * Transform PARTITION and ORDER specs, if any. These are treated
2811 : * almost exactly like top-level GROUP BY and ORDER BY clauses,
2812 : * including the special handling of nondefault operator semantics.
2813 : */
2814 2676 : orderClause = transformSortClause(pstate,
2815 : windef->orderClause,
2816 : targetlist,
2817 : EXPR_KIND_WINDOW_ORDER,
2818 : true /* force SQL99 rules */ );
2819 2670 : partitionClause = transformGroupClause(pstate,
2820 : windef->partitionClause,
2821 : NULL,
2822 : targetlist,
2823 : orderClause,
2824 : EXPR_KIND_WINDOW_PARTITION,
2825 : true /* force SQL99 rules */ );
2826 :
2827 : /*
2828 : * And prepare the new WindowClause.
2829 : */
2830 2670 : wc = makeNode(WindowClause);
2831 2670 : wc->name = windef->name;
2832 2670 : wc->refname = windef->refname;
2833 :
2834 : /*
2835 : * Per spec, a windowdef that references a previous one copies the
2836 : * previous partition clause (and mustn't specify its own). It can
2837 : * specify its own ordering clause, but only if the previous one had
2838 : * none. It always specifies its own frame clause, and the previous
2839 : * one must not have a frame clause. Yeah, it's bizarre that each of
2840 : * these cases works differently, but SQL:2008 says so; see 7.11
2841 : * <window clause> syntax rule 10 and general rule 1. The frame
2842 : * clause rule is especially bizarre because it makes "OVER foo"
2843 : * different from "OVER (foo)", and requires the latter to throw an
2844 : * error if foo has a nondefault frame clause. Well, ours not to
2845 : * reason why, but we do go out of our way to throw a useful error
2846 : * message for such cases.
2847 : */
2848 2670 : if (refwc)
2849 : {
2850 24 : if (partitionClause)
2851 0 : ereport(ERROR,
2852 : (errcode(ERRCODE_WINDOWING_ERROR),
2853 : errmsg("cannot override PARTITION BY clause of window \"%s\"",
2854 : windef->refname),
2855 : parser_errposition(pstate, windef->location)));
2856 24 : wc->partitionClause = copyObject(refwc->partitionClause);
2857 : }
2858 : else
2859 2646 : wc->partitionClause = partitionClause;
2860 2670 : if (refwc)
2861 : {
2862 24 : if (orderClause && refwc->orderClause)
2863 0 : ereport(ERROR,
2864 : (errcode(ERRCODE_WINDOWING_ERROR),
2865 : errmsg("cannot override ORDER BY clause of window \"%s\"",
2866 : windef->refname),
2867 : parser_errposition(pstate, windef->location)));
2868 24 : if (orderClause)
2869 : {
2870 0 : wc->orderClause = orderClause;
2871 0 : wc->copiedOrder = false;
2872 : }
2873 : else
2874 : {
2875 24 : wc->orderClause = copyObject(refwc->orderClause);
2876 24 : wc->copiedOrder = true;
2877 : }
2878 : }
2879 : else
2880 : {
2881 2646 : wc->orderClause = orderClause;
2882 2646 : wc->copiedOrder = false;
2883 : }
2884 2670 : if (refwc && refwc->frameOptions != FRAMEOPTION_DEFAULTS)
2885 : {
2886 : /*
2887 : * Use this message if this is a WINDOW clause, or if it's an OVER
2888 : * clause that includes ORDER BY or framing clauses. (We already
2889 : * rejected PARTITION BY above, so no need to check that.)
2890 : */
2891 0 : if (windef->name ||
2892 0 : orderClause || windef->frameOptions != FRAMEOPTION_DEFAULTS)
2893 0 : ereport(ERROR,
2894 : (errcode(ERRCODE_WINDOWING_ERROR),
2895 : errmsg("cannot copy window \"%s\" because it has a frame clause",
2896 : windef->refname),
2897 : parser_errposition(pstate, windef->location)));
2898 : /* Else this clause is just OVER (foo), so say this: */
2899 0 : ereport(ERROR,
2900 : (errcode(ERRCODE_WINDOWING_ERROR),
2901 : errmsg("cannot copy window \"%s\" because it has a frame clause",
2902 : windef->refname),
2903 : errhint("Omit the parentheses in this OVER clause."),
2904 : parser_errposition(pstate, windef->location)));
2905 : }
2906 2670 : wc->frameOptions = windef->frameOptions;
2907 :
2908 : /*
2909 : * RANGE offset PRECEDING/FOLLOWING requires exactly one ORDER BY
2910 : * column; check that and get its sort opfamily info.
2911 : */
2912 2670 : if ((wc->frameOptions & FRAMEOPTION_RANGE) &&
2913 1896 : (wc->frameOptions & (FRAMEOPTION_START_OFFSET |
2914 : FRAMEOPTION_END_OFFSET)))
2915 : {
2916 : SortGroupClause *sortcl;
2917 : Node *sortkey;
2918 : int16 rangestrategy;
2919 :
2920 636 : if (list_length(wc->orderClause) != 1)
2921 18 : ereport(ERROR,
2922 : (errcode(ERRCODE_WINDOWING_ERROR),
2923 : errmsg("RANGE with offset PRECEDING/FOLLOWING requires exactly one ORDER BY column"),
2924 : parser_errposition(pstate, windef->location)));
2925 618 : sortcl = linitial_node(SortGroupClause, wc->orderClause);
2926 618 : sortkey = get_sortgroupclause_expr(sortcl, *targetlist);
2927 : /* Find the sort operator in pg_amop */
2928 618 : if (!get_ordering_op_properties(sortcl->sortop,
2929 : &rangeopfamily,
2930 : &rangeopcintype,
2931 : &rangestrategy))
2932 0 : elog(ERROR, "operator %u is not a valid ordering operator",
2933 : sortcl->sortop);
2934 : /* Record properties of sort ordering */
2935 618 : wc->inRangeColl = exprCollation(sortkey);
2936 618 : wc->inRangeAsc = (rangestrategy == BTLessStrategyNumber);
2937 618 : wc->inRangeNullsFirst = sortcl->nulls_first;
2938 : }
2939 :
2940 : /* Per spec, GROUPS mode requires an ORDER BY clause */
2941 2652 : if (wc->frameOptions & FRAMEOPTION_GROUPS)
2942 : {
2943 174 : if (wc->orderClause == NIL)
2944 6 : ereport(ERROR,
2945 : (errcode(ERRCODE_WINDOWING_ERROR),
2946 : errmsg("GROUPS mode requires an ORDER BY clause"),
2947 : parser_errposition(pstate, windef->location)));
2948 : }
2949 :
2950 : /* Process frame offset expressions */
2951 2646 : wc->startOffset = transformFrameOffset(pstate, wc->frameOptions,
2952 : rangeopfamily, rangeopcintype,
2953 : &wc->startInRangeFunc,
2954 : windef->startOffset);
2955 2622 : wc->endOffset = transformFrameOffset(pstate, wc->frameOptions,
2956 : rangeopfamily, rangeopcintype,
2957 : &wc->endInRangeFunc,
2958 : windef->endOffset);
2959 2616 : wc->winref = winref;
2960 :
2961 2616 : result = lappend(result, wc);
2962 : }
2963 :
2964 436974 : return result;
2965 : }
2966 :
2967 : /*
2968 : * transformDistinctClause -
2969 : * transform a DISTINCT clause
2970 : *
2971 : * Since we may need to add items to the query's targetlist, that list
2972 : * is passed by reference.
2973 : *
2974 : * As with GROUP BY, we absorb the sorting semantics of ORDER BY as much as
2975 : * possible into the distinctClause. This avoids a possible need to re-sort,
2976 : * and allows the user to choose the equality semantics used by DISTINCT,
2977 : * should she be working with a datatype that has more than one equality
2978 : * operator.
2979 : *
2980 : * is_agg is true if we are transforming an aggregate(DISTINCT ...)
2981 : * function call. This does not affect any behavior, only the phrasing
2982 : * of error messages.
2983 : */
2984 : List *
2985 3378 : transformDistinctClause(ParseState *pstate,
2986 : List **targetlist, List *sortClause, bool is_agg)
2987 : {
2988 3378 : List *result = NIL;
2989 : ListCell *slitem;
2990 : ListCell *tlitem;
2991 :
2992 : /*
2993 : * The distinctClause should consist of all ORDER BY items followed by all
2994 : * other non-resjunk targetlist items. There must not be any resjunk
2995 : * ORDER BY items --- that would imply that we are sorting by a value that
2996 : * isn't necessarily unique within a DISTINCT group, so the results
2997 : * wouldn't be well-defined. This construction ensures we follow the rule
2998 : * that sortClause and distinctClause match; in fact the sortClause will
2999 : * always be a prefix of distinctClause.
3000 : *
3001 : * Note a corner case: the same TLE could be in the ORDER BY list multiple
3002 : * times with different sortops. We have to include it in the
3003 : * distinctClause the same way to preserve the prefix property. The net
3004 : * effect will be that the TLE value will be made unique according to both
3005 : * sortops.
3006 : */
3007 3988 : foreach(slitem, sortClause)
3008 : {
3009 646 : SortGroupClause *scl = (SortGroupClause *) lfirst(slitem);
3010 646 : TargetEntry *tle = get_sortgroupclause_tle(scl, *targetlist);
3011 :
3012 646 : if (tle->resjunk)
3013 36 : ereport(ERROR,
3014 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3015 : is_agg ?
3016 : errmsg("in an aggregate with DISTINCT, ORDER BY expressions must appear in argument list") :
3017 : errmsg("for SELECT DISTINCT, ORDER BY expressions must appear in select list"),
3018 : parser_errposition(pstate,
3019 : exprLocation((Node *) tle->expr))));
3020 610 : result = lappend(result, copyObject(scl));
3021 : }
3022 :
3023 : /*
3024 : * Now add any remaining non-resjunk tlist items, using default sort/group
3025 : * semantics for their data types.
3026 : */
3027 12814 : foreach(tlitem, *targetlist)
3028 : {
3029 9472 : TargetEntry *tle = (TargetEntry *) lfirst(tlitem);
3030 :
3031 9472 : if (tle->resjunk)
3032 4 : continue; /* ignore junk */
3033 9468 : result = addTargetToGroupList(pstate, tle,
3034 : result, *targetlist,
3035 9468 : exprLocation((Node *) tle->expr));
3036 : }
3037 :
3038 : /*
3039 : * Complain if we found nothing to make DISTINCT. Returning an empty list
3040 : * would cause the parsed Query to look like it didn't have DISTINCT, with
3041 : * results that would probably surprise the user. Note: this case is
3042 : * presently impossible for aggregates because of grammar restrictions,
3043 : * but we check anyway.
3044 : */
3045 3342 : if (result == NIL)
3046 0 : ereport(ERROR,
3047 : (errcode(ERRCODE_SYNTAX_ERROR),
3048 : is_agg ?
3049 : errmsg("an aggregate with DISTINCT must have at least one argument") :
3050 : errmsg("SELECT DISTINCT must have at least one column")));
3051 :
3052 3342 : return result;
3053 : }
3054 :
3055 : /*
3056 : * transformDistinctOnClause -
3057 : * transform a DISTINCT ON clause
3058 : *
3059 : * Since we may need to add items to the query's targetlist, that list
3060 : * is passed by reference.
3061 : *
3062 : * As with GROUP BY, we absorb the sorting semantics of ORDER BY as much as
3063 : * possible into the distinctClause. This avoids a possible need to re-sort,
3064 : * and allows the user to choose the equality semantics used by DISTINCT,
3065 : * should she be working with a datatype that has more than one equality
3066 : * operator.
3067 : */
3068 : List *
3069 180 : transformDistinctOnClause(ParseState *pstate, List *distinctlist,
3070 : List **targetlist, List *sortClause)
3071 : {
3072 180 : List *result = NIL;
3073 180 : List *sortgrouprefs = NIL;
3074 : bool skipped_sortitem;
3075 : ListCell *lc;
3076 : ListCell *lc2;
3077 :
3078 : /*
3079 : * Add all the DISTINCT ON expressions to the tlist (if not already
3080 : * present, they are added as resjunk items). Assign sortgroupref numbers
3081 : * to them, and make a list of these numbers. (NB: we rely below on the
3082 : * sortgrouprefs list being one-for-one with the original distinctlist.
3083 : * Also notice that we could have duplicate DISTINCT ON expressions and
3084 : * hence duplicate entries in sortgrouprefs.)
3085 : */
3086 420 : foreach(lc, distinctlist)
3087 : {
3088 246 : Node *dexpr = (Node *) lfirst(lc);
3089 : int sortgroupref;
3090 : TargetEntry *tle;
3091 :
3092 246 : tle = findTargetlistEntrySQL92(pstate, dexpr, targetlist,
3093 : EXPR_KIND_DISTINCT_ON);
3094 240 : sortgroupref = assignSortGroupRef(tle, *targetlist);
3095 240 : sortgrouprefs = lappend_int(sortgrouprefs, sortgroupref);
3096 : }
3097 :
3098 : /*
3099 : * If the user writes both DISTINCT ON and ORDER BY, adopt the sorting
3100 : * semantics from ORDER BY items that match DISTINCT ON items, and also
3101 : * adopt their column sort order. We insist that the distinctClause and
3102 : * sortClause match, so throw error if we find the need to add any more
3103 : * distinctClause items after we've skipped an ORDER BY item that wasn't
3104 : * in DISTINCT ON.
3105 : */
3106 174 : skipped_sortitem = false;
3107 408 : foreach(lc, sortClause)
3108 : {
3109 240 : SortGroupClause *scl = (SortGroupClause *) lfirst(lc);
3110 :
3111 240 : if (list_member_int(sortgrouprefs, scl->tleSortGroupRef))
3112 : {
3113 180 : if (skipped_sortitem)
3114 6 : ereport(ERROR,
3115 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3116 : errmsg("SELECT DISTINCT ON expressions must match initial ORDER BY expressions"),
3117 : parser_errposition(pstate,
3118 : get_matching_location(scl->tleSortGroupRef,
3119 : sortgrouprefs,
3120 : distinctlist))));
3121 : else
3122 174 : result = lappend(result, copyObject(scl));
3123 : }
3124 : else
3125 60 : skipped_sortitem = true;
3126 : }
3127 :
3128 : /*
3129 : * Now add any remaining DISTINCT ON items, using default sort/group
3130 : * semantics for their data types. (Note: this is pretty questionable; if
3131 : * the ORDER BY list doesn't include all the DISTINCT ON items and more
3132 : * besides, you certainly aren't using DISTINCT ON in the intended way,
3133 : * and you probably aren't going to get consistent results. It might be
3134 : * better to throw an error or warning here. But historically we've
3135 : * allowed it, so keep doing so.)
3136 : */
3137 396 : forboth(lc, distinctlist, lc2, sortgrouprefs)
3138 : {
3139 228 : Node *dexpr = (Node *) lfirst(lc);
3140 228 : int sortgroupref = lfirst_int(lc2);
3141 228 : TargetEntry *tle = get_sortgroupref_tle(sortgroupref, *targetlist);
3142 :
3143 228 : if (targetIsInSortList(tle, InvalidOid, result))
3144 168 : continue; /* already in list (with some semantics) */
3145 60 : if (skipped_sortitem)
3146 0 : ereport(ERROR,
3147 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3148 : errmsg("SELECT DISTINCT ON expressions must match initial ORDER BY expressions"),
3149 : parser_errposition(pstate, exprLocation(dexpr))));
3150 60 : result = addTargetToGroupList(pstate, tle,
3151 : result, *targetlist,
3152 : exprLocation(dexpr));
3153 : }
3154 :
3155 : /*
3156 : * An empty result list is impossible here because of grammar
3157 : * restrictions.
3158 : */
3159 : Assert(result != NIL);
3160 :
3161 168 : return result;
3162 : }
3163 :
3164 : /*
3165 : * get_matching_location
3166 : * Get the exprLocation of the exprs member corresponding to the
3167 : * (first) member of sortgrouprefs that equals sortgroupref.
3168 : *
3169 : * This is used so that we can point at a troublesome DISTINCT ON entry.
3170 : * (Note that we need to use the original untransformed DISTINCT ON list
3171 : * item, as whatever TLE it corresponds to will very possibly have a
3172 : * parse location pointing to some matching entry in the SELECT list
3173 : * or ORDER BY list.)
3174 : */
3175 : static int
3176 6 : get_matching_location(int sortgroupref, List *sortgrouprefs, List *exprs)
3177 : {
3178 : ListCell *lcs;
3179 : ListCell *lce;
3180 :
3181 12 : forboth(lcs, sortgrouprefs, lce, exprs)
3182 : {
3183 12 : if (lfirst_int(lcs) == sortgroupref)
3184 6 : return exprLocation((Node *) lfirst(lce));
3185 : }
3186 : /* if no match, caller blew it */
3187 0 : elog(ERROR, "get_matching_location: no matching sortgroupref");
3188 : return -1; /* keep compiler quiet */
3189 : }
3190 :
3191 : /*
3192 : * resolve_unique_index_expr
3193 : * Infer a unique index from a list of indexElems, for ON
3194 : * CONFLICT clause
3195 : *
3196 : * Perform parse analysis of expressions and columns appearing within ON
3197 : * CONFLICT clause. During planning, the returned list of expressions is used
3198 : * to infer which unique index to use.
3199 : */
3200 : static List *
3201 1262 : resolve_unique_index_expr(ParseState *pstate, InferClause *infer,
3202 : Relation heapRel)
3203 : {
3204 1262 : List *result = NIL;
3205 : ListCell *l;
3206 :
3207 2816 : foreach(l, infer->indexElems)
3208 : {
3209 1560 : IndexElem *ielem = (IndexElem *) lfirst(l);
3210 1560 : InferenceElem *pInfer = makeNode(InferenceElem);
3211 : Node *parse;
3212 :
3213 : /*
3214 : * Raw grammar re-uses CREATE INDEX infrastructure for unique index
3215 : * inference clause, and so will accept opclasses by name and so on.
3216 : *
3217 : * Make no attempt to match ASC or DESC ordering or NULLS FIRST/NULLS
3218 : * LAST ordering, since those are not significant for inference
3219 : * purposes (any unique index matching the inference specification in
3220 : * other regards is accepted indifferently). Actively reject this as
3221 : * wrong-headed.
3222 : */
3223 1560 : if (ielem->ordering != SORTBY_DEFAULT)
3224 0 : ereport(ERROR,
3225 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3226 : errmsg("ASC/DESC is not allowed in ON CONFLICT clause"),
3227 : parser_errposition(pstate,
3228 : exprLocation((Node *) infer))));
3229 1560 : if (ielem->nulls_ordering != SORTBY_NULLS_DEFAULT)
3230 0 : ereport(ERROR,
3231 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3232 : errmsg("NULLS FIRST/LAST is not allowed in ON CONFLICT clause"),
3233 : parser_errposition(pstate,
3234 : exprLocation((Node *) infer))));
3235 :
3236 1560 : if (!ielem->expr)
3237 : {
3238 : /* Simple index attribute */
3239 : ColumnRef *n;
3240 :
3241 : /*
3242 : * Grammar won't have built raw expression for us in event of
3243 : * plain column reference. Create one directly, and perform
3244 : * expression transformation. Planner expects this, and performs
3245 : * its own normalization for the purposes of matching against
3246 : * pg_index.
3247 : */
3248 1398 : n = makeNode(ColumnRef);
3249 1398 : n->fields = list_make1(makeString(ielem->name));
3250 : /* Location is approximately that of inference specification */
3251 1398 : n->location = infer->location;
3252 1398 : parse = (Node *) n;
3253 : }
3254 : else
3255 : {
3256 : /* Do parse transformation of the raw expression */
3257 162 : parse = (Node *) ielem->expr;
3258 : }
3259 :
3260 : /*
3261 : * transformExpr() will reject subqueries, aggregates, window
3262 : * functions, and SRFs, based on being passed
3263 : * EXPR_KIND_INDEX_EXPRESSION. So we needn't worry about those
3264 : * further ... not that they would match any available index
3265 : * expression anyway.
3266 : */
3267 1560 : pInfer->expr = transformExpr(pstate, parse, EXPR_KIND_INDEX_EXPRESSION);
3268 :
3269 : /* Perform lookup of collation and operator class as required */
3270 1554 : if (!ielem->collation)
3271 1512 : pInfer->infercollid = InvalidOid;
3272 : else
3273 42 : pInfer->infercollid = LookupCollation(pstate, ielem->collation,
3274 42 : exprLocation(pInfer->expr));
3275 :
3276 1554 : if (!ielem->opclass)
3277 1512 : pInfer->inferopclass = InvalidOid;
3278 : else
3279 42 : pInfer->inferopclass = get_opclass_oid(BTREE_AM_OID,
3280 : ielem->opclass, false);
3281 :
3282 1554 : result = lappend(result, pInfer);
3283 : }
3284 :
3285 1256 : return result;
3286 : }
3287 :
3288 : /*
3289 : * transformOnConflictArbiter -
3290 : * transform arbiter expressions in an ON CONFLICT clause.
3291 : *
3292 : * Transformed expressions used to infer one unique index relation to serve as
3293 : * an ON CONFLICT arbiter. Partial unique indexes may be inferred using WHERE
3294 : * clause from inference specification clause.
3295 : */
3296 : void
3297 1472 : transformOnConflictArbiter(ParseState *pstate,
3298 : OnConflictClause *onConflictClause,
3299 : List **arbiterExpr, Node **arbiterWhere,
3300 : Oid *constraint)
3301 : {
3302 1472 : InferClause *infer = onConflictClause->infer;
3303 :
3304 1472 : *arbiterExpr = NIL;
3305 1472 : *arbiterWhere = NULL;
3306 1472 : *constraint = InvalidOid;
3307 :
3308 1472 : if (onConflictClause->action == ONCONFLICT_UPDATE && !infer)
3309 6 : ereport(ERROR,
3310 : (errcode(ERRCODE_SYNTAX_ERROR),
3311 : errmsg("ON CONFLICT DO UPDATE requires inference specification or constraint name"),
3312 : errhint("For example, ON CONFLICT (column_name)."),
3313 : parser_errposition(pstate,
3314 : exprLocation((Node *) onConflictClause))));
3315 :
3316 : /*
3317 : * To simplify certain aspects of its design, speculative insertion into
3318 : * system catalogs is disallowed
3319 : */
3320 1466 : if (IsCatalogRelation(pstate->p_target_relation))
3321 0 : ereport(ERROR,
3322 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3323 : errmsg("ON CONFLICT is not supported with system catalog tables"),
3324 : parser_errposition(pstate,
3325 : exprLocation((Node *) onConflictClause))));
3326 :
3327 : /* Same applies to table used by logical decoding as catalog table */
3328 1466 : if (RelationIsUsedAsCatalogTable(pstate->p_target_relation))
3329 0 : ereport(ERROR,
3330 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3331 : errmsg("ON CONFLICT is not supported on table \"%s\" used as a catalog table",
3332 : RelationGetRelationName(pstate->p_target_relation)),
3333 : parser_errposition(pstate,
3334 : exprLocation((Node *) onConflictClause))));
3335 :
3336 : /* ON CONFLICT DO NOTHING does not require an inference clause */
3337 1466 : if (infer)
3338 : {
3339 1310 : if (infer->indexElems)
3340 1262 : *arbiterExpr = resolve_unique_index_expr(pstate, infer,
3341 : pstate->p_target_relation);
3342 :
3343 : /*
3344 : * Handling inference WHERE clause (for partial unique index
3345 : * inference)
3346 : */
3347 1304 : if (infer->whereClause)
3348 42 : *arbiterWhere = transformExpr(pstate, infer->whereClause,
3349 : EXPR_KIND_INDEX_PREDICATE);
3350 :
3351 : /*
3352 : * If the arbiter is specified by constraint name, get the constraint
3353 : * OID and mark the constrained columns as requiring SELECT privilege,
3354 : * in the same way as would have happened if the arbiter had been
3355 : * specified by explicit reference to the constraint's index columns.
3356 : */
3357 1304 : if (infer->conname)
3358 : {
3359 48 : Oid relid = RelationGetRelid(pstate->p_target_relation);
3360 48 : RTEPermissionInfo *perminfo = pstate->p_target_nsitem->p_perminfo;
3361 : Bitmapset *conattnos;
3362 :
3363 48 : conattnos = get_relation_constraint_attnos(relid, infer->conname,
3364 : false, constraint);
3365 :
3366 : /* Make sure the rel as a whole is marked for SELECT access */
3367 48 : perminfo->requiredPerms |= ACL_SELECT;
3368 : /* Mark the constrained columns as requiring SELECT access */
3369 48 : perminfo->selectedCols = bms_add_members(perminfo->selectedCols,
3370 : conattnos);
3371 : }
3372 : }
3373 :
3374 : /*
3375 : * It's convenient to form a list of expressions based on the
3376 : * representation used by CREATE INDEX, since the same restrictions are
3377 : * appropriate (e.g. on subqueries). However, from here on, a dedicated
3378 : * primnode representation is used for inference elements, and so
3379 : * assign_query_collations() can be trusted to do the right thing with the
3380 : * post parse analysis query tree inference clause representation.
3381 : */
3382 1460 : }
3383 :
3384 : /*
3385 : * addTargetToSortList
3386 : * If the given targetlist entry isn't already in the SortGroupClause
3387 : * list, add it to the end of the list, using the given sort ordering
3388 : * info.
3389 : *
3390 : * Returns the updated SortGroupClause list.
3391 : */
3392 : List *
3393 86724 : addTargetToSortList(ParseState *pstate, TargetEntry *tle,
3394 : List *sortlist, List *targetlist, SortBy *sortby)
3395 : {
3396 86724 : Oid restype = exprType((Node *) tle->expr);
3397 : Oid sortop;
3398 : Oid eqop;
3399 : bool hashable;
3400 : bool reverse;
3401 : int location;
3402 : ParseCallbackState pcbstate;
3403 :
3404 : /* if tlist item is an UNKNOWN literal, change it to TEXT */
3405 86724 : if (restype == UNKNOWNOID)
3406 : {
3407 12 : tle->expr = (Expr *) coerce_type(pstate, (Node *) tle->expr,
3408 : restype, TEXTOID, -1,
3409 : COERCION_IMPLICIT,
3410 : COERCE_IMPLICIT_CAST,
3411 : -1);
3412 12 : restype = TEXTOID;
3413 : }
3414 :
3415 : /*
3416 : * Rather than clutter the API of get_sort_group_operators and the other
3417 : * functions we're about to use, make use of error context callback to
3418 : * mark any error reports with a parse position. We point to the operator
3419 : * location if present, else to the expression being sorted. (NB: use the
3420 : * original untransformed expression here; the TLE entry might well point
3421 : * at a duplicate expression in the regular SELECT list.)
3422 : */
3423 86724 : location = sortby->location;
3424 86724 : if (location < 0)
3425 86504 : location = exprLocation(sortby->node);
3426 86724 : setup_parser_errposition_callback(&pcbstate, pstate, location);
3427 :
3428 : /* determine the sortop, eqop, and directionality */
3429 86724 : switch (sortby->sortby_dir)
3430 : {
3431 83796 : case SORTBY_DEFAULT:
3432 : case SORTBY_ASC:
3433 83796 : get_sort_group_operators(restype,
3434 : true, true, false,
3435 : &sortop, &eqop, NULL,
3436 : &hashable);
3437 83790 : reverse = false;
3438 83790 : break;
3439 2708 : case SORTBY_DESC:
3440 2708 : get_sort_group_operators(restype,
3441 : false, true, true,
3442 : NULL, &eqop, &sortop,
3443 : &hashable);
3444 2708 : reverse = true;
3445 2708 : break;
3446 220 : case SORTBY_USING:
3447 : Assert(sortby->useOp != NIL);
3448 220 : sortop = compatible_oper_opid(sortby->useOp,
3449 : restype,
3450 : restype,
3451 : false);
3452 :
3453 : /*
3454 : * Verify it's a valid ordering operator, fetch the corresponding
3455 : * equality operator, and determine whether to consider it like
3456 : * ASC or DESC.
3457 : */
3458 220 : eqop = get_equality_op_for_ordering_op(sortop, &reverse);
3459 220 : if (!OidIsValid(eqop))
3460 0 : ereport(ERROR,
3461 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
3462 : errmsg("operator %s is not a valid ordering operator",
3463 : strVal(llast(sortby->useOp))),
3464 : errhint("Ordering operators must be \"<\" or \">\" members of btree operator families.")));
3465 :
3466 : /*
3467 : * Also see if the equality operator is hashable.
3468 : */
3469 220 : hashable = op_hashjoinable(eqop, restype);
3470 220 : break;
3471 0 : default:
3472 0 : elog(ERROR, "unrecognized sortby_dir: %d", sortby->sortby_dir);
3473 : sortop = InvalidOid; /* keep compiler quiet */
3474 : eqop = InvalidOid;
3475 : hashable = false;
3476 : reverse = false;
3477 : break;
3478 : }
3479 :
3480 86718 : cancel_parser_errposition_callback(&pcbstate);
3481 :
3482 : /* avoid making duplicate sortlist entries */
3483 86718 : if (!targetIsInSortList(tle, sortop, sortlist))
3484 : {
3485 86718 : SortGroupClause *sortcl = makeNode(SortGroupClause);
3486 :
3487 86718 : sortcl->tleSortGroupRef = assignSortGroupRef(tle, targetlist);
3488 :
3489 86718 : sortcl->eqop = eqop;
3490 86718 : sortcl->sortop = sortop;
3491 86718 : sortcl->hashable = hashable;
3492 :
3493 86718 : switch (sortby->sortby_nulls)
3494 : {
3495 84774 : case SORTBY_NULLS_DEFAULT:
3496 : /* NULLS FIRST is default for DESC; other way for ASC */
3497 84774 : sortcl->nulls_first = reverse;
3498 84774 : break;
3499 276 : case SORTBY_NULLS_FIRST:
3500 276 : sortcl->nulls_first = true;
3501 276 : break;
3502 1668 : case SORTBY_NULLS_LAST:
3503 1668 : sortcl->nulls_first = false;
3504 1668 : break;
3505 0 : default:
3506 0 : elog(ERROR, "unrecognized sortby_nulls: %d",
3507 : sortby->sortby_nulls);
3508 : break;
3509 : }
3510 :
3511 86718 : sortlist = lappend(sortlist, sortcl);
3512 : }
3513 :
3514 86718 : return sortlist;
3515 : }
3516 :
3517 : /*
3518 : * addTargetToGroupList
3519 : * If the given targetlist entry isn't already in the SortGroupClause
3520 : * list, add it to the end of the list, using default sort/group
3521 : * semantics.
3522 : *
3523 : * This is very similar to addTargetToSortList, except that we allow the
3524 : * case where only a grouping (equality) operator can be found, and that
3525 : * the TLE is considered "already in the list" if it appears there with any
3526 : * sorting semantics.
3527 : *
3528 : * location is the parse location to be fingered in event of trouble. Note
3529 : * that we can't rely on exprLocation(tle->expr), because that might point
3530 : * to a SELECT item that matches the GROUP BY item; it'd be pretty confusing
3531 : * to report such a location.
3532 : *
3533 : * Returns the updated SortGroupClause list.
3534 : */
3535 : static List *
3536 15610 : addTargetToGroupList(ParseState *pstate, TargetEntry *tle,
3537 : List *grouplist, List *targetlist, int location)
3538 : {
3539 15610 : Oid restype = exprType((Node *) tle->expr);
3540 :
3541 : /* if tlist item is an UNKNOWN literal, change it to TEXT */
3542 15610 : if (restype == UNKNOWNOID)
3543 : {
3544 16 : tle->expr = (Expr *) coerce_type(pstate, (Node *) tle->expr,
3545 : restype, TEXTOID, -1,
3546 : COERCION_IMPLICIT,
3547 : COERCE_IMPLICIT_CAST,
3548 : -1);
3549 16 : restype = TEXTOID;
3550 : }
3551 :
3552 : /* avoid making duplicate grouplist entries */
3553 15610 : if (!targetIsInSortList(tle, InvalidOid, grouplist))
3554 : {
3555 15024 : SortGroupClause *grpcl = makeNode(SortGroupClause);
3556 : Oid sortop;
3557 : Oid eqop;
3558 : bool hashable;
3559 : ParseCallbackState pcbstate;
3560 :
3561 15024 : setup_parser_errposition_callback(&pcbstate, pstate, location);
3562 :
3563 : /* determine the eqop and optional sortop */
3564 15024 : get_sort_group_operators(restype,
3565 : false, true, false,
3566 : &sortop, &eqop, NULL,
3567 : &hashable);
3568 :
3569 15024 : cancel_parser_errposition_callback(&pcbstate);
3570 :
3571 15024 : grpcl->tleSortGroupRef = assignSortGroupRef(tle, targetlist);
3572 15024 : grpcl->eqop = eqop;
3573 15024 : grpcl->sortop = sortop;
3574 15024 : grpcl->nulls_first = false; /* OK with or without sortop */
3575 15024 : grpcl->hashable = hashable;
3576 :
3577 15024 : grouplist = lappend(grouplist, grpcl);
3578 : }
3579 :
3580 15610 : return grouplist;
3581 : }
3582 :
3583 : /*
3584 : * assignSortGroupRef
3585 : * Assign the targetentry an unused ressortgroupref, if it doesn't
3586 : * already have one. Return the assigned or pre-existing refnumber.
3587 : *
3588 : * 'tlist' is the targetlist containing (or to contain) the given targetentry.
3589 : */
3590 : Index
3591 133746 : assignSortGroupRef(TargetEntry *tle, List *tlist)
3592 : {
3593 : Index maxRef;
3594 : ListCell *l;
3595 :
3596 133746 : if (tle->ressortgroupref) /* already has one? */
3597 748 : return tle->ressortgroupref;
3598 :
3599 : /* easiest way to pick an unused refnumber: max used + 1 */
3600 132998 : maxRef = 0;
3601 751022 : foreach(l, tlist)
3602 : {
3603 618024 : Index ref = ((TargetEntry *) lfirst(l))->ressortgroupref;
3604 :
3605 618024 : if (ref > maxRef)
3606 90726 : maxRef = ref;
3607 : }
3608 132998 : tle->ressortgroupref = maxRef + 1;
3609 132998 : return tle->ressortgroupref;
3610 : }
3611 :
3612 : /*
3613 : * targetIsInSortList
3614 : * Is the given target item already in the sortlist?
3615 : * If sortop is not InvalidOid, also test for a match to the sortop.
3616 : *
3617 : * It is not an oversight that this function ignores the nulls_first flag.
3618 : * We check sortop when determining if an ORDER BY item is redundant with
3619 : * earlier ORDER BY items, because it's conceivable that "ORDER BY
3620 : * foo USING <, foo USING <<<" is not redundant, if <<< distinguishes
3621 : * values that < considers equal. We need not check nulls_first
3622 : * however, because a lower-order column with the same sortop but
3623 : * opposite nulls direction is redundant. Also, we can consider
3624 : * ORDER BY foo ASC, foo DESC redundant, so check for a commutator match.
3625 : *
3626 : * Works for both ordering and grouping lists (sortop would normally be
3627 : * InvalidOid when considering grouping). Note that the main reason we need
3628 : * this routine (and not just a quick test for nonzeroness of ressortgroupref)
3629 : * is that a TLE might be in only one of the lists.
3630 : */
3631 : bool
3632 106732 : targetIsInSortList(TargetEntry *tle, Oid sortop, List *sortList)
3633 : {
3634 106732 : Index ref = tle->ressortgroupref;
3635 : ListCell *l;
3636 :
3637 : /* no need to scan list if tle has no marker */
3638 106732 : if (ref == 0)
3639 102360 : return false;
3640 :
3641 5532 : foreach(l, sortList)
3642 : {
3643 2508 : SortGroupClause *scl = (SortGroupClause *) lfirst(l);
3644 :
3645 2508 : if (scl->tleSortGroupRef == ref &&
3646 0 : (sortop == InvalidOid ||
3647 0 : sortop == scl->sortop ||
3648 0 : sortop == get_commutator(scl->sortop)))
3649 1348 : return true;
3650 : }
3651 3024 : return false;
3652 : }
3653 :
3654 : /*
3655 : * findWindowClause
3656 : * Find the named WindowClause in the list, or return NULL if not there
3657 : */
3658 : static WindowClause *
3659 564 : findWindowClause(List *wclist, const char *name)
3660 : {
3661 : ListCell *l;
3662 :
3663 570 : foreach(l, wclist)
3664 : {
3665 36 : WindowClause *wc = (WindowClause *) lfirst(l);
3666 :
3667 36 : if (wc->name && strcmp(wc->name, name) == 0)
3668 30 : return wc;
3669 : }
3670 :
3671 534 : return NULL;
3672 : }
3673 :
3674 : /*
3675 : * transformFrameOffset
3676 : * Process a window frame offset expression
3677 : *
3678 : * In RANGE mode, rangeopfamily is the sort opfamily for the input ORDER BY
3679 : * column, and rangeopcintype is the input data type the sort operator is
3680 : * registered with. We expect the in_range function to be registered with
3681 : * that same type. (In binary-compatible cases, it might be different from
3682 : * the input column's actual type, so we can't use that for the lookups.)
3683 : * We'll return the OID of the in_range function to *inRangeFunc.
3684 : */
3685 : static Node *
3686 5268 : transformFrameOffset(ParseState *pstate, int frameOptions,
3687 : Oid rangeopfamily, Oid rangeopcintype, Oid *inRangeFunc,
3688 : Node *clause)
3689 : {
3690 5268 : const char *constructName = NULL;
3691 : Node *node;
3692 :
3693 5268 : *inRangeFunc = InvalidOid; /* default result */
3694 :
3695 : /* Quick exit if no offset expression */
3696 5268 : if (clause == NULL)
3697 3438 : return NULL;
3698 :
3699 1830 : if (frameOptions & FRAMEOPTION_ROWS)
3700 : {
3701 : /* Transform the raw expression tree */
3702 396 : node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_ROWS);
3703 :
3704 : /*
3705 : * Like LIMIT clause, simply coerce to int8
3706 : */
3707 396 : constructName = "ROWS";
3708 396 : node = coerce_to_specific_type(pstate, node, INT8OID, constructName);
3709 : }
3710 1434 : else if (frameOptions & FRAMEOPTION_RANGE)
3711 : {
3712 : /*
3713 : * We must look up the in_range support function that's to be used,
3714 : * possibly choosing one of several, and coerce the "offset" value to
3715 : * the appropriate input type.
3716 : */
3717 : Oid nodeType;
3718 : Oid preferredType;
3719 1152 : int nfuncs = 0;
3720 1152 : int nmatches = 0;
3721 1152 : Oid selectedType = InvalidOid;
3722 1152 : Oid selectedFunc = InvalidOid;
3723 : CatCList *proclist;
3724 : int i;
3725 :
3726 : /* Transform the raw expression tree */
3727 1152 : node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_RANGE);
3728 1152 : nodeType = exprType(node);
3729 :
3730 : /*
3731 : * If there are multiple candidates, we'll prefer the one that exactly
3732 : * matches nodeType; or if nodeType is as yet unknown, prefer the one
3733 : * that exactly matches the sort column type. (The second rule is
3734 : * like what we do for "known_type operator unknown".)
3735 : */
3736 1152 : preferredType = (nodeType != UNKNOWNOID) ? nodeType : rangeopcintype;
3737 :
3738 : /* Find the in_range support functions applicable to this case */
3739 1152 : proclist = SearchSysCacheList2(AMPROCNUM,
3740 : ObjectIdGetDatum(rangeopfamily),
3741 : ObjectIdGetDatum(rangeopcintype));
3742 7344 : for (i = 0; i < proclist->n_members; i++)
3743 : {
3744 6192 : HeapTuple proctup = &proclist->members[i]->tuple;
3745 6192 : Form_pg_amproc procform = (Form_pg_amproc) GETSTRUCT(proctup);
3746 :
3747 : /* The search will find all support proc types; ignore others */
3748 6192 : if (procform->amprocnum != BTINRANGE_PROC)
3749 4434 : continue;
3750 1758 : nfuncs++;
3751 :
3752 : /* Ignore function if given value can't be coerced to that type */
3753 1758 : if (!can_coerce_type(1, &nodeType, &procform->amprocrighttype,
3754 : COERCION_IMPLICIT))
3755 330 : continue;
3756 1428 : nmatches++;
3757 :
3758 : /* Remember preferred match, or any match if didn't find that */
3759 1428 : if (selectedType != preferredType)
3760 : {
3761 1368 : selectedType = procform->amprocrighttype;
3762 1368 : selectedFunc = procform->amproc;
3763 : }
3764 : }
3765 1152 : ReleaseCatCacheList(proclist);
3766 :
3767 : /*
3768 : * Throw error if needed. It seems worth taking the trouble to
3769 : * distinguish "no support at all" from "you didn't match any
3770 : * available offset type".
3771 : */
3772 1152 : if (nfuncs == 0)
3773 6 : ereport(ERROR,
3774 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3775 : errmsg("RANGE with offset PRECEDING/FOLLOWING is not supported for column type %s",
3776 : format_type_be(rangeopcintype)),
3777 : parser_errposition(pstate, exprLocation(node))));
3778 1146 : if (nmatches == 0)
3779 18 : ereport(ERROR,
3780 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3781 : errmsg("RANGE with offset PRECEDING/FOLLOWING is not supported for column type %s and offset type %s",
3782 : format_type_be(rangeopcintype),
3783 : format_type_be(nodeType)),
3784 : errhint("Cast the offset value to an appropriate type."),
3785 : parser_errposition(pstate, exprLocation(node))));
3786 1128 : if (nmatches != 1 && selectedType != preferredType)
3787 0 : ereport(ERROR,
3788 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3789 : errmsg("RANGE with offset PRECEDING/FOLLOWING has multiple interpretations for column type %s and offset type %s",
3790 : format_type_be(rangeopcintype),
3791 : format_type_be(nodeType)),
3792 : errhint("Cast the offset value to the exact intended type."),
3793 : parser_errposition(pstate, exprLocation(node))));
3794 :
3795 : /* OK, coerce the offset to the right type */
3796 1128 : constructName = "RANGE";
3797 1128 : node = coerce_to_specific_type(pstate, node,
3798 : selectedType, constructName);
3799 1128 : *inRangeFunc = selectedFunc;
3800 : }
3801 282 : else if (frameOptions & FRAMEOPTION_GROUPS)
3802 : {
3803 : /* Transform the raw expression tree */
3804 282 : node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_GROUPS);
3805 :
3806 : /*
3807 : * Like LIMIT clause, simply coerce to int8
3808 : */
3809 282 : constructName = "GROUPS";
3810 282 : node = coerce_to_specific_type(pstate, node, INT8OID, constructName);
3811 : }
3812 : else
3813 : {
3814 : Assert(false);
3815 0 : node = NULL;
3816 : }
3817 :
3818 : /* Disallow variables in frame offsets */
3819 1806 : checkExprIsVarFree(pstate, node, constructName);
3820 :
3821 1800 : return node;
3822 : }
|