Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * parse_clause.c
4 : * handle clauses in parser
5 : *
6 : * Portions Copyright (c) 1996-2025, 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 "rewrite/rewriteManip.h"
44 : #include "utils/builtins.h"
45 : #include "utils/catcache.h"
46 : #include "utils/lsyscache.h"
47 : #include "utils/rel.h"
48 : #include "utils/syscache.h"
49 :
50 :
51 : static int extractRemainingColumns(ParseState *pstate,
52 : ParseNamespaceColumn *src_nscolumns,
53 : List *src_colnames,
54 : List **src_colnos,
55 : List **res_colnames, List **res_colvars,
56 : ParseNamespaceColumn *res_nscolumns);
57 : static Node *transformJoinUsingClause(ParseState *pstate,
58 : List *leftVars, List *rightVars);
59 : static Node *transformJoinOnClause(ParseState *pstate, JoinExpr *j,
60 : List *namespace);
61 : static ParseNamespaceItem *transformTableEntry(ParseState *pstate, RangeVar *r);
62 : static ParseNamespaceItem *transformRangeSubselect(ParseState *pstate,
63 : RangeSubselect *r);
64 : static ParseNamespaceItem *transformRangeFunction(ParseState *pstate,
65 : RangeFunction *r);
66 : static ParseNamespaceItem *transformRangeTableFunc(ParseState *pstate,
67 : RangeTableFunc *rtf);
68 : static TableSampleClause *transformRangeTableSample(ParseState *pstate,
69 : RangeTableSample *rts);
70 : static ParseNamespaceItem *getNSItemForSpecialRelationTypes(ParseState *pstate,
71 : RangeVar *rv);
72 : static Node *transformFromClauseItem(ParseState *pstate, Node *n,
73 : ParseNamespaceItem **top_nsitem,
74 : List **namespace);
75 : static Var *buildVarFromNSColumn(ParseState *pstate,
76 : ParseNamespaceColumn *nscol);
77 : static Node *buildMergedJoinVar(ParseState *pstate, JoinType jointype,
78 : Var *l_colvar, Var *r_colvar);
79 : static void markRelsAsNulledBy(ParseState *pstate, Node *n, int jindex);
80 : static void setNamespaceColumnVisibility(List *namespace, bool cols_visible);
81 : static void setNamespaceLateralState(List *namespace,
82 : bool lateral_only, bool lateral_ok);
83 : static void checkExprIsVarFree(ParseState *pstate, Node *n,
84 : const char *constructName);
85 : static TargetEntry *findTargetlistEntrySQL92(ParseState *pstate, Node *node,
86 : List **tlist, ParseExprKind exprKind);
87 : static TargetEntry *findTargetlistEntrySQL99(ParseState *pstate, Node *node,
88 : List **tlist, ParseExprKind exprKind);
89 : static int get_matching_location(int sortgroupref,
90 : List *sortgrouprefs, List *exprs);
91 : static List *resolve_unique_index_expr(ParseState *pstate, InferClause *infer,
92 : Relation heapRel);
93 : static List *addTargetToGroupList(ParseState *pstate, TargetEntry *tle,
94 : List *grouplist, List *targetlist, int location);
95 : static WindowClause *findWindowClause(List *wclist, const char *name);
96 : static Node *transformFrameOffset(ParseState *pstate, int frameOptions,
97 : Oid rangeopfamily, Oid rangeopcintype, Oid *inRangeFunc,
98 : Node *clause);
99 :
100 :
101 : /*
102 : * transformFromClause -
103 : * Process the FROM clause and add items to the query's range table,
104 : * joinlist, and namespace.
105 : *
106 : * Note: we assume that the pstate's p_rtable, p_joinlist, and p_namespace
107 : * lists were initialized to NIL when the pstate was created.
108 : * We will add onto any entries already present --- this is needed for rule
109 : * processing, as well as for UPDATE and DELETE.
110 : */
111 : void
112 479506 : transformFromClause(ParseState *pstate, List *frmList)
113 : {
114 : ListCell *fl;
115 :
116 : /*
117 : * The grammar will have produced a list of RangeVars, RangeSubselects,
118 : * RangeFunctions, and/or JoinExprs. Transform each one (possibly adding
119 : * entries to the rtable), check for duplicate refnames, and then add it
120 : * to the joinlist and namespace.
121 : *
122 : * Note we must process the items left-to-right for proper handling of
123 : * LATERAL references.
124 : */
125 822314 : foreach(fl, frmList)
126 : {
127 343484 : Node *n = lfirst(fl);
128 : ParseNamespaceItem *nsitem;
129 : List *namespace;
130 :
131 343484 : n = transformFromClauseItem(pstate, n,
132 : &nsitem,
133 : &namespace);
134 :
135 342814 : checkNameSpaceConflicts(pstate, pstate->p_namespace, namespace);
136 :
137 : /* Mark the new namespace items as visible only to LATERAL */
138 342808 : setNamespaceLateralState(namespace, true, true);
139 :
140 342808 : pstate->p_joinlist = lappend(pstate->p_joinlist, n);
141 342808 : pstate->p_namespace = list_concat(pstate->p_namespace, namespace);
142 : }
143 :
144 : /*
145 : * We're done parsing the FROM list, so make all namespace items
146 : * unconditionally visible. Note that this will also reset lateral_only
147 : * for any namespace items that were already present when we were called;
148 : * but those should have been that way already.
149 : */
150 478830 : setNamespaceLateralState(pstate->p_namespace, false, true);
151 478830 : }
152 :
153 : /*
154 : * setTargetTable
155 : * Add the target relation of INSERT/UPDATE/DELETE/MERGE to the range table,
156 : * and make the special links to it in the ParseState.
157 : *
158 : * We also open the target relation and acquire a write lock on it.
159 : * This must be done before processing the FROM list, in case the target
160 : * is also mentioned as a source relation --- we want to be sure to grab
161 : * the write lock before any read lock.
162 : *
163 : * If alsoSource is true, add the target to the query's joinlist and
164 : * namespace. For INSERT, we don't want the target to be joined to;
165 : * it's a destination of tuples, not a source. MERGE is actually
166 : * both, but we'll add it separately to joinlist and namespace, so
167 : * doing nothing (like INSERT) is correct here. For UPDATE/DELETE,
168 : * we do need to scan or join the target. (NOTE: we do not bother
169 : * to check for namespace conflict; we assume that the namespace was
170 : * initially empty in these cases.)
171 : *
172 : * Finally, we mark the relation as requiring the permissions specified
173 : * by requiredPerms.
174 : *
175 : * Returns the rangetable index of the target relation.
176 : */
177 : int
178 92492 : setTargetTable(ParseState *pstate, RangeVar *relation,
179 : bool inh, bool alsoSource, AclMode requiredPerms)
180 : {
181 : ParseNamespaceItem *nsitem;
182 :
183 : /*
184 : * ENRs hide tables of the same name, so we need to check for them first.
185 : * In contrast, CTEs don't hide tables (for this purpose).
186 : */
187 174372 : if (relation->schemaname == NULL &&
188 81880 : scanNameSpaceForENR(pstate, relation->relname))
189 6 : ereport(ERROR,
190 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
191 : errmsg("relation \"%s\" cannot be the target of a modifying statement",
192 : relation->relname)));
193 :
194 : /* Close old target; this could only happen for multi-action rules */
195 92486 : if (pstate->p_target_relation != NULL)
196 0 : table_close(pstate->p_target_relation, NoLock);
197 :
198 : /*
199 : * Open target rel and grab suitable lock (which we will hold till end of
200 : * transaction).
201 : *
202 : * free_parsestate() will eventually do the corresponding table_close(),
203 : * but *not* release the lock.
204 : */
205 92486 : pstate->p_target_relation = parserOpenTable(pstate, relation,
206 : RowExclusiveLock);
207 :
208 : /*
209 : * Now build an RTE and a ParseNamespaceItem.
210 : */
211 92466 : nsitem = addRangeTableEntryForRelation(pstate, pstate->p_target_relation,
212 : RowExclusiveLock,
213 : relation->alias, inh, false);
214 :
215 : /* remember the RTE/nsitem as being the query target */
216 92466 : pstate->p_target_nsitem = nsitem;
217 :
218 : /*
219 : * Override addRangeTableEntry's default ACL_SELECT permissions check, and
220 : * instead mark target table as requiring exactly the specified
221 : * permissions.
222 : *
223 : * If we find an explicit reference to the rel later during parse
224 : * analysis, we will add the ACL_SELECT bit back again; see
225 : * markVarForSelectPriv and its callers.
226 : */
227 92466 : nsitem->p_perminfo->requiredPerms = requiredPerms;
228 :
229 : /*
230 : * If UPDATE/DELETE, add table to joinlist and namespace.
231 : */
232 92466 : if (alsoSource)
233 17702 : addNSItemToQuery(pstate, nsitem, true, true, true);
234 :
235 92466 : return nsitem->p_rtindex;
236 : }
237 :
238 : /*
239 : * Extract all not-in-common columns from column lists of a source table
240 : *
241 : * src_nscolumns and src_colnames describe the source table.
242 : *
243 : * *src_colnos initially contains the column numbers of the already-merged
244 : * columns. We add to it the column number of each additional column.
245 : * Also append to *res_colnames the name of each additional column,
246 : * append to *res_colvars a Var for each additional column, and copy the
247 : * columns' nscolumns data into res_nscolumns[] (which is caller-allocated
248 : * space that had better be big enough).
249 : *
250 : * Returns the number of columns added.
251 : */
252 : static int
253 151040 : extractRemainingColumns(ParseState *pstate,
254 : ParseNamespaceColumn *src_nscolumns,
255 : List *src_colnames,
256 : List **src_colnos,
257 : List **res_colnames, List **res_colvars,
258 : ParseNamespaceColumn *res_nscolumns)
259 : {
260 151040 : int colcount = 0;
261 : Bitmapset *prevcols;
262 : int attnum;
263 : ListCell *lc;
264 :
265 : /*
266 : * While we could just test "list_member_int(*src_colnos, attnum)" to
267 : * detect already-merged columns in the loop below, that would be O(N^2)
268 : * for a wide input table. Instead build a bitmapset of just the merged
269 : * USING columns, which we won't add to within the main loop.
270 : */
271 151040 : prevcols = NULL;
272 154360 : foreach(lc, *src_colnos)
273 : {
274 3320 : prevcols = bms_add_member(prevcols, lfirst_int(lc));
275 : }
276 :
277 151040 : attnum = 0;
278 2987322 : foreach(lc, src_colnames)
279 : {
280 2836282 : char *colname = strVal(lfirst(lc));
281 :
282 2836282 : attnum++;
283 : /* Non-dropped and not already merged? */
284 2836282 : if (colname[0] != '\0' && !bms_is_member(attnum, prevcols))
285 : {
286 : /* Yes, so emit it as next output column */
287 2832482 : *src_colnos = lappend_int(*src_colnos, attnum);
288 2832482 : *res_colnames = lappend(*res_colnames, lfirst(lc));
289 2832482 : *res_colvars = lappend(*res_colvars,
290 2832482 : buildVarFromNSColumn(pstate,
291 2832482 : src_nscolumns + attnum - 1));
292 : /* Copy the input relation's nscolumn data for this column */
293 2832482 : res_nscolumns[colcount] = src_nscolumns[attnum - 1];
294 2832482 : colcount++;
295 : }
296 : }
297 151040 : return colcount;
298 : }
299 :
300 : /* transformJoinUsingClause()
301 : * Build a complete ON clause from a partially-transformed USING list.
302 : * We are given lists of nodes representing left and right match columns.
303 : * Result is a transformed qualification expression.
304 : */
305 : static Node *
306 1460 : transformJoinUsingClause(ParseState *pstate,
307 : List *leftVars, List *rightVars)
308 : {
309 : Node *result;
310 1460 : List *andargs = NIL;
311 : ListCell *lvars,
312 : *rvars;
313 :
314 : /*
315 : * We cheat a little bit here by building an untransformed operator tree
316 : * whose leaves are the already-transformed Vars. This requires collusion
317 : * from transformExpr(), which normally could be expected to complain
318 : * about already-transformed subnodes. However, this does mean that we
319 : * have to mark the columns as requiring SELECT privilege for ourselves;
320 : * transformExpr() won't do it.
321 : */
322 3120 : forboth(lvars, leftVars, rvars, rightVars)
323 : {
324 1660 : Var *lvar = (Var *) lfirst(lvars);
325 1660 : Var *rvar = (Var *) lfirst(rvars);
326 : A_Expr *e;
327 :
328 : /* Require read access to the join variables */
329 1660 : markVarForSelectPriv(pstate, lvar);
330 1660 : markVarForSelectPriv(pstate, rvar);
331 :
332 : /* Now create the lvar = rvar join condition */
333 1660 : e = makeSimpleA_Expr(AEXPR_OP, "=",
334 1660 : (Node *) copyObject(lvar), (Node *) copyObject(rvar),
335 : -1);
336 :
337 : /* Prepare to combine into an AND clause, if multiple join columns */
338 1660 : andargs = lappend(andargs, e);
339 : }
340 :
341 : /* Only need an AND if there's more than one join column */
342 1460 : if (list_length(andargs) == 1)
343 1282 : result = (Node *) linitial(andargs);
344 : else
345 178 : result = (Node *) makeBoolExpr(AND_EXPR, andargs, -1);
346 :
347 : /*
348 : * Since the references are already Vars, and are certainly from the input
349 : * relations, we don't have to go through the same pushups that
350 : * transformJoinOnClause() does. Just invoke transformExpr() to fix up
351 : * the operators, and we're done.
352 : */
353 1460 : result = transformExpr(pstate, result, EXPR_KIND_JOIN_USING);
354 :
355 1460 : result = coerce_to_boolean(pstate, result, "JOIN/USING");
356 :
357 1460 : return result;
358 : }
359 :
360 : /* transformJoinOnClause()
361 : * Transform the qual conditions for JOIN/ON.
362 : * Result is a transformed qualification expression.
363 : */
364 : static Node *
365 73768 : transformJoinOnClause(ParseState *pstate, JoinExpr *j, List *namespace)
366 : {
367 : Node *result;
368 : List *save_namespace;
369 :
370 : /*
371 : * The namespace that the join expression should see is just the two
372 : * subtrees of the JOIN plus any outer references from upper pstate
373 : * levels. Temporarily set this pstate's namespace accordingly. (We need
374 : * not check for refname conflicts, because transformFromClauseItem()
375 : * already did.) All namespace items are marked visible regardless of
376 : * LATERAL state.
377 : */
378 73768 : setNamespaceLateralState(namespace, false, true);
379 :
380 73768 : save_namespace = pstate->p_namespace;
381 73768 : pstate->p_namespace = namespace;
382 :
383 73768 : result = transformWhereClause(pstate, j->quals,
384 : EXPR_KIND_JOIN_ON, "JOIN/ON");
385 :
386 73750 : pstate->p_namespace = save_namespace;
387 :
388 73750 : return result;
389 : }
390 :
391 : /*
392 : * transformTableEntry --- transform a RangeVar (simple relation reference)
393 : */
394 : static ParseNamespaceItem *
395 354094 : transformTableEntry(ParseState *pstate, RangeVar *r)
396 : {
397 : /* addRangeTableEntry does all the work */
398 354094 : return addRangeTableEntry(pstate, r, r->alias, r->inh, true);
399 : }
400 :
401 : /*
402 : * transformRangeSubselect --- transform a sub-SELECT appearing in FROM
403 : */
404 : static ParseNamespaceItem *
405 14728 : transformRangeSubselect(ParseState *pstate, RangeSubselect *r)
406 : {
407 : Query *query;
408 :
409 : /*
410 : * Set p_expr_kind to show this parse level is recursing to a subselect.
411 : * We can't be nested within any expression, so don't need save-restore
412 : * logic here.
413 : */
414 : Assert(pstate->p_expr_kind == EXPR_KIND_NONE);
415 14728 : pstate->p_expr_kind = EXPR_KIND_FROM_SUBSELECT;
416 :
417 : /*
418 : * If the subselect is LATERAL, make lateral_only names of this level
419 : * visible to it. (LATERAL can't nest within a single pstate level, so we
420 : * don't need save/restore logic here.)
421 : */
422 : Assert(!pstate->p_lateral_active);
423 14728 : pstate->p_lateral_active = r->lateral;
424 :
425 : /*
426 : * Analyze and transform the subquery. Note that if the subquery doesn't
427 : * have an alias, it can't be explicitly selected for locking, but locking
428 : * might still be required (if there is an all-tables locking clause).
429 : */
430 14728 : query = parse_sub_analyze(r->subquery, pstate, NULL,
431 14728 : isLockedRefname(pstate,
432 14728 : r->alias == NULL ? NULL :
433 14572 : r->alias->aliasname),
434 : true);
435 :
436 : /* Restore state */
437 14620 : pstate->p_lateral_active = false;
438 14620 : pstate->p_expr_kind = EXPR_KIND_NONE;
439 :
440 : /*
441 : * Check that we got a SELECT. Anything else should be impossible given
442 : * restrictions of the grammar, but check anyway.
443 : */
444 14620 : if (!IsA(query, Query) ||
445 14620 : query->commandType != CMD_SELECT)
446 0 : elog(ERROR, "unexpected non-SELECT command in subquery in FROM");
447 :
448 : /*
449 : * OK, build an RTE and nsitem for the subquery.
450 : */
451 29234 : return addRangeTableEntryForSubquery(pstate,
452 : query,
453 : r->alias,
454 14620 : r->lateral,
455 : true);
456 : }
457 :
458 :
459 : /*
460 : * transformRangeFunction --- transform a function call appearing in FROM
461 : */
462 : static ParseNamespaceItem *
463 43366 : transformRangeFunction(ParseState *pstate, RangeFunction *r)
464 : {
465 43366 : List *funcexprs = NIL;
466 43366 : List *funcnames = NIL;
467 43366 : List *coldeflists = NIL;
468 : bool is_lateral;
469 : ListCell *lc;
470 :
471 : /*
472 : * We make lateral_only names of this level visible, whether or not the
473 : * RangeFunction is explicitly marked LATERAL. This is needed for SQL
474 : * spec compliance in the case of UNNEST(), and seems useful on
475 : * convenience grounds for all functions in FROM.
476 : *
477 : * (LATERAL can't nest within a single pstate level, so we don't need
478 : * save/restore logic here.)
479 : */
480 : Assert(!pstate->p_lateral_active);
481 43366 : pstate->p_lateral_active = true;
482 :
483 : /*
484 : * Transform the raw expressions.
485 : *
486 : * While transforming, also save function names for possible use as alias
487 : * and column names. We use the same transformation rules as for a SELECT
488 : * output expression. For a FuncCall node, the result will be the
489 : * function name, but it is possible for the grammar to hand back other
490 : * node types.
491 : *
492 : * We have to get this info now, because FigureColname only works on raw
493 : * parsetrees. Actually deciding what to do with the names is left up to
494 : * addRangeTableEntryForFunction.
495 : *
496 : * Likewise, collect column definition lists if there were any. But
497 : * complain if we find one here and the RangeFunction has one too.
498 : */
499 86750 : foreach(lc, r->functions)
500 : {
501 43552 : List *pair = (List *) lfirst(lc);
502 : Node *fexpr;
503 : List *coldeflist;
504 : Node *newfexpr;
505 : Node *last_srf;
506 :
507 : /* Disassemble the function-call/column-def-list pairs */
508 : Assert(list_length(pair) == 2);
509 43552 : fexpr = (Node *) linitial(pair);
510 43552 : coldeflist = (List *) lsecond(pair);
511 :
512 : /*
513 : * If we find a function call unnest() with more than one argument and
514 : * no special decoration, transform it into separate unnest() calls on
515 : * each argument. This is a kluge, for sure, but it's less nasty than
516 : * other ways of implementing the SQL-standard UNNEST() syntax.
517 : *
518 : * If there is any decoration (including a coldeflist), we don't
519 : * transform, which probably means a no-such-function error later. We
520 : * could alternatively throw an error right now, but that doesn't seem
521 : * tremendously helpful. If someone is using any such decoration,
522 : * then they're not using the SQL-standard syntax, and they're more
523 : * likely expecting an un-tweaked function call.
524 : *
525 : * Note: the transformation changes a non-schema-qualified unnest()
526 : * function name into schema-qualified pg_catalog.unnest(). This
527 : * choice is also a bit debatable, but it seems reasonable to force
528 : * use of built-in unnest() when we make this transformation.
529 : */
530 43552 : if (IsA(fexpr, FuncCall))
531 : {
532 43408 : FuncCall *fc = (FuncCall *) fexpr;
533 :
534 43408 : if (list_length(fc->funcname) == 1 &&
535 32178 : strcmp(strVal(linitial(fc->funcname)), "unnest") == 0 &&
536 2506 : list_length(fc->args) > 1 &&
537 66 : fc->agg_order == NIL &&
538 66 : fc->agg_filter == NULL &&
539 66 : fc->over == NULL &&
540 66 : !fc->agg_star &&
541 66 : !fc->agg_distinct &&
542 66 : !fc->func_variadic &&
543 : coldeflist == NIL)
544 : {
545 : ListCell *lc2;
546 :
547 240 : foreach(lc2, fc->args)
548 : {
549 174 : Node *arg = (Node *) lfirst(lc2);
550 : FuncCall *newfc;
551 :
552 174 : last_srf = pstate->p_last_srf;
553 :
554 174 : newfc = makeFuncCall(SystemFuncName("unnest"),
555 174 : list_make1(arg),
556 : COERCE_EXPLICIT_CALL,
557 : fc->location);
558 :
559 174 : newfexpr = transformExpr(pstate, (Node *) newfc,
560 : EXPR_KIND_FROM_FUNCTION);
561 :
562 : /* nodeFunctionscan.c requires SRFs to be at top level */
563 174 : if (pstate->p_last_srf != last_srf &&
564 174 : pstate->p_last_srf != newfexpr)
565 0 : ereport(ERROR,
566 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
567 : errmsg("set-returning functions must appear at top level of FROM"),
568 : parser_errposition(pstate,
569 : exprLocation(pstate->p_last_srf))));
570 :
571 174 : funcexprs = lappend(funcexprs, newfexpr);
572 :
573 174 : funcnames = lappend(funcnames,
574 174 : FigureColname((Node *) newfc));
575 :
576 : /* coldeflist is empty, so no error is possible */
577 :
578 174 : coldeflists = lappend(coldeflists, coldeflist);
579 : }
580 66 : continue; /* done with this function item */
581 : }
582 : }
583 :
584 : /* normal case ... */
585 43486 : last_srf = pstate->p_last_srf;
586 :
587 43486 : newfexpr = transformExpr(pstate, fexpr,
588 : EXPR_KIND_FROM_FUNCTION);
589 :
590 : /* nodeFunctionscan.c requires SRFs to be at top level */
591 43324 : if (pstate->p_last_srf != last_srf &&
592 37922 : pstate->p_last_srf != newfexpr)
593 6 : ereport(ERROR,
594 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
595 : errmsg("set-returning functions must appear at top level of FROM"),
596 : parser_errposition(pstate,
597 : exprLocation(pstate->p_last_srf))));
598 :
599 43318 : funcexprs = lappend(funcexprs, newfexpr);
600 :
601 43318 : funcnames = lappend(funcnames,
602 43318 : FigureColname(fexpr));
603 :
604 43318 : if (coldeflist && r->coldeflist)
605 0 : ereport(ERROR,
606 : (errcode(ERRCODE_SYNTAX_ERROR),
607 : errmsg("multiple column definition lists are not allowed for the same function"),
608 : parser_errposition(pstate,
609 : exprLocation((Node *) r->coldeflist))));
610 :
611 43318 : coldeflists = lappend(coldeflists, coldeflist);
612 : }
613 :
614 43198 : pstate->p_lateral_active = false;
615 :
616 : /*
617 : * We must assign collations now so that the RTE exposes correct collation
618 : * info for Vars created from it.
619 : */
620 43198 : assign_list_collations(pstate, funcexprs);
621 :
622 : /*
623 : * Install the top-level coldeflist if there was one (we already checked
624 : * that there was no conflicting per-function coldeflist).
625 : *
626 : * We only allow this when there's a single function (even after UNNEST
627 : * expansion) and no WITH ORDINALITY. The reason for the latter
628 : * restriction is that it's not real clear whether the ordinality column
629 : * should be in the coldeflist, and users are too likely to make mistakes
630 : * in one direction or the other. Putting the coldeflist inside ROWS
631 : * FROM() is much clearer in this case.
632 : */
633 43198 : if (r->coldeflist)
634 : {
635 748 : if (list_length(funcexprs) != 1)
636 : {
637 0 : if (r->is_rowsfrom)
638 0 : ereport(ERROR,
639 : (errcode(ERRCODE_SYNTAX_ERROR),
640 : errmsg("ROWS FROM() with multiple functions cannot have a column definition list"),
641 : errhint("Put a separate column definition list for each function inside ROWS FROM()."),
642 : parser_errposition(pstate,
643 : exprLocation((Node *) r->coldeflist))));
644 : else
645 0 : ereport(ERROR,
646 : (errcode(ERRCODE_SYNTAX_ERROR),
647 : errmsg("UNNEST() with multiple arguments cannot have a column definition list"),
648 : errhint("Use separate UNNEST() calls inside ROWS FROM(), and attach a column definition list to each one."),
649 : parser_errposition(pstate,
650 : exprLocation((Node *) r->coldeflist))));
651 : }
652 748 : if (r->ordinality)
653 0 : ereport(ERROR,
654 : (errcode(ERRCODE_SYNTAX_ERROR),
655 : errmsg("WITH ORDINALITY cannot be used with a column definition list"),
656 : errhint("Put the column definition list inside ROWS FROM()."),
657 : parser_errposition(pstate,
658 : exprLocation((Node *) r->coldeflist))));
659 :
660 748 : coldeflists = list_make1(r->coldeflist);
661 : }
662 :
663 : /*
664 : * Mark the RTE as LATERAL if the user said LATERAL explicitly, or if
665 : * there are any lateral cross-references in it.
666 : */
667 43198 : is_lateral = r->lateral || contain_vars_of_level((Node *) funcexprs, 0);
668 :
669 : /*
670 : * OK, build an RTE and nsitem for the function.
671 : */
672 43198 : return addRangeTableEntryForFunction(pstate,
673 : funcnames, funcexprs, coldeflists,
674 : r, is_lateral, true);
675 : }
676 :
677 : /*
678 : * transformRangeTableFunc -
679 : * Transform a raw RangeTableFunc into TableFunc.
680 : *
681 : * Transform the namespace clauses, the document-generating expression, the
682 : * row-generating expression, the column-generating expressions, and the
683 : * default value expressions.
684 : */
685 : static ParseNamespaceItem *
686 220 : transformRangeTableFunc(ParseState *pstate, RangeTableFunc *rtf)
687 : {
688 220 : TableFunc *tf = makeNode(TableFunc);
689 : const char *constructName;
690 : Oid docType;
691 : bool is_lateral;
692 : ListCell *col;
693 : char **names;
694 : int colno;
695 :
696 : /*
697 : * Currently we only support XMLTABLE here. See transformJsonTable() for
698 : * JSON_TABLE support.
699 : */
700 220 : tf->functype = TFT_XMLTABLE;
701 220 : constructName = "XMLTABLE";
702 220 : docType = XMLOID;
703 :
704 : /*
705 : * We make lateral_only names of this level visible, whether or not the
706 : * RangeTableFunc is explicitly marked LATERAL. This is needed for SQL
707 : * spec compliance and seems useful on convenience grounds for all
708 : * functions in FROM.
709 : *
710 : * (LATERAL can't nest within a single pstate level, so we don't need
711 : * save/restore logic here.)
712 : */
713 : Assert(!pstate->p_lateral_active);
714 220 : pstate->p_lateral_active = true;
715 :
716 : /* Transform and apply typecast to the row-generating expression ... */
717 : Assert(rtf->rowexpr != NULL);
718 220 : tf->rowexpr = coerce_to_specific_type(pstate,
719 : transformExpr(pstate, rtf->rowexpr, EXPR_KIND_FROM_FUNCTION),
720 : TEXTOID,
721 : constructName);
722 220 : assign_expr_collations(pstate, tf->rowexpr);
723 :
724 : /* ... and to the document itself */
725 : Assert(rtf->docexpr != NULL);
726 220 : tf->docexpr = coerce_to_specific_type(pstate,
727 : transformExpr(pstate, rtf->docexpr, EXPR_KIND_FROM_FUNCTION),
728 : docType,
729 : constructName);
730 220 : assign_expr_collations(pstate, tf->docexpr);
731 :
732 : /* undef ordinality column number */
733 220 : tf->ordinalitycol = -1;
734 :
735 : /* Process column specs */
736 220 : names = palloc(sizeof(char *) * list_length(rtf->columns));
737 :
738 220 : colno = 0;
739 970 : foreach(col, rtf->columns)
740 : {
741 750 : RangeTableFuncCol *rawc = (RangeTableFuncCol *) lfirst(col);
742 : Oid typid;
743 : int32 typmod;
744 : Node *colexpr;
745 : Node *coldefexpr;
746 : int j;
747 :
748 750 : tf->colnames = lappend(tf->colnames,
749 750 : makeString(pstrdup(rawc->colname)));
750 :
751 : /*
752 : * Determine the type and typmod for the new column. FOR ORDINALITY
753 : * columns are INTEGER per spec; the others are user-specified.
754 : */
755 750 : if (rawc->for_ordinality)
756 : {
757 62 : if (tf->ordinalitycol != -1)
758 0 : ereport(ERROR,
759 : (errcode(ERRCODE_SYNTAX_ERROR),
760 : errmsg("only one FOR ORDINALITY column is allowed"),
761 : parser_errposition(pstate, rawc->location)));
762 :
763 62 : typid = INT4OID;
764 62 : typmod = -1;
765 62 : tf->ordinalitycol = colno;
766 : }
767 : else
768 : {
769 688 : if (rawc->typeName->setof)
770 0 : ereport(ERROR,
771 : (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
772 : errmsg("column \"%s\" cannot be declared SETOF",
773 : rawc->colname),
774 : parser_errposition(pstate, rawc->location)));
775 :
776 688 : typenameTypeIdAndMod(pstate, rawc->typeName,
777 : &typid, &typmod);
778 : }
779 :
780 750 : tf->coltypes = lappend_oid(tf->coltypes, typid);
781 750 : tf->coltypmods = lappend_int(tf->coltypmods, typmod);
782 750 : tf->colcollations = lappend_oid(tf->colcollations,
783 : get_typcollation(typid));
784 :
785 : /* Transform the PATH and DEFAULT expressions */
786 750 : if (rawc->colexpr)
787 : {
788 490 : colexpr = coerce_to_specific_type(pstate,
789 : transformExpr(pstate, rawc->colexpr,
790 : EXPR_KIND_FROM_FUNCTION),
791 : TEXTOID,
792 : constructName);
793 490 : assign_expr_collations(pstate, colexpr);
794 : }
795 : else
796 260 : colexpr = NULL;
797 :
798 750 : if (rawc->coldefexpr)
799 : {
800 56 : coldefexpr = coerce_to_specific_type_typmod(pstate,
801 : transformExpr(pstate, rawc->coldefexpr,
802 : EXPR_KIND_FROM_FUNCTION),
803 : typid, typmod,
804 : constructName);
805 56 : assign_expr_collations(pstate, coldefexpr);
806 : }
807 : else
808 694 : coldefexpr = NULL;
809 :
810 750 : tf->colexprs = lappend(tf->colexprs, colexpr);
811 750 : tf->coldefexprs = lappend(tf->coldefexprs, coldefexpr);
812 :
813 750 : if (rawc->is_not_null)
814 56 : tf->notnulls = bms_add_member(tf->notnulls, colno);
815 :
816 : /* make sure column names are unique */
817 2534 : for (j = 0; j < colno; j++)
818 1784 : if (strcmp(names[j], rawc->colname) == 0)
819 0 : ereport(ERROR,
820 : (errcode(ERRCODE_SYNTAX_ERROR),
821 : errmsg("column name \"%s\" is not unique",
822 : rawc->colname),
823 : parser_errposition(pstate, rawc->location)));
824 750 : names[colno] = rawc->colname;
825 :
826 750 : colno++;
827 : }
828 220 : pfree(names);
829 :
830 : /* Namespaces, if any, also need to be transformed */
831 220 : if (rtf->namespaces != NIL)
832 : {
833 : ListCell *ns;
834 : ListCell *lc2;
835 20 : List *ns_uris = NIL;
836 20 : List *ns_names = NIL;
837 20 : bool default_ns_seen = false;
838 :
839 40 : foreach(ns, rtf->namespaces)
840 : {
841 20 : ResTarget *r = (ResTarget *) lfirst(ns);
842 : Node *ns_uri;
843 :
844 : Assert(IsA(r, ResTarget));
845 20 : ns_uri = transformExpr(pstate, r->val, EXPR_KIND_FROM_FUNCTION);
846 20 : ns_uri = coerce_to_specific_type(pstate, ns_uri,
847 : TEXTOID, constructName);
848 20 : assign_expr_collations(pstate, ns_uri);
849 20 : ns_uris = lappend(ns_uris, ns_uri);
850 :
851 : /* Verify consistency of name list: no dupes, only one DEFAULT */
852 20 : if (r->name != NULL)
853 : {
854 14 : foreach(lc2, ns_names)
855 : {
856 0 : String *ns_node = lfirst_node(String, lc2);
857 :
858 0 : if (ns_node == NULL)
859 0 : continue;
860 0 : if (strcmp(strVal(ns_node), r->name) == 0)
861 0 : ereport(ERROR,
862 : (errcode(ERRCODE_SYNTAX_ERROR),
863 : errmsg("namespace name \"%s\" is not unique",
864 : r->name),
865 : parser_errposition(pstate, r->location)));
866 : }
867 : }
868 : else
869 : {
870 6 : if (default_ns_seen)
871 0 : ereport(ERROR,
872 : (errcode(ERRCODE_SYNTAX_ERROR),
873 : errmsg("only one default namespace is allowed"),
874 : parser_errposition(pstate, r->location)));
875 6 : default_ns_seen = true;
876 : }
877 :
878 : /* We represent DEFAULT by a null pointer */
879 20 : ns_names = lappend(ns_names,
880 20 : r->name ? makeString(r->name) : NULL);
881 : }
882 :
883 20 : tf->ns_uris = ns_uris;
884 20 : tf->ns_names = ns_names;
885 : }
886 :
887 220 : tf->location = rtf->location;
888 :
889 220 : pstate->p_lateral_active = false;
890 :
891 : /*
892 : * Mark the RTE as LATERAL if the user said LATERAL explicitly, or if
893 : * there are any lateral cross-references in it.
894 : */
895 220 : is_lateral = rtf->lateral || contain_vars_of_level((Node *) tf, 0);
896 :
897 220 : return addRangeTableEntryForTableFunc(pstate,
898 : tf, rtf->alias, is_lateral, true);
899 : }
900 :
901 : /*
902 : * transformRangeTableSample --- transform a TABLESAMPLE clause
903 : *
904 : * Caller has already transformed rts->relation, we just have to validate
905 : * the remaining fields and create a TableSampleClause node.
906 : */
907 : static TableSampleClause *
908 242 : transformRangeTableSample(ParseState *pstate, RangeTableSample *rts)
909 : {
910 : TableSampleClause *tablesample;
911 : Oid handlerOid;
912 : Oid funcargtypes[1];
913 : TsmRoutine *tsm;
914 : List *fargs;
915 : ListCell *larg,
916 : *ltyp;
917 :
918 : /*
919 : * To validate the sample method name, look up the handler function, which
920 : * has the same name, one dummy INTERNAL argument, and a result type of
921 : * tsm_handler. (Note: tablesample method names are not schema-qualified
922 : * in the SQL standard; but since they are just functions to us, we allow
923 : * schema qualification to resolve any potential ambiguity.)
924 : */
925 242 : funcargtypes[0] = INTERNALOID;
926 :
927 242 : handlerOid = LookupFuncName(rts->method, 1, funcargtypes, true);
928 :
929 : /* we want error to complain about no-such-method, not no-such-function */
930 242 : if (!OidIsValid(handlerOid))
931 6 : ereport(ERROR,
932 : (errcode(ERRCODE_UNDEFINED_OBJECT),
933 : errmsg("tablesample method %s does not exist",
934 : NameListToString(rts->method)),
935 : parser_errposition(pstate, rts->location)));
936 :
937 : /* check that handler has correct return type */
938 236 : if (get_func_rettype(handlerOid) != TSM_HANDLEROID)
939 0 : ereport(ERROR,
940 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
941 : errmsg("function %s must return type %s",
942 : NameListToString(rts->method), "tsm_handler"),
943 : parser_errposition(pstate, rts->location)));
944 :
945 : /* OK, run the handler to get TsmRoutine, for argument type info */
946 236 : tsm = GetTsmRoutine(handlerOid);
947 :
948 236 : tablesample = makeNode(TableSampleClause);
949 236 : tablesample->tsmhandler = handlerOid;
950 :
951 : /* check user provided the expected number of arguments */
952 236 : if (list_length(rts->args) != list_length(tsm->parameterTypes))
953 0 : ereport(ERROR,
954 : (errcode(ERRCODE_INVALID_TABLESAMPLE_ARGUMENT),
955 : errmsg_plural("tablesample method %s requires %d argument, not %d",
956 : "tablesample method %s requires %d arguments, not %d",
957 : list_length(tsm->parameterTypes),
958 : NameListToString(rts->method),
959 : list_length(tsm->parameterTypes),
960 : list_length(rts->args)),
961 : parser_errposition(pstate, rts->location)));
962 :
963 : /*
964 : * Transform the arguments, typecasting them as needed. Note we must also
965 : * assign collations now, because assign_query_collations() doesn't
966 : * examine any substructure of RTEs.
967 : */
968 236 : fargs = NIL;
969 472 : forboth(larg, rts->args, ltyp, tsm->parameterTypes)
970 : {
971 236 : Node *arg = (Node *) lfirst(larg);
972 236 : Oid argtype = lfirst_oid(ltyp);
973 :
974 236 : arg = transformExpr(pstate, arg, EXPR_KIND_FROM_FUNCTION);
975 236 : arg = coerce_to_specific_type(pstate, arg, argtype, "TABLESAMPLE");
976 236 : assign_expr_collations(pstate, arg);
977 236 : fargs = lappend(fargs, arg);
978 : }
979 236 : tablesample->args = fargs;
980 :
981 : /* Process REPEATABLE (seed) */
982 236 : if (rts->repeatable != NULL)
983 : {
984 : Node *arg;
985 :
986 102 : if (!tsm->repeatable_across_queries)
987 4 : ereport(ERROR,
988 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
989 : errmsg("tablesample method %s does not support REPEATABLE",
990 : NameListToString(rts->method)),
991 : parser_errposition(pstate, rts->location)));
992 :
993 98 : arg = transformExpr(pstate, rts->repeatable, EXPR_KIND_FROM_FUNCTION);
994 98 : arg = coerce_to_specific_type(pstate, arg, FLOAT8OID, "REPEATABLE");
995 98 : assign_expr_collations(pstate, arg);
996 98 : tablesample->repeatable = (Expr *) arg;
997 : }
998 : else
999 134 : tablesample->repeatable = NULL;
1000 :
1001 232 : return tablesample;
1002 : }
1003 :
1004 : /*
1005 : * getNSItemForSpecialRelationTypes
1006 : *
1007 : * If given RangeVar refers to a CTE or an EphemeralNamedRelation,
1008 : * build and return an appropriate ParseNamespaceItem, otherwise return NULL
1009 : */
1010 : static ParseNamespaceItem *
1011 360220 : getNSItemForSpecialRelationTypes(ParseState *pstate, RangeVar *rv)
1012 : {
1013 : ParseNamespaceItem *nsitem;
1014 : CommonTableExpr *cte;
1015 : Index levelsup;
1016 :
1017 : /*
1018 : * if it is a qualified name, it can't be a CTE or tuplestore reference
1019 : */
1020 360220 : if (rv->schemaname)
1021 179214 : return NULL;
1022 :
1023 181006 : cte = scanNameSpaceForCTE(pstate, rv->relname, &levelsup);
1024 181006 : if (cte)
1025 5676 : nsitem = addRangeTableEntryForCTE(pstate, cte, levelsup, rv, true);
1026 175330 : else if (scanNameSpaceForENR(pstate, rv->relname))
1027 450 : nsitem = addRangeTableEntryForENR(pstate, rv, true);
1028 : else
1029 174880 : nsitem = NULL;
1030 :
1031 180994 : return nsitem;
1032 : }
1033 :
1034 : /*
1035 : * transformFromClauseItem -
1036 : * Transform a FROM-clause item, adding any required entries to the
1037 : * range table list being built in the ParseState, and return the
1038 : * transformed item ready to include in the joinlist. Also build a
1039 : * ParseNamespaceItem list describing the names exposed by this item.
1040 : * This routine can recurse to handle SQL92 JOIN expressions.
1041 : *
1042 : * The function return value is the node to add to the jointree (a
1043 : * RangeTblRef or JoinExpr). Additional output parameters are:
1044 : *
1045 : * *top_nsitem: receives the ParseNamespaceItem directly corresponding to the
1046 : * jointree item. (This is only used during internal recursion, not by
1047 : * outside callers.)
1048 : *
1049 : * *namespace: receives a List of ParseNamespaceItems for the RTEs exposed
1050 : * as table/column names by this item. (The lateral_only flags in these items
1051 : * are indeterminate and should be explicitly set by the caller before use.)
1052 : */
1053 : static Node *
1054 494886 : transformFromClauseItem(ParseState *pstate, Node *n,
1055 : ParseNamespaceItem **top_nsitem,
1056 : List **namespace)
1057 : {
1058 : /* Guard against stack overflow due to overly deep subtree */
1059 494886 : check_stack_depth();
1060 :
1061 494886 : if (IsA(n, RangeVar))
1062 : {
1063 : /* Plain relation reference, or perhaps a CTE reference */
1064 360220 : RangeVar *rv = (RangeVar *) n;
1065 : RangeTblRef *rtr;
1066 : ParseNamespaceItem *nsitem;
1067 :
1068 : /* Check if it's a CTE or tuplestore reference */
1069 360220 : nsitem = getNSItemForSpecialRelationTypes(pstate, rv);
1070 :
1071 : /* if not found above, must be a table reference */
1072 360208 : if (!nsitem)
1073 354094 : nsitem = transformTableEntry(pstate, rv);
1074 :
1075 360028 : *top_nsitem = nsitem;
1076 360028 : *namespace = list_make1(nsitem);
1077 360028 : rtr = makeNode(RangeTblRef);
1078 360028 : rtr->rtindex = nsitem->p_rtindex;
1079 360028 : return (Node *) rtr;
1080 : }
1081 134666 : else if (IsA(n, RangeSubselect))
1082 : {
1083 : /* sub-SELECT is like a plain relation */
1084 : RangeTblRef *rtr;
1085 : ParseNamespaceItem *nsitem;
1086 :
1087 14728 : nsitem = transformRangeSubselect(pstate, (RangeSubselect *) n);
1088 14614 : *top_nsitem = nsitem;
1089 14614 : *namespace = list_make1(nsitem);
1090 14614 : rtr = makeNode(RangeTblRef);
1091 14614 : rtr->rtindex = nsitem->p_rtindex;
1092 14614 : return (Node *) rtr;
1093 : }
1094 119938 : else if (IsA(n, RangeFunction))
1095 : {
1096 : /* function is like a plain relation */
1097 : RangeTblRef *rtr;
1098 : ParseNamespaceItem *nsitem;
1099 :
1100 43366 : nsitem = transformRangeFunction(pstate, (RangeFunction *) n);
1101 43144 : *top_nsitem = nsitem;
1102 43144 : *namespace = list_make1(nsitem);
1103 43144 : rtr = makeNode(RangeTblRef);
1104 43144 : rtr->rtindex = nsitem->p_rtindex;
1105 43144 : return (Node *) rtr;
1106 : }
1107 76572 : else if (IsA(n, RangeTableFunc) || IsA(n, JsonTable))
1108 : {
1109 : /* table function is like a plain relation */
1110 : RangeTblRef *rtr;
1111 : ParseNamespaceItem *nsitem;
1112 :
1113 744 : if (IsA(n, JsonTable))
1114 524 : nsitem = transformJsonTable(pstate, (JsonTable *) n);
1115 : else
1116 220 : nsitem = transformRangeTableFunc(pstate, (RangeTableFunc *) n);
1117 :
1118 654 : *top_nsitem = nsitem;
1119 654 : *namespace = list_make1(nsitem);
1120 654 : rtr = makeNode(RangeTblRef);
1121 654 : rtr->rtindex = nsitem->p_rtindex;
1122 654 : return (Node *) rtr;
1123 : }
1124 75828 : else if (IsA(n, RangeTableSample))
1125 : {
1126 : /* TABLESAMPLE clause (wrapping some other valid FROM node) */
1127 254 : RangeTableSample *rts = (RangeTableSample *) n;
1128 : Node *rel;
1129 : RangeTblEntry *rte;
1130 :
1131 : /* Recursively transform the contained relation */
1132 254 : rel = transformFromClauseItem(pstate, rts->relation,
1133 : top_nsitem, namespace);
1134 254 : rte = (*top_nsitem)->p_rte;
1135 : /* We only support this on plain relations and matviews */
1136 254 : if (rte->rtekind != RTE_RELATION ||
1137 248 : (rte->relkind != RELKIND_RELATION &&
1138 24 : rte->relkind != RELKIND_MATVIEW &&
1139 24 : rte->relkind != RELKIND_PARTITIONED_TABLE))
1140 12 : ereport(ERROR,
1141 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1142 : errmsg("TABLESAMPLE clause can only be applied to tables and materialized views"),
1143 : parser_errposition(pstate, exprLocation(rts->relation))));
1144 :
1145 : /* Transform TABLESAMPLE details and attach to the RTE */
1146 242 : rte->tablesample = transformRangeTableSample(pstate, rts);
1147 232 : return rel;
1148 : }
1149 75574 : else if (IsA(n, JoinExpr))
1150 : {
1151 : /* A newfangled join expression */
1152 75574 : JoinExpr *j = (JoinExpr *) n;
1153 : ParseNamespaceItem *nsitem;
1154 : ParseNamespaceItem *l_nsitem;
1155 : ParseNamespaceItem *r_nsitem;
1156 : List *l_namespace,
1157 : *r_namespace,
1158 : *my_namespace,
1159 : *l_colnames,
1160 : *r_colnames,
1161 : *res_colnames,
1162 : *l_colnos,
1163 : *r_colnos,
1164 : *res_colvars;
1165 : ParseNamespaceColumn *l_nscolumns,
1166 : *r_nscolumns,
1167 : *res_nscolumns;
1168 : int res_colindex;
1169 : bool lateral_ok;
1170 : int sv_namespace_length;
1171 : int k;
1172 :
1173 : /*
1174 : * Recursively process the left subtree, then the right. We must do
1175 : * it in this order for correct visibility of LATERAL references.
1176 : */
1177 75574 : j->larg = transformFromClauseItem(pstate, j->larg,
1178 : &l_nsitem,
1179 : &l_namespace);
1180 :
1181 : /*
1182 : * Make the left-side RTEs available for LATERAL access within the
1183 : * right side, by temporarily adding them to the pstate's namespace
1184 : * list. Per SQL:2008, if the join type is not INNER or LEFT then the
1185 : * left-side names must still be exposed, but it's an error to
1186 : * reference them. (Stupid design, but that's what it says.) Hence,
1187 : * we always push them into the namespace, but mark them as not
1188 : * lateral_ok if the jointype is wrong.
1189 : *
1190 : * Notice that we don't require the merged namespace list to be
1191 : * conflict-free. See the comments for scanNameSpaceForRefname().
1192 : */
1193 75574 : lateral_ok = (j->jointype == JOIN_INNER || j->jointype == JOIN_LEFT);
1194 75574 : setNamespaceLateralState(l_namespace, true, lateral_ok);
1195 :
1196 75574 : sv_namespace_length = list_length(pstate->p_namespace);
1197 75574 : pstate->p_namespace = list_concat(pstate->p_namespace, l_namespace);
1198 :
1199 : /* And now we can process the RHS */
1200 75574 : j->rarg = transformFromClauseItem(pstate, j->rarg,
1201 : &r_nsitem,
1202 : &r_namespace);
1203 :
1204 : /* Remove the left-side RTEs from the namespace list again */
1205 75538 : pstate->p_namespace = list_truncate(pstate->p_namespace,
1206 : sv_namespace_length);
1207 :
1208 : /*
1209 : * Check for conflicting refnames in left and right subtrees. Must do
1210 : * this because higher levels will assume I hand back a self-
1211 : * consistent namespace list.
1212 : */
1213 75538 : checkNameSpaceConflicts(pstate, l_namespace, r_namespace);
1214 :
1215 : /*
1216 : * Generate combined namespace info for possible use below.
1217 : */
1218 75538 : my_namespace = list_concat(l_namespace, r_namespace);
1219 :
1220 : /*
1221 : * We'll work from the nscolumns data and eref alias column names for
1222 : * each of the input nsitems. Note that these include dropped
1223 : * columns, which is helpful because we can keep track of physical
1224 : * input column numbers more easily.
1225 : */
1226 75538 : l_nscolumns = l_nsitem->p_nscolumns;
1227 75538 : l_colnames = l_nsitem->p_names->colnames;
1228 75538 : r_nscolumns = r_nsitem->p_nscolumns;
1229 75538 : r_colnames = r_nsitem->p_names->colnames;
1230 :
1231 : /*
1232 : * Natural join does not explicitly specify columns; must generate
1233 : * columns to join. Need to run through the list of columns from each
1234 : * table or join result and match up the column names. Use the first
1235 : * table, and check every column in the second table for a match.
1236 : * (We'll check that the matches were unique later on.) The result of
1237 : * this step is a list of column names just like an explicitly-written
1238 : * USING list.
1239 : */
1240 75538 : if (j->isNatural)
1241 : {
1242 258 : List *rlist = NIL;
1243 : ListCell *lx,
1244 : *rx;
1245 :
1246 : Assert(j->usingClause == NIL); /* shouldn't have USING() too */
1247 :
1248 1140 : foreach(lx, l_colnames)
1249 : {
1250 882 : char *l_colname = strVal(lfirst(lx));
1251 882 : String *m_name = NULL;
1252 :
1253 882 : if (l_colname[0] == '\0')
1254 12 : continue; /* ignore dropped columns */
1255 :
1256 2412 : foreach(rx, r_colnames)
1257 : {
1258 1848 : char *r_colname = strVal(lfirst(rx));
1259 :
1260 1848 : if (strcmp(l_colname, r_colname) == 0)
1261 : {
1262 306 : m_name = makeString(l_colname);
1263 306 : break;
1264 : }
1265 : }
1266 :
1267 : /* matched a right column? then keep as join column... */
1268 870 : if (m_name != NULL)
1269 306 : rlist = lappend(rlist, m_name);
1270 : }
1271 :
1272 258 : j->usingClause = rlist;
1273 : }
1274 :
1275 : /*
1276 : * If a USING clause alias was specified, save the USING columns as
1277 : * its column list.
1278 : */
1279 75538 : if (j->join_using_alias)
1280 84 : j->join_using_alias->colnames = j->usingClause;
1281 :
1282 : /*
1283 : * Now transform the join qualifications, if any.
1284 : */
1285 75538 : l_colnos = NIL;
1286 75538 : r_colnos = NIL;
1287 75538 : res_colnames = NIL;
1288 75538 : res_colvars = NIL;
1289 :
1290 : /* this may be larger than needed, but it's not worth being exact */
1291 : res_nscolumns = (ParseNamespaceColumn *)
1292 75538 : palloc0((list_length(l_colnames) + list_length(r_colnames)) *
1293 : sizeof(ParseNamespaceColumn));
1294 75538 : res_colindex = 0;
1295 :
1296 75538 : if (j->usingClause)
1297 : {
1298 : /*
1299 : * JOIN/USING (or NATURAL JOIN, as transformed above). Transform
1300 : * the list into an explicit ON-condition.
1301 : */
1302 1460 : List *ucols = j->usingClause;
1303 1460 : List *l_usingvars = NIL;
1304 1460 : List *r_usingvars = NIL;
1305 : ListCell *ucol;
1306 :
1307 : Assert(j->quals == NULL); /* shouldn't have ON() too */
1308 :
1309 3120 : foreach(ucol, ucols)
1310 : {
1311 1660 : char *u_colname = strVal(lfirst(ucol));
1312 : ListCell *col;
1313 : int ndx;
1314 1660 : int l_index = -1;
1315 1660 : int r_index = -1;
1316 : Var *l_colvar,
1317 : *r_colvar;
1318 :
1319 : Assert(u_colname[0] != '\0');
1320 :
1321 : /* Check for USING(foo,foo) */
1322 1888 : foreach(col, res_colnames)
1323 : {
1324 228 : char *res_colname = strVal(lfirst(col));
1325 :
1326 228 : if (strcmp(res_colname, u_colname) == 0)
1327 0 : ereport(ERROR,
1328 : (errcode(ERRCODE_DUPLICATE_COLUMN),
1329 : errmsg("column name \"%s\" appears more than once in USING clause",
1330 : u_colname)));
1331 : }
1332 :
1333 : /* Find it in left input */
1334 1660 : ndx = 0;
1335 8134 : foreach(col, l_colnames)
1336 : {
1337 6474 : char *l_colname = strVal(lfirst(col));
1338 :
1339 6474 : if (strcmp(l_colname, u_colname) == 0)
1340 : {
1341 1660 : if (l_index >= 0)
1342 0 : ereport(ERROR,
1343 : (errcode(ERRCODE_AMBIGUOUS_COLUMN),
1344 : errmsg("common column name \"%s\" appears more than once in left table",
1345 : u_colname)));
1346 1660 : l_index = ndx;
1347 : }
1348 6474 : ndx++;
1349 : }
1350 1660 : if (l_index < 0)
1351 0 : ereport(ERROR,
1352 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1353 : errmsg("column \"%s\" specified in USING clause does not exist in left table",
1354 : u_colname)));
1355 1660 : l_colnos = lappend_int(l_colnos, l_index + 1);
1356 :
1357 : /* Find it in right input */
1358 1660 : ndx = 0;
1359 8034 : foreach(col, r_colnames)
1360 : {
1361 6374 : char *r_colname = strVal(lfirst(col));
1362 :
1363 6374 : if (strcmp(r_colname, u_colname) == 0)
1364 : {
1365 1660 : if (r_index >= 0)
1366 0 : ereport(ERROR,
1367 : (errcode(ERRCODE_AMBIGUOUS_COLUMN),
1368 : errmsg("common column name \"%s\" appears more than once in right table",
1369 : u_colname)));
1370 1660 : r_index = ndx;
1371 : }
1372 6374 : ndx++;
1373 : }
1374 1660 : if (r_index < 0)
1375 0 : ereport(ERROR,
1376 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1377 : errmsg("column \"%s\" specified in USING clause does not exist in right table",
1378 : u_colname)));
1379 1660 : r_colnos = lappend_int(r_colnos, r_index + 1);
1380 :
1381 : /* Build Vars to use in the generated JOIN ON clause */
1382 1660 : l_colvar = buildVarFromNSColumn(pstate, l_nscolumns + l_index);
1383 1660 : l_usingvars = lappend(l_usingvars, l_colvar);
1384 1660 : r_colvar = buildVarFromNSColumn(pstate, r_nscolumns + r_index);
1385 1660 : r_usingvars = lappend(r_usingvars, r_colvar);
1386 :
1387 : /*
1388 : * While we're here, add column names to the res_colnames
1389 : * list. It's a bit ugly to do this here while the
1390 : * corresponding res_colvars entries are not made till later,
1391 : * but doing this later would require an additional traversal
1392 : * of the usingClause list.
1393 : */
1394 1660 : res_colnames = lappend(res_colnames, lfirst(ucol));
1395 : }
1396 :
1397 : /* Construct the generated JOIN ON clause */
1398 1460 : j->quals = transformJoinUsingClause(pstate,
1399 : l_usingvars,
1400 : r_usingvars);
1401 : }
1402 74078 : else if (j->quals)
1403 : {
1404 : /* User-written ON-condition; transform it */
1405 73768 : j->quals = transformJoinOnClause(pstate, j, my_namespace);
1406 : }
1407 : else
1408 : {
1409 : /* CROSS JOIN: no quals */
1410 : }
1411 :
1412 : /*
1413 : * If this is an outer join, now mark the appropriate child RTEs as
1414 : * being nulled by this join. We have finished processing the child
1415 : * join expressions as well as the current join's quals, which deal in
1416 : * non-nulled input columns. All future references to those RTEs will
1417 : * see possibly-nulled values, and we should mark generated Vars to
1418 : * account for that. In particular, the join alias Vars that we're
1419 : * about to build should reflect the nulling effects of this join.
1420 : *
1421 : * A difficulty with doing this is that we need the join's RT index,
1422 : * which we don't officially have yet. However, no other RTE can get
1423 : * made between here and the addRangeTableEntryForJoin call, so we can
1424 : * predict what the assignment will be. (Alternatively, we could call
1425 : * addRangeTableEntryForJoin before we have all the data computed, but
1426 : * this seems less ugly.)
1427 : */
1428 75520 : j->rtindex = list_length(pstate->p_rtable) + 1;
1429 :
1430 75520 : switch (j->jointype)
1431 : {
1432 36034 : case JOIN_INNER:
1433 36034 : break;
1434 38140 : case JOIN_LEFT:
1435 38140 : markRelsAsNulledBy(pstate, j->rarg, j->rtindex);
1436 38140 : break;
1437 1008 : case JOIN_FULL:
1438 1008 : markRelsAsNulledBy(pstate, j->larg, j->rtindex);
1439 1008 : markRelsAsNulledBy(pstate, j->rarg, j->rtindex);
1440 1008 : break;
1441 338 : case JOIN_RIGHT:
1442 338 : markRelsAsNulledBy(pstate, j->larg, j->rtindex);
1443 338 : break;
1444 0 : default:
1445 : /* shouldn't see any other types here */
1446 0 : elog(ERROR, "unrecognized join type: %d",
1447 : (int) j->jointype);
1448 : break;
1449 : }
1450 :
1451 : /*
1452 : * Now we can construct join alias expressions for the USING columns.
1453 : */
1454 75520 : if (j->usingClause)
1455 : {
1456 : ListCell *lc1,
1457 : *lc2;
1458 :
1459 : /* Scan the colnos lists to recover info from the previous loop */
1460 3120 : forboth(lc1, l_colnos, lc2, r_colnos)
1461 : {
1462 1660 : int l_index = lfirst_int(lc1) - 1;
1463 1660 : int r_index = lfirst_int(lc2) - 1;
1464 : Var *l_colvar,
1465 : *r_colvar;
1466 : Node *u_colvar;
1467 : ParseNamespaceColumn *res_nscolumn;
1468 :
1469 : /*
1470 : * Note we re-build these Vars: they might have different
1471 : * varnullingrels than the ones made in the previous loop.
1472 : */
1473 1660 : l_colvar = buildVarFromNSColumn(pstate, l_nscolumns + l_index);
1474 1660 : r_colvar = buildVarFromNSColumn(pstate, r_nscolumns + r_index);
1475 :
1476 : /* Construct the join alias Var for this column */
1477 1660 : u_colvar = buildMergedJoinVar(pstate,
1478 : j->jointype,
1479 : l_colvar,
1480 : r_colvar);
1481 1660 : res_colvars = lappend(res_colvars, u_colvar);
1482 :
1483 : /* Construct column's res_nscolumns[] entry */
1484 1660 : res_nscolumn = res_nscolumns + res_colindex;
1485 1660 : res_colindex++;
1486 1660 : if (u_colvar == (Node *) l_colvar)
1487 : {
1488 : /* Merged column is equivalent to left input */
1489 1214 : *res_nscolumn = l_nscolumns[l_index];
1490 : }
1491 446 : else if (u_colvar == (Node *) r_colvar)
1492 : {
1493 : /* Merged column is equivalent to right input */
1494 42 : *res_nscolumn = r_nscolumns[r_index];
1495 : }
1496 : else
1497 : {
1498 : /*
1499 : * Merged column is not semantically equivalent to either
1500 : * input, so it needs to be referenced as the join output
1501 : * column.
1502 : */
1503 404 : res_nscolumn->p_varno = j->rtindex;
1504 404 : res_nscolumn->p_varattno = res_colindex;
1505 404 : res_nscolumn->p_vartype = exprType(u_colvar);
1506 404 : res_nscolumn->p_vartypmod = exprTypmod(u_colvar);
1507 404 : res_nscolumn->p_varcollid = exprCollation(u_colvar);
1508 404 : res_nscolumn->p_varnosyn = j->rtindex;
1509 404 : res_nscolumn->p_varattnosyn = res_colindex;
1510 : }
1511 : }
1512 : }
1513 :
1514 : /* Add remaining columns from each side to the output columns */
1515 75520 : res_colindex +=
1516 75520 : extractRemainingColumns(pstate,
1517 : l_nscolumns, l_colnames, &l_colnos,
1518 : &res_colnames, &res_colvars,
1519 75520 : res_nscolumns + res_colindex);
1520 75520 : res_colindex +=
1521 75520 : extractRemainingColumns(pstate,
1522 : r_nscolumns, r_colnames, &r_colnos,
1523 : &res_colnames, &res_colvars,
1524 75520 : res_nscolumns + res_colindex);
1525 :
1526 : /* If join has an alias, it syntactically hides all inputs */
1527 75520 : if (j->alias)
1528 : {
1529 978 : for (k = 0; k < res_colindex; k++)
1530 : {
1531 804 : ParseNamespaceColumn *nscol = res_nscolumns + k;
1532 :
1533 804 : nscol->p_varnosyn = j->rtindex;
1534 804 : nscol->p_varattnosyn = k + 1;
1535 : }
1536 : }
1537 :
1538 : /*
1539 : * Now build an RTE and nsitem for the result of the join.
1540 : */
1541 75520 : nsitem = addRangeTableEntryForJoin(pstate,
1542 : res_colnames,
1543 : res_nscolumns,
1544 : j->jointype,
1545 75520 : list_length(j->usingClause),
1546 : res_colvars,
1547 : l_colnos,
1548 : r_colnos,
1549 : j->join_using_alias,
1550 : j->alias,
1551 : true);
1552 :
1553 : /* Verify that we correctly predicted the join's RT index */
1554 : Assert(j->rtindex == nsitem->p_rtindex);
1555 : /* Cross-check number of columns, too */
1556 : Assert(res_colindex == list_length(nsitem->p_names->colnames));
1557 :
1558 : /*
1559 : * Save a link to the JoinExpr in the proper element of p_joinexprs.
1560 : * Since we maintain that list lazily, it may be necessary to fill in
1561 : * empty entries before we can add the JoinExpr in the right place.
1562 : */
1563 197536 : for (k = list_length(pstate->p_joinexprs) + 1; k < j->rtindex; k++)
1564 122022 : pstate->p_joinexprs = lappend(pstate->p_joinexprs, NULL);
1565 75514 : pstate->p_joinexprs = lappend(pstate->p_joinexprs, j);
1566 : Assert(list_length(pstate->p_joinexprs) == j->rtindex);
1567 :
1568 : /*
1569 : * If the join has a USING alias, build a ParseNamespaceItem for that
1570 : * and add it to the list of nsitems in the join's input.
1571 : */
1572 75514 : if (j->join_using_alias)
1573 : {
1574 : ParseNamespaceItem *jnsitem;
1575 :
1576 84 : jnsitem = (ParseNamespaceItem *) palloc(sizeof(ParseNamespaceItem));
1577 84 : jnsitem->p_names = j->join_using_alias;
1578 84 : jnsitem->p_rte = nsitem->p_rte;
1579 84 : jnsitem->p_rtindex = nsitem->p_rtindex;
1580 84 : jnsitem->p_perminfo = NULL;
1581 : /* no need to copy the first N columns, just use res_nscolumns */
1582 84 : jnsitem->p_nscolumns = res_nscolumns;
1583 : /* set default visibility flags; might get changed later */
1584 84 : jnsitem->p_rel_visible = true;
1585 84 : jnsitem->p_cols_visible = true;
1586 84 : jnsitem->p_lateral_only = false;
1587 84 : jnsitem->p_lateral_ok = true;
1588 84 : jnsitem->p_returning_type = VAR_RETURNING_DEFAULT;
1589 : /* Per SQL, we must check for alias conflicts */
1590 84 : checkNameSpaceConflicts(pstate, list_make1(jnsitem), my_namespace);
1591 78 : my_namespace = lappend(my_namespace, jnsitem);
1592 : }
1593 :
1594 : /*
1595 : * Prepare returned namespace list. If the JOIN has an alias then it
1596 : * hides the contained RTEs completely; otherwise, the contained RTEs
1597 : * are still visible as table names, but are not visible for
1598 : * unqualified column-name access.
1599 : *
1600 : * Note: if there are nested alias-less JOINs, the lower-level ones
1601 : * will remain in the list although they have neither p_rel_visible
1602 : * nor p_cols_visible set. We could delete such list items, but it's
1603 : * unclear that it's worth expending cycles to do so.
1604 : */
1605 75508 : if (j->alias != NULL)
1606 168 : my_namespace = NIL;
1607 : else
1608 75340 : setNamespaceColumnVisibility(my_namespace, false);
1609 :
1610 : /*
1611 : * The join RTE itself is always made visible for unqualified column
1612 : * names. It's visible as a relation name only if it has an alias.
1613 : */
1614 75508 : nsitem->p_rel_visible = (j->alias != NULL);
1615 75508 : nsitem->p_cols_visible = true;
1616 75508 : nsitem->p_lateral_only = false;
1617 75508 : nsitem->p_lateral_ok = true;
1618 :
1619 75508 : *top_nsitem = nsitem;
1620 75508 : *namespace = lappend(my_namespace, nsitem);
1621 :
1622 75508 : return (Node *) j;
1623 : }
1624 : else
1625 0 : elog(ERROR, "unrecognized node type: %d", (int) nodeTag(n));
1626 : return NULL; /* can't get here, keep compiler quiet */
1627 : }
1628 :
1629 : /*
1630 : * buildVarFromNSColumn -
1631 : * build a Var node using ParseNamespaceColumn data
1632 : *
1633 : * This is used to construct joinaliasvars entries.
1634 : * We can assume varlevelsup should be 0, and no location is specified.
1635 : * Note also that no column SELECT privilege is requested here; that would
1636 : * happen only if the column is actually referenced in the query.
1637 : */
1638 : static Var *
1639 2839122 : buildVarFromNSColumn(ParseState *pstate, ParseNamespaceColumn *nscol)
1640 : {
1641 : Var *var;
1642 :
1643 : Assert(nscol->p_varno > 0); /* i.e., not deleted column */
1644 2839122 : var = makeVar(nscol->p_varno,
1645 2839122 : nscol->p_varattno,
1646 : nscol->p_vartype,
1647 : nscol->p_vartypmod,
1648 : nscol->p_varcollid,
1649 : 0);
1650 : /* makeVar doesn't offer parameters for these, so set by hand: */
1651 2839122 : var->varreturningtype = nscol->p_varreturningtype;
1652 2839122 : var->varnosyn = nscol->p_varnosyn;
1653 2839122 : var->varattnosyn = nscol->p_varattnosyn;
1654 :
1655 : /* ... and update varnullingrels */
1656 2839122 : markNullableIfNeeded(pstate, var);
1657 :
1658 2839122 : return var;
1659 : }
1660 :
1661 : /*
1662 : * buildMergedJoinVar -
1663 : * generate a suitable replacement expression for a merged join column
1664 : */
1665 : static Node *
1666 1660 : 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 1660 : outcoltype = select_common_type(pstate,
1676 1660 : list_make2(l_colvar, r_colvar),
1677 : "JOIN/USING",
1678 : NULL);
1679 1660 : outcoltypmod = select_common_typmod(pstate,
1680 1660 : 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 1660 : 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 1576 : 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 1576 : l_node = (Node *) l_colvar;
1700 :
1701 1660 : 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 1630 : 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 1630 : r_node = (Node *) r_colvar;
1712 :
1713 : /*
1714 : * Choose what to emit
1715 : */
1716 1660 : 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 350 : 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 350 : CoalesceExpr *c = makeNode(CoalesceExpr);
1745 :
1746 350 : c->coalescetype = outcoltype;
1747 : /* coalescecollid will get set below */
1748 350 : c->args = list_make2(l_node, r_node);
1749 350 : c->location = -1;
1750 350 : res_node = (Node *) c;
1751 350 : 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 1660 : assign_expr_collations(pstate, res_node);
1765 :
1766 1660 : 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 43810 : markRelsAsNulledBy(ParseState *pstate, Node *n, int jindex)
1775 : {
1776 : int varno;
1777 : ListCell *lc;
1778 :
1779 : /* Note: we can't see FromExpr here */
1780 43810 : if (IsA(n, RangeTblRef))
1781 : {
1782 42152 : varno = ((RangeTblRef *) n)->rtindex;
1783 : }
1784 1658 : else if (IsA(n, JoinExpr))
1785 : {
1786 1658 : JoinExpr *j = (JoinExpr *) n;
1787 :
1788 : /* recurse to children */
1789 1658 : markRelsAsNulledBy(pstate, j->larg, jindex);
1790 1658 : markRelsAsNulledBy(pstate, j->rarg, jindex);
1791 1658 : 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 142434 : while (list_length(pstate->p_nullingrels) < varno)
1805 98624 : pstate->p_nullingrels = lappend(pstate->p_nullingrels, NULL);
1806 43810 : lc = list_nth_cell(pstate->p_nullingrels, varno - 1);
1807 43810 : lfirst(lc) = bms_add_member((Bitmapset *) lfirst(lc), jindex);
1808 43810 : }
1809 :
1810 : /*
1811 : * setNamespaceColumnVisibility -
1812 : * Convenience subroutine to update cols_visible flags in a namespace list.
1813 : */
1814 : static void
1815 75340 : setNamespaceColumnVisibility(List *namespace, bool cols_visible)
1816 : {
1817 : ListCell *lc;
1818 :
1819 319946 : foreach(lc, namespace)
1820 : {
1821 244606 : ParseNamespaceItem *nsitem = (ParseNamespaceItem *) lfirst(lc);
1822 :
1823 244606 : nsitem->p_cols_visible = cols_visible;
1824 : }
1825 75340 : }
1826 :
1827 : /*
1828 : * setNamespaceLateralState -
1829 : * Convenience subroutine to update LATERAL flags in a namespace list.
1830 : */
1831 : static void
1832 970980 : setNamespaceLateralState(List *namespace, bool lateral_only, bool lateral_ok)
1833 : {
1834 : ListCell *lc;
1835 :
1836 2383518 : foreach(lc, namespace)
1837 : {
1838 1412538 : ParseNamespaceItem *nsitem = (ParseNamespaceItem *) lfirst(lc);
1839 :
1840 1412538 : nsitem->p_lateral_only = lateral_only;
1841 1412538 : nsitem->p_lateral_ok = lateral_ok;
1842 : }
1843 970980 : }
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 1008762 : transformWhereClause(ParseState *pstate, Node *clause,
1855 : ParseExprKind exprKind, const char *constructName)
1856 : {
1857 : Node *qual;
1858 :
1859 1008762 : if (clause == NULL)
1860 727510 : return NULL;
1861 :
1862 281252 : qual = transformExpr(pstate, clause, exprKind);
1863 :
1864 281042 : qual = coerce_to_boolean(pstate, qual, constructName);
1865 :
1866 281036 : 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 937836 : transformLimitClause(ParseState *pstate, Node *clause,
1882 : ParseExprKind exprKind, const char *constructName,
1883 : LimitOption limitOption)
1884 : {
1885 : Node *qual;
1886 :
1887 937836 : if (clause == NULL)
1888 932560 : return NULL;
1889 :
1890 5276 : qual = transformExpr(pstate, clause, exprKind);
1891 :
1892 5270 : qual = coerce_to_specific_type(pstate, qual, INT8OID, constructName);
1893 :
1894 : /* LIMIT can't refer to any variables of the current query */
1895 5270 : 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 5270 : if (exprKind == EXPR_KIND_LIMIT && limitOption == LIMIT_OPTION_WITH_TIES &&
1904 52 : 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 5264 : 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 7076 : checkExprIsVarFree(ParseState *pstate, Node *n, const char *constructName)
1926 : {
1927 7076 : 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 7070 : }
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 61186 : checkTargetlistEntrySQL92(ParseState *pstate, TargetEntry *tle,
1951 : ParseExprKind exprKind)
1952 : {
1953 61186 : switch (exprKind)
1954 : {
1955 794 : case EXPR_KIND_GROUP_BY:
1956 : /* reject aggregates and window functions */
1957 1454 : if (pstate->p_hasAggs &&
1958 660 : 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 806 : 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 788 : break;
1976 60152 : case EXPR_KIND_ORDER_BY:
1977 : /* no extra checks needed */
1978 60152 : break;
1979 240 : case EXPR_KIND_DISTINCT_ON:
1980 : /* no extra checks needed */
1981 240 : break;
1982 0 : default:
1983 0 : elog(ERROR, "unexpected exprKind in checkTargetlistEntrySQL92");
1984 : break;
1985 : }
1986 61180 : }
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 97278 : 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 153220 : if (IsA(node, ColumnRef) &&
2050 55942 : list_length(((ColumnRef *) node)->fields) == 1 &&
2051 38788 : IsA(linitial(((ColumnRef *) node)->fields), String))
2052 : {
2053 38788 : char *name = strVal(linitial(((ColumnRef *) node)->fields));
2054 38788 : int location = ((ColumnRef *) node)->location;
2055 :
2056 38788 : 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 4676 : if (colNameToVar(pstate, name, true, location) != NULL)
2075 4566 : name = NULL;
2076 : }
2077 :
2078 38788 : if (name != NULL)
2079 : {
2080 34222 : TargetEntry *target_result = NULL;
2081 :
2082 189458 : foreach(tl, *tlist)
2083 : {
2084 155236 : TargetEntry *tle = (TargetEntry *) lfirst(tl);
2085 :
2086 155236 : if (!tle->resjunk &&
2087 154612 : strcmp(tle->resname, name) == 0)
2088 : {
2089 29168 : 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 29162 : target_result = tle;
2104 : /* Stay in loop to check for ambiguity */
2105 : }
2106 : }
2107 34222 : if (target_result != NULL)
2108 : {
2109 : /* return the first match, after suitable validation */
2110 29162 : checkTargetlistEntrySQL92(pstate, target_result, exprKind);
2111 29162 : return target_result;
2112 : }
2113 : }
2114 : }
2115 68116 : if (IsA(node, A_Const))
2116 : {
2117 32030 : A_Const *aconst = castNode(A_Const, node);
2118 32030 : int targetlist_pos = 0;
2119 : int target_pos;
2120 :
2121 32030 : 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 32030 : target_pos = intVal(&aconst->val);
2130 56524 : foreach(tl, *tlist)
2131 : {
2132 56518 : TargetEntry *tle = (TargetEntry *) lfirst(tl);
2133 :
2134 56518 : if (!tle->resjunk)
2135 : {
2136 56518 : if (++targetlist_pos == target_pos)
2137 : {
2138 : /* return the unique match, after suitable validation */
2139 32024 : checkTargetlistEntrySQL92(pstate, tle, exprKind);
2140 32018 : 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 36086 : 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 41440 : 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 41440 : expr = transformExpr(pstate, node, exprKind);
2187 :
2188 153420 : foreach(tl, *tlist)
2189 : {
2190 128816 : 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 128816 : texpr = strip_implicit_coercions((Node *) tle->expr);
2203 :
2204 128816 : if (equal(expr, texpr))
2205 16782 : 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 24604 : target_result = transformTargetEntry(pstate, node, expr, exprKind,
2214 : NULL, true);
2215 :
2216 24604 : *tlist = lappend(*tlist, target_result);
2217 :
2218 24604 : 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 468068 : flatten_grouping_sets(Node *expr, bool toplevel, bool *hasGroupingSets)
2259 : {
2260 : /* just in case of pathological input */
2261 468068 : check_stack_depth();
2262 :
2263 468068 : if (expr == (Node *) NIL)
2264 451832 : return (Node *) NIL;
2265 :
2266 16236 : switch (expr->type)
2267 : {
2268 352 : case T_RowExpr:
2269 : {
2270 352 : RowExpr *r = (RowExpr *) expr;
2271 :
2272 352 : if (r->row_format == COERCE_IMPLICIT_CAST)
2273 352 : return flatten_grouping_sets((Node *) r->args,
2274 : false, NULL);
2275 : }
2276 0 : break;
2277 1248 : case T_GroupingSet:
2278 : {
2279 1248 : GroupingSet *gset = (GroupingSet *) expr;
2280 : ListCell *l2;
2281 1248 : List *result_set = NIL;
2282 :
2283 1248 : if (hasGroupingSets)
2284 924 : *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 1248 : if (toplevel && gset->kind == GROUPING_SET_EMPTY)
2293 18 : return (Node *) NIL;
2294 :
2295 3292 : foreach(l2, gset->content)
2296 : {
2297 2062 : Node *n1 = lfirst(l2);
2298 2062 : Node *n2 = flatten_grouping_sets(n1, false, NULL);
2299 :
2300 2062 : 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 1966 : 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 1230 : if (toplevel || (gset->kind != GROUPING_SET_SETS))
2314 : {
2315 1134 : return (Node *) makeGroupingSet(gset->kind, result_set, gset->location);
2316 : }
2317 : else
2318 96 : return (Node *) result_set;
2319 : }
2320 5592 : case T_List:
2321 : {
2322 5592 : List *result = NIL;
2323 : ListCell *l;
2324 :
2325 14174 : foreach(l, (List *) expr)
2326 : {
2327 8582 : Node *n = flatten_grouping_sets(lfirst(l), toplevel, hasGroupingSets);
2328 :
2329 8582 : if (n != (Node *) NIL)
2330 : {
2331 8564 : if (IsA(n, List))
2332 46 : result = list_concat(result, (List *) n);
2333 : else
2334 8518 : result = lappend(result, n);
2335 : }
2336 : }
2337 :
2338 5592 : return (Node *) result;
2339 : }
2340 9044 : default:
2341 9044 : break;
2342 : }
2343 :
2344 9044 : 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 9044 : 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 9044 : bool found = false;
2374 :
2375 9044 : if (useSQL99)
2376 1154 : tle = findTargetlistEntrySQL99(pstate, gexpr,
2377 : targetlist, exprKind);
2378 : else
2379 7890 : tle = findTargetlistEntrySQL92(pstate, gexpr,
2380 : targetlist, exprKind);
2381 :
2382 9020 : 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 2652 : 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 2628 : found = targetIsInSortList(tle, InvalidOid, *flatresult);
2402 2628 : if (found)
2403 262 : 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 3300 : foreach(sl, sortClause)
2425 : {
2426 3110 : SortGroupClause *sc = (SortGroupClause *) lfirst(sl);
2427 :
2428 3110 : if (sc->tleSortGroupRef == tle->ressortgroupref)
2429 : {
2430 2176 : SortGroupClause *grpc = copyObject(sc);
2431 :
2432 2176 : if (!toplevel)
2433 592 : grpc->nulls_first = false;
2434 2176 : *flatresult = lappend(*flatresult, grpc);
2435 2176 : found = true;
2436 2176 : 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 8734 : if (!found)
2446 6558 : *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 8734 : 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 306 : transformGroupClauseList(List **flatresult,
2476 : ParseState *pstate, List *list,
2477 : List **targetlist, List *sortClause,
2478 : ParseExprKind exprKind, bool useSQL99, bool toplevel)
2479 : {
2480 306 : Bitmapset *seen_local = NULL;
2481 306 : List *result = NIL;
2482 : ListCell *gl;
2483 :
2484 942 : foreach(gl, list)
2485 : {
2486 636 : Node *gexpr = (Node *) lfirst(gl);
2487 :
2488 636 : Index ref = transformGroupClauseExpr(flatresult,
2489 : seen_local,
2490 : pstate,
2491 : gexpr,
2492 : targetlist,
2493 : sortClause,
2494 : exprKind,
2495 : useSQL99,
2496 : toplevel);
2497 :
2498 636 : if (ref > 0)
2499 : {
2500 624 : seen_local = bms_add_member(seen_local, ref);
2501 624 : result = lappend_int(result, ref);
2502 : }
2503 : }
2504 :
2505 306 : 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 1134 : 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 1134 : List *content = NIL;
2535 :
2536 : Assert(toplevel || gset->kind != GROUPING_SET_SETS);
2537 :
2538 3100 : foreach(gl, gset->content)
2539 : {
2540 1966 : Node *n = lfirst(gl);
2541 :
2542 1966 : if (IsA(n, List))
2543 : {
2544 306 : List *l = transformGroupClauseList(flatresult,
2545 : pstate, (List *) n,
2546 : targetlist, sortClause,
2547 : exprKind, useSQL99, false);
2548 :
2549 306 : content = lappend(content, makeGroupingSet(GROUPING_SET_SIMPLE,
2550 : l,
2551 : exprLocation(n)));
2552 : }
2553 1660 : 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 1432 : Index ref = transformGroupClauseExpr(flatresult,
2565 : NULL,
2566 : pstate,
2567 : n,
2568 : targetlist,
2569 : sortClause,
2570 : exprKind,
2571 : useSQL99,
2572 : false);
2573 :
2574 2864 : content = lappend(content, makeGroupingSet(GROUPING_SET_SIMPLE,
2575 1432 : list_make1_int(ref),
2576 : exprLocation(n)));
2577 : }
2578 : }
2579 :
2580 : /* Arbitrarily cap the size of CUBE, which has exponential growth */
2581 1134 : 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 1134 : 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 457072 : transformGroupClause(ParseState *pstate, List *grouplist, List **groupingSets,
2633 : List **targetlist, List *sortClause,
2634 : ParseExprKind exprKind, bool useSQL99)
2635 : {
2636 457072 : List *result = NIL;
2637 : List *flat_grouplist;
2638 457072 : List *gsets = NIL;
2639 : ListCell *gl;
2640 457072 : bool hasGroupingSets = false;
2641 457072 : 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 457072 : 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 457072 : 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 464948 : foreach(gl, flat_grouplist)
2666 : {
2667 7900 : Node *gexpr = (Node *) lfirst(gl);
2668 :
2669 7900 : if (IsA(gexpr, GroupingSet))
2670 : {
2671 924 : GroupingSet *gset = (GroupingSet *) gexpr;
2672 :
2673 924 : 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 906 : case GROUPING_SET_SETS:
2683 : case GROUPING_SET_CUBE:
2684 : case GROUPING_SET_ROLLUP:
2685 906 : gsets = lappend(gsets,
2686 906 : transformGroupingSet(&result,
2687 : pstate, gset,
2688 : targetlist, sortClause,
2689 : exprKind, useSQL99, true));
2690 906 : break;
2691 : }
2692 924 : }
2693 : else
2694 : {
2695 6976 : Index ref = transformGroupClauseExpr(&result, seen_local,
2696 : pstate, gexpr,
2697 : targetlist, sortClause,
2698 : exprKind, useSQL99, true);
2699 :
2700 6952 : if (ref > 0)
2701 : {
2702 6940 : seen_local = bms_add_member(seen_local, ref);
2703 6940 : 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 457048 : if (groupingSets)
2716 454332 : *groupingSets = gsets;
2717 :
2718 457048 : 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 513492 : transformSortClause(ParseState *pstate,
2733 : List *orderlist,
2734 : List **targetlist,
2735 : ParseExprKind exprKind,
2736 : bool useSQL99)
2737 : {
2738 513492 : List *sortlist = NIL;
2739 : ListCell *olitem;
2740 :
2741 606648 : foreach(olitem, orderlist)
2742 : {
2743 93198 : SortBy *sortby = (SortBy *) lfirst(olitem);
2744 : TargetEntry *tle;
2745 :
2746 93198 : if (useSQL99)
2747 4200 : tle = findTargetlistEntrySQL99(pstate, sortby->node,
2748 : targetlist, exprKind);
2749 : else
2750 88998 : tle = findTargetlistEntrySQL92(pstate, sortby->node,
2751 : targetlist, exprKind);
2752 :
2753 93162 : sortlist = addTargetToSortList(pstate, tle,
2754 : sortlist, *targetlist, sortby);
2755 : }
2756 :
2757 513450 : return sortlist;
2758 : }
2759 :
2760 : /*
2761 : * transformWindowDefinitions -
2762 : * transform window definitions (WindowDef to WindowClause)
2763 : */
2764 : List *
2765 454308 : transformWindowDefinitions(ParseState *pstate,
2766 : List *windowdefs,
2767 : List **targetlist)
2768 : {
2769 454308 : List *result = NIL;
2770 454308 : Index winref = 0;
2771 : ListCell *lc;
2772 :
2773 456970 : foreach(lc, windowdefs)
2774 : {
2775 2728 : WindowDef *windef = (WindowDef *) lfirst(lc);
2776 2728 : WindowClause *refwc = NULL;
2777 : List *partitionClause;
2778 : List *orderClause;
2779 2728 : Oid rangeopfamily = InvalidOid;
2780 2728 : Oid rangeopcintype = InvalidOid;
2781 : WindowClause *wc;
2782 :
2783 2728 : winref++;
2784 :
2785 : /*
2786 : * Check for duplicate window names.
2787 : */
2788 3268 : 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 2722 : 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 2722 : orderClause = transformSortClause(pstate,
2815 : windef->orderClause,
2816 : targetlist,
2817 : EXPR_KIND_WINDOW_ORDER,
2818 : true /* force SQL99 rules */ );
2819 2716 : 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 2716 : wc = makeNode(WindowClause);
2831 2716 : wc->name = windef->name;
2832 2716 : 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 2716 : 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 2692 : wc->partitionClause = partitionClause;
2860 2716 : 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 2692 : wc->orderClause = orderClause;
2882 2692 : wc->copiedOrder = false;
2883 : }
2884 2716 : 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 2716 : 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 2716 : if ((wc->frameOptions & FRAMEOPTION_RANGE) &&
2913 1942 : (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 = !sortcl->reverse_sort;
2937 618 : wc->inRangeNullsFirst = sortcl->nulls_first;
2938 : }
2939 :
2940 : /* Per spec, GROUPS mode requires an ORDER BY clause */
2941 2698 : 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 2692 : wc->startOffset = transformFrameOffset(pstate, wc->frameOptions,
2952 : rangeopfamily, rangeopcintype,
2953 : &wc->startInRangeFunc,
2954 : windef->startOffset);
2955 2668 : wc->endOffset = transformFrameOffset(pstate, wc->frameOptions,
2956 : rangeopfamily, rangeopcintype,
2957 : &wc->endInRangeFunc,
2958 : windef->endOffset);
2959 2662 : wc->winref = winref;
2960 :
2961 2662 : result = lappend(result, wc);
2962 : }
2963 :
2964 454242 : 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 3638 : transformDistinctClause(ParseState *pstate,
2986 : List **targetlist, List *sortClause, bool is_agg)
2987 : {
2988 3638 : 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 4264 : foreach(slitem, sortClause)
3008 : {
3009 662 : SortGroupClause *scl = (SortGroupClause *) lfirst(slitem);
3010 662 : TargetEntry *tle = get_sortgroupclause_tle(scl, *targetlist);
3011 :
3012 662 : 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 626 : 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 14068 : foreach(tlitem, *targetlist)
3028 : {
3029 10466 : TargetEntry *tle = (TargetEntry *) lfirst(tlitem);
3030 :
3031 10466 : if (tle->resjunk)
3032 4 : continue; /* ignore junk */
3033 10462 : result = addTargetToGroupList(pstate, tle,
3034 : result, *targetlist,
3035 10462 : 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 3602 : 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 3602 : 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 252 : transformDistinctOnClause(ParseState *pstate, List *distinctlist,
3070 : List **targetlist, List *sortClause)
3071 : {
3072 252 : List *result = NIL;
3073 252 : 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 636 : foreach(lc, distinctlist)
3087 : {
3088 390 : Node *dexpr = (Node *) lfirst(lc);
3089 : int sortgroupref;
3090 : TargetEntry *tle;
3091 :
3092 390 : tle = findTargetlistEntrySQL92(pstate, dexpr, targetlist,
3093 : EXPR_KIND_DISTINCT_ON);
3094 384 : sortgroupref = assignSortGroupRef(tle, *targetlist);
3095 384 : 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 246 : skipped_sortitem = false;
3107 576 : foreach(lc, sortClause)
3108 : {
3109 336 : SortGroupClause *scl = (SortGroupClause *) lfirst(lc);
3110 :
3111 336 : if (list_member_int(sortgrouprefs, scl->tleSortGroupRef))
3112 : {
3113 264 : 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 258 : result = lappend(result, copyObject(scl));
3123 : }
3124 : else
3125 72 : 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 612 : forboth(lc, distinctlist, lc2, sortgrouprefs)
3138 : {
3139 372 : Node *dexpr = (Node *) lfirst(lc);
3140 372 : int sortgroupref = lfirst_int(lc2);
3141 372 : TargetEntry *tle = get_sortgroupref_tle(sortgroupref, *targetlist);
3142 :
3143 372 : if (targetIsInSortList(tle, InvalidOid, result))
3144 252 : continue; /* already in list (with some semantics) */
3145 120 : 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 120 : 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 240 : 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 1430 : resolve_unique_index_expr(ParseState *pstate, InferClause *infer,
3202 : Relation heapRel)
3203 : {
3204 1430 : List *result = NIL;
3205 : ListCell *l;
3206 :
3207 3314 : foreach(l, infer->indexElems)
3208 : {
3209 1890 : IndexElem *ielem = (IndexElem *) lfirst(l);
3210 1890 : 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 1890 : 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 1890 : 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 1890 : 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 1716 : n = makeNode(ColumnRef);
3249 1716 : n->fields = list_make1(makeString(ielem->name));
3250 : /* Location is approximately that of inference specification */
3251 1716 : n->location = infer->location;
3252 1716 : parse = (Node *) n;
3253 : }
3254 : else
3255 : {
3256 : /* Do parse transformation of the raw expression */
3257 174 : 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 1890 : pInfer->expr = transformExpr(pstate, parse, EXPR_KIND_INDEX_EXPRESSION);
3268 :
3269 : /* Perform lookup of collation and operator class as required */
3270 1884 : if (!ielem->collation)
3271 1842 : pInfer->infercollid = InvalidOid;
3272 : else
3273 42 : pInfer->infercollid = LookupCollation(pstate, ielem->collation,
3274 42 : exprLocation(pInfer->expr));
3275 :
3276 1884 : if (!ielem->opclass)
3277 1842 : pInfer->inferopclass = InvalidOid;
3278 : else
3279 42 : pInfer->inferopclass = get_opclass_oid(BTREE_AM_OID,
3280 : ielem->opclass, false);
3281 :
3282 1884 : result = lappend(result, pInfer);
3283 : }
3284 :
3285 1424 : 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 1856 : transformOnConflictArbiter(ParseState *pstate,
3298 : OnConflictClause *onConflictClause,
3299 : List **arbiterExpr, Node **arbiterWhere,
3300 : Oid *constraint)
3301 : {
3302 1856 : InferClause *infer = onConflictClause->infer;
3303 :
3304 1856 : *arbiterExpr = NIL;
3305 1856 : *arbiterWhere = NULL;
3306 1856 : *constraint = InvalidOid;
3307 :
3308 1856 : 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 1850 : 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 1850 : 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 1850 : if (infer)
3338 : {
3339 1622 : if (infer->indexElems)
3340 1430 : *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 1616 : if (infer->whereClause)
3348 48 : *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 1616 : if (infer->conname)
3358 : {
3359 192 : Oid relid = RelationGetRelid(pstate->p_target_relation);
3360 192 : RTEPermissionInfo *perminfo = pstate->p_target_nsitem->p_perminfo;
3361 : Bitmapset *conattnos;
3362 :
3363 192 : 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 192 : perminfo->requiredPerms |= ACL_SELECT;
3368 : /* Mark the constrained columns as requiring SELECT access */
3369 192 : 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 1844 : }
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 93534 : addTargetToSortList(ParseState *pstate, TargetEntry *tle,
3394 : List *sortlist, List *targetlist, SortBy *sortby)
3395 : {
3396 93534 : 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 93534 : 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 93534 : location = sortby->location;
3424 93534 : if (location < 0)
3425 93314 : location = exprLocation(sortby->node);
3426 93534 : setup_parser_errposition_callback(&pcbstate, pstate, location);
3427 :
3428 : /* determine the sortop, eqop, and directionality */
3429 93534 : switch (sortby->sortby_dir)
3430 : {
3431 90458 : case SORTBY_DEFAULT:
3432 : case SORTBY_ASC:
3433 90458 : get_sort_group_operators(restype,
3434 : true, true, false,
3435 : &sortop, &eqop, NULL,
3436 : &hashable);
3437 90452 : reverse = false;
3438 90452 : break;
3439 2856 : case SORTBY_DESC:
3440 2856 : get_sort_group_operators(restype,
3441 : false, true, true,
3442 : NULL, &eqop, &sortop,
3443 : &hashable);
3444 2856 : reverse = true;
3445 2856 : 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 93528 : cancel_parser_errposition_callback(&pcbstate);
3481 :
3482 : /* avoid making duplicate sortlist entries */
3483 93528 : if (!targetIsInSortList(tle, sortop, sortlist))
3484 : {
3485 93528 : SortGroupClause *sortcl = makeNode(SortGroupClause);
3486 :
3487 93528 : sortcl->tleSortGroupRef = assignSortGroupRef(tle, targetlist);
3488 :
3489 93528 : sortcl->eqop = eqop;
3490 93528 : sortcl->sortop = sortop;
3491 93528 : sortcl->hashable = hashable;
3492 93528 : sortcl->reverse_sort = reverse;
3493 :
3494 93528 : switch (sortby->sortby_nulls)
3495 : {
3496 91512 : case SORTBY_NULLS_DEFAULT:
3497 : /* NULLS FIRST is default for DESC; other way for ASC */
3498 91512 : sortcl->nulls_first = reverse;
3499 91512 : break;
3500 336 : case SORTBY_NULLS_FIRST:
3501 336 : sortcl->nulls_first = true;
3502 336 : break;
3503 1680 : case SORTBY_NULLS_LAST:
3504 1680 : sortcl->nulls_first = false;
3505 1680 : break;
3506 0 : default:
3507 0 : elog(ERROR, "unrecognized sortby_nulls: %d",
3508 : sortby->sortby_nulls);
3509 : break;
3510 : }
3511 :
3512 93528 : sortlist = lappend(sortlist, sortcl);
3513 : }
3514 :
3515 93528 : return sortlist;
3516 : }
3517 :
3518 : /*
3519 : * addTargetToGroupList
3520 : * If the given targetlist entry isn't already in the SortGroupClause
3521 : * list, add it to the end of the list, using default sort/group
3522 : * semantics.
3523 : *
3524 : * This is very similar to addTargetToSortList, except that we allow the
3525 : * case where only a grouping (equality) operator can be found, and that
3526 : * the TLE is considered "already in the list" if it appears there with any
3527 : * sorting semantics.
3528 : *
3529 : * location is the parse location to be fingered in event of trouble. Note
3530 : * that we can't rely on exprLocation(tle->expr), because that might point
3531 : * to a SELECT item that matches the GROUP BY item; it'd be pretty confusing
3532 : * to report such a location.
3533 : *
3534 : * Returns the updated SortGroupClause list.
3535 : */
3536 : static List *
3537 17140 : addTargetToGroupList(ParseState *pstate, TargetEntry *tle,
3538 : List *grouplist, List *targetlist, int location)
3539 : {
3540 17140 : Oid restype = exprType((Node *) tle->expr);
3541 :
3542 : /* if tlist item is an UNKNOWN literal, change it to TEXT */
3543 17140 : if (restype == UNKNOWNOID)
3544 : {
3545 16 : tle->expr = (Expr *) coerce_type(pstate, (Node *) tle->expr,
3546 : restype, TEXTOID, -1,
3547 : COERCION_IMPLICIT,
3548 : COERCE_IMPLICIT_CAST,
3549 : -1);
3550 16 : restype = TEXTOID;
3551 : }
3552 :
3553 : /* avoid making duplicate grouplist entries */
3554 17140 : if (!targetIsInSortList(tle, InvalidOid, grouplist))
3555 : {
3556 16538 : SortGroupClause *grpcl = makeNode(SortGroupClause);
3557 : Oid sortop;
3558 : Oid eqop;
3559 : bool hashable;
3560 : ParseCallbackState pcbstate;
3561 :
3562 16538 : setup_parser_errposition_callback(&pcbstate, pstate, location);
3563 :
3564 : /* determine the eqop and optional sortop */
3565 16538 : get_sort_group_operators(restype,
3566 : false, true, false,
3567 : &sortop, &eqop, NULL,
3568 : &hashable);
3569 :
3570 16538 : cancel_parser_errposition_callback(&pcbstate);
3571 :
3572 16538 : grpcl->tleSortGroupRef = assignSortGroupRef(tle, targetlist);
3573 16538 : grpcl->eqop = eqop;
3574 16538 : grpcl->sortop = sortop;
3575 16538 : grpcl->reverse_sort = false; /* sortop is "less than", or
3576 : * InvalidOid */
3577 16538 : grpcl->nulls_first = false; /* OK with or without sortop */
3578 16538 : grpcl->hashable = hashable;
3579 :
3580 16538 : grouplist = lappend(grouplist, grpcl);
3581 : }
3582 :
3583 17140 : return grouplist;
3584 : }
3585 :
3586 : /*
3587 : * assignSortGroupRef
3588 : * Assign the targetentry an unused ressortgroupref, if it doesn't
3589 : * already have one. Return the assigned or pre-existing refnumber.
3590 : *
3591 : * 'tlist' is the targetlist containing (or to contain) the given targetentry.
3592 : */
3593 : Index
3594 146326 : assignSortGroupRef(TargetEntry *tle, List *tlist)
3595 : {
3596 : Index maxRef;
3597 : ListCell *l;
3598 :
3599 146326 : if (tle->ressortgroupref) /* already has one? */
3600 934 : return tle->ressortgroupref;
3601 :
3602 : /* easiest way to pick an unused refnumber: max used + 1 */
3603 145392 : maxRef = 0;
3604 910072 : foreach(l, tlist)
3605 : {
3606 764680 : Index ref = ((TargetEntry *) lfirst(l))->ressortgroupref;
3607 :
3608 764680 : if (ref > maxRef)
3609 144744 : maxRef = ref;
3610 : }
3611 145392 : tle->ressortgroupref = maxRef + 1;
3612 145392 : return tle->ressortgroupref;
3613 : }
3614 :
3615 : /*
3616 : * targetIsInSortList
3617 : * Is the given target item already in the sortlist?
3618 : * If sortop is not InvalidOid, also test for a match to the sortop.
3619 : *
3620 : * It is not an oversight that this function ignores the nulls_first flag.
3621 : * We check sortop when determining if an ORDER BY item is redundant with
3622 : * earlier ORDER BY items, because it's conceivable that "ORDER BY
3623 : * foo USING <, foo USING <<<" is not redundant, if <<< distinguishes
3624 : * values that < considers equal. We need not check nulls_first
3625 : * however, because a lower-order column with the same sortop but
3626 : * opposite nulls direction is redundant. Also, we can consider
3627 : * ORDER BY foo ASC, foo DESC redundant, so check for a commutator match.
3628 : *
3629 : * Works for both ordering and grouping lists (sortop would normally be
3630 : * InvalidOid when considering grouping). Note that the main reason we need
3631 : * this routine (and not just a quick test for nonzeroness of ressortgroupref)
3632 : * is that a TLE might be in only one of the lists.
3633 : */
3634 : bool
3635 115430 : targetIsInSortList(TargetEntry *tle, Oid sortop, List *sortList)
3636 : {
3637 115430 : Index ref = tle->ressortgroupref;
3638 : ListCell *l;
3639 :
3640 : /* no need to scan list if tle has no marker */
3641 115430 : if (ref == 0)
3642 110608 : return false;
3643 :
3644 6138 : foreach(l, sortList)
3645 : {
3646 2824 : SortGroupClause *scl = (SortGroupClause *) lfirst(l);
3647 :
3648 2824 : if (scl->tleSortGroupRef == ref &&
3649 0 : (sortop == InvalidOid ||
3650 0 : sortop == scl->sortop ||
3651 0 : sortop == get_commutator(scl->sortop)))
3652 1508 : return true;
3653 : }
3654 3314 : return false;
3655 : }
3656 :
3657 : /*
3658 : * findWindowClause
3659 : * Find the named WindowClause in the list, or return NULL if not there
3660 : */
3661 : static WindowClause *
3662 564 : findWindowClause(List *wclist, const char *name)
3663 : {
3664 : ListCell *l;
3665 :
3666 570 : foreach(l, wclist)
3667 : {
3668 36 : WindowClause *wc = (WindowClause *) lfirst(l);
3669 :
3670 36 : if (wc->name && strcmp(wc->name, name) == 0)
3671 30 : return wc;
3672 : }
3673 :
3674 534 : return NULL;
3675 : }
3676 :
3677 : /*
3678 : * transformFrameOffset
3679 : * Process a window frame offset expression
3680 : *
3681 : * In RANGE mode, rangeopfamily is the sort opfamily for the input ORDER BY
3682 : * column, and rangeopcintype is the input data type the sort operator is
3683 : * registered with. We expect the in_range function to be registered with
3684 : * that same type. (In binary-compatible cases, it might be different from
3685 : * the input column's actual type, so we can't use that for the lookups.)
3686 : * We'll return the OID of the in_range function to *inRangeFunc.
3687 : */
3688 : static Node *
3689 5360 : transformFrameOffset(ParseState *pstate, int frameOptions,
3690 : Oid rangeopfamily, Oid rangeopcintype, Oid *inRangeFunc,
3691 : Node *clause)
3692 : {
3693 5360 : const char *constructName = NULL;
3694 : Node *node;
3695 :
3696 5360 : *inRangeFunc = InvalidOid; /* default result */
3697 :
3698 : /* Quick exit if no offset expression */
3699 5360 : if (clause == NULL)
3700 3530 : return NULL;
3701 :
3702 1830 : if (frameOptions & FRAMEOPTION_ROWS)
3703 : {
3704 : /* Transform the raw expression tree */
3705 396 : node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_ROWS);
3706 :
3707 : /*
3708 : * Like LIMIT clause, simply coerce to int8
3709 : */
3710 396 : constructName = "ROWS";
3711 396 : node = coerce_to_specific_type(pstate, node, INT8OID, constructName);
3712 : }
3713 1434 : else if (frameOptions & FRAMEOPTION_RANGE)
3714 : {
3715 : /*
3716 : * We must look up the in_range support function that's to be used,
3717 : * possibly choosing one of several, and coerce the "offset" value to
3718 : * the appropriate input type.
3719 : */
3720 : Oid nodeType;
3721 : Oid preferredType;
3722 1152 : int nfuncs = 0;
3723 1152 : int nmatches = 0;
3724 1152 : Oid selectedType = InvalidOid;
3725 1152 : Oid selectedFunc = InvalidOid;
3726 : CatCList *proclist;
3727 : int i;
3728 :
3729 : /* Transform the raw expression tree */
3730 1152 : node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_RANGE);
3731 1152 : nodeType = exprType(node);
3732 :
3733 : /*
3734 : * If there are multiple candidates, we'll prefer the one that exactly
3735 : * matches nodeType; or if nodeType is as yet unknown, prefer the one
3736 : * that exactly matches the sort column type. (The second rule is
3737 : * like what we do for "known_type operator unknown".)
3738 : */
3739 1152 : preferredType = (nodeType != UNKNOWNOID) ? nodeType : rangeopcintype;
3740 :
3741 : /* Find the in_range support functions applicable to this case */
3742 1152 : proclist = SearchSysCacheList2(AMPROCNUM,
3743 : ObjectIdGetDatum(rangeopfamily),
3744 : ObjectIdGetDatum(rangeopcintype));
3745 7344 : for (i = 0; i < proclist->n_members; i++)
3746 : {
3747 6192 : HeapTuple proctup = &proclist->members[i]->tuple;
3748 6192 : Form_pg_amproc procform = (Form_pg_amproc) GETSTRUCT(proctup);
3749 :
3750 : /* The search will find all support proc types; ignore others */
3751 6192 : if (procform->amprocnum != BTINRANGE_PROC)
3752 4434 : continue;
3753 1758 : nfuncs++;
3754 :
3755 : /* Ignore function if given value can't be coerced to that type */
3756 1758 : if (!can_coerce_type(1, &nodeType, &procform->amprocrighttype,
3757 : COERCION_IMPLICIT))
3758 330 : continue;
3759 1428 : nmatches++;
3760 :
3761 : /* Remember preferred match, or any match if didn't find that */
3762 1428 : if (selectedType != preferredType)
3763 : {
3764 1368 : selectedType = procform->amprocrighttype;
3765 1368 : selectedFunc = procform->amproc;
3766 : }
3767 : }
3768 1152 : ReleaseCatCacheList(proclist);
3769 :
3770 : /*
3771 : * Throw error if needed. It seems worth taking the trouble to
3772 : * distinguish "no support at all" from "you didn't match any
3773 : * available offset type".
3774 : */
3775 1152 : if (nfuncs == 0)
3776 6 : ereport(ERROR,
3777 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3778 : errmsg("RANGE with offset PRECEDING/FOLLOWING is not supported for column type %s",
3779 : format_type_be(rangeopcintype)),
3780 : parser_errposition(pstate, exprLocation(node))));
3781 1146 : if (nmatches == 0)
3782 18 : ereport(ERROR,
3783 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3784 : errmsg("RANGE with offset PRECEDING/FOLLOWING is not supported for column type %s and offset type %s",
3785 : format_type_be(rangeopcintype),
3786 : format_type_be(nodeType)),
3787 : errhint("Cast the offset value to an appropriate type."),
3788 : parser_errposition(pstate, exprLocation(node))));
3789 1128 : if (nmatches != 1 && selectedType != preferredType)
3790 0 : ereport(ERROR,
3791 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3792 : errmsg("RANGE with offset PRECEDING/FOLLOWING has multiple interpretations for column type %s and offset type %s",
3793 : format_type_be(rangeopcintype),
3794 : format_type_be(nodeType)),
3795 : errhint("Cast the offset value to the exact intended type."),
3796 : parser_errposition(pstate, exprLocation(node))));
3797 :
3798 : /* OK, coerce the offset to the right type */
3799 1128 : constructName = "RANGE";
3800 1128 : node = coerce_to_specific_type(pstate, node,
3801 : selectedType, constructName);
3802 1128 : *inRangeFunc = selectedFunc;
3803 : }
3804 282 : else if (frameOptions & FRAMEOPTION_GROUPS)
3805 : {
3806 : /* Transform the raw expression tree */
3807 282 : node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_GROUPS);
3808 :
3809 : /*
3810 : * Like LIMIT clause, simply coerce to int8
3811 : */
3812 282 : constructName = "GROUPS";
3813 282 : node = coerce_to_specific_type(pstate, node, INT8OID, constructName);
3814 : }
3815 : else
3816 : {
3817 : Assert(false);
3818 0 : node = NULL;
3819 : }
3820 :
3821 : /* Disallow variables in frame offsets */
3822 1806 : checkExprIsVarFree(pstate, node, constructName);
3823 :
3824 1800 : return node;
3825 : }
|
