Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * parse_clause.c
4 : * handle clauses in parser
5 : *
6 : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/parser/parse_clause.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 :
16 : #include "postgres.h"
17 :
18 : #include "access/htup_details.h"
19 : #include "access/nbtree.h"
20 : #include "access/table.h"
21 : #include "access/tsmapi.h"
22 : #include "catalog/catalog.h"
23 : #include "catalog/pg_am.h"
24 : #include "catalog/pg_amproc.h"
25 : #include "catalog/pg_constraint.h"
26 : #include "catalog/pg_type.h"
27 : #include "commands/defrem.h"
28 : #include "miscadmin.h"
29 : #include "nodes/makefuncs.h"
30 : #include "nodes/nodeFuncs.h"
31 : #include "optimizer/optimizer.h"
32 : #include "parser/analyze.h"
33 : #include "parser/parse_clause.h"
34 : #include "parser/parse_coerce.h"
35 : #include "parser/parse_collate.h"
36 : #include "parser/parse_expr.h"
37 : #include "parser/parse_func.h"
38 : #include "parser/parse_oper.h"
39 : #include "parser/parse_relation.h"
40 : #include "parser/parse_target.h"
41 : #include "parser/parse_type.h"
42 : #include "parser/parser.h"
43 : #include "parser/parsetree.h"
44 : #include "rewrite/rewriteManip.h"
45 : #include "utils/builtins.h"
46 : #include "utils/catcache.h"
47 : #include "utils/guc.h"
48 : #include "utils/lsyscache.h"
49 : #include "utils/rel.h"
50 : #include "utils/syscache.h"
51 :
52 :
53 : static int extractRemainingColumns(ParseState *pstate,
54 : ParseNamespaceColumn *src_nscolumns,
55 : List *src_colnames,
56 : List **src_colnos,
57 : List **res_colnames, List **res_colvars,
58 : ParseNamespaceColumn *res_nscolumns);
59 : static Node *transformJoinUsingClause(ParseState *pstate,
60 : List *leftVars, List *rightVars);
61 : static Node *transformJoinOnClause(ParseState *pstate, JoinExpr *j,
62 : List *namespace);
63 : static ParseNamespaceItem *transformTableEntry(ParseState *pstate, RangeVar *r);
64 : static ParseNamespaceItem *transformRangeSubselect(ParseState *pstate,
65 : RangeSubselect *r);
66 : static ParseNamespaceItem *transformRangeFunction(ParseState *pstate,
67 : RangeFunction *r);
68 : static ParseNamespaceItem *transformRangeTableFunc(ParseState *pstate,
69 : RangeTableFunc *rtf);
70 : static TableSampleClause *transformRangeTableSample(ParseState *pstate,
71 : RangeTableSample *rts);
72 : static ParseNamespaceItem *getNSItemForSpecialRelationTypes(ParseState *pstate,
73 : RangeVar *rv);
74 : static Node *transformFromClauseItem(ParseState *pstate, Node *n,
75 : ParseNamespaceItem **top_nsitem,
76 : List **namespace);
77 : static Var *buildVarFromNSColumn(ParseState *pstate,
78 : ParseNamespaceColumn *nscol);
79 : static Node *buildMergedJoinVar(ParseState *pstate, JoinType jointype,
80 : Var *l_colvar, Var *r_colvar);
81 : static void markRelsAsNulledBy(ParseState *pstate, Node *n, int jindex);
82 : static void setNamespaceColumnVisibility(List *namespace, bool cols_visible);
83 : static void setNamespaceLateralState(List *namespace,
84 : bool lateral_only, bool lateral_ok);
85 : static void checkExprIsVarFree(ParseState *pstate, Node *n,
86 : const char *constructName);
87 : static TargetEntry *findTargetlistEntrySQL92(ParseState *pstate, Node *node,
88 : List **tlist, ParseExprKind exprKind);
89 : static TargetEntry *findTargetlistEntrySQL99(ParseState *pstate, Node *node,
90 : List **tlist, ParseExprKind exprKind);
91 : static int get_matching_location(int sortgroupref,
92 : List *sortgrouprefs, List *exprs);
93 : static List *resolve_unique_index_expr(ParseState *pstate, InferClause *infer,
94 : Relation heapRel);
95 : static List *addTargetToGroupList(ParseState *pstate, TargetEntry *tle,
96 : List *grouplist, List *targetlist, int location);
97 : static WindowClause *findWindowClause(List *wclist, const char *name);
98 : static Node *transformFrameOffset(ParseState *pstate, int frameOptions,
99 : Oid rangeopfamily, Oid rangeopcintype, Oid *inRangeFunc,
100 : Node *clause);
101 :
102 :
103 : /*
104 : * transformFromClause -
105 : * Process the FROM clause and add items to the query's range table,
106 : * joinlist, and namespace.
107 : *
108 : * Note: we assume that the pstate's p_rtable, p_joinlist, and p_namespace
109 : * lists were initialized to NIL when the pstate was created.
110 : * We will add onto any entries already present --- this is needed for rule
111 : * processing, as well as for UPDATE and DELETE.
112 : */
113 : void
114 458794 : transformFromClause(ParseState *pstate, List *frmList)
115 : {
116 : ListCell *fl;
117 :
118 : /*
119 : * The grammar will have produced a list of RangeVars, RangeSubselects,
120 : * RangeFunctions, and/or JoinExprs. Transform each one (possibly adding
121 : * entries to the rtable), check for duplicate refnames, and then add it
122 : * to the joinlist and namespace.
123 : *
124 : * Note we must process the items left-to-right for proper handling of
125 : * LATERAL references.
126 : */
127 778336 : foreach(fl, frmList)
128 : {
129 320134 : Node *n = lfirst(fl);
130 : ParseNamespaceItem *nsitem;
131 : List *namespace;
132 :
133 320134 : n = transformFromClauseItem(pstate, n,
134 : &nsitem,
135 : &namespace);
136 :
137 319548 : checkNameSpaceConflicts(pstate, pstate->p_namespace, namespace);
138 :
139 : /* Mark the new namespace items as visible only to LATERAL */
140 319542 : setNamespaceLateralState(namespace, true, true);
141 :
142 319542 : pstate->p_joinlist = lappend(pstate->p_joinlist, n);
143 319542 : pstate->p_namespace = list_concat(pstate->p_namespace, namespace);
144 : }
145 :
146 : /*
147 : * We're done parsing the FROM list, so make all namespace items
148 : * unconditionally visible. Note that this will also reset lateral_only
149 : * for any namespace items that were already present when we were called;
150 : * but those should have been that way already.
151 : */
152 458202 : setNamespaceLateralState(pstate->p_namespace, false, true);
153 458202 : }
154 :
155 : /*
156 : * setTargetTable
157 : * Add the target relation of INSERT/UPDATE/DELETE/MERGE to the range table,
158 : * and make the special links to it in the ParseState.
159 : *
160 : * We also open the target relation and acquire a write lock on it.
161 : * This must be done before processing the FROM list, in case the target
162 : * is also mentioned as a source relation --- we want to be sure to grab
163 : * the write lock before any read lock.
164 : *
165 : * If alsoSource is true, add the target to the query's joinlist and
166 : * namespace. For INSERT, we don't want the target to be joined to;
167 : * it's a destination of tuples, not a source. MERGE is actually
168 : * both, but we'll add it separately to joinlist and namespace, so
169 : * doing nothing (like INSERT) is correct here. For UPDATE/DELETE,
170 : * we do need to scan or join the target. (NOTE: we do not bother
171 : * to check for namespace conflict; we assume that the namespace was
172 : * initially empty in these cases.)
173 : *
174 : * Finally, we mark the relation as requiring the permissions specified
175 : * by requiredPerms.
176 : *
177 : * Returns the rangetable index of the target relation.
178 : */
179 : int
180 89850 : setTargetTable(ParseState *pstate, RangeVar *relation,
181 : bool inh, bool alsoSource, AclMode requiredPerms)
182 : {
183 : ParseNamespaceItem *nsitem;
184 :
185 : /*
186 : * ENRs hide tables of the same name, so we need to check for them first.
187 : * In contrast, CTEs don't hide tables (for this purpose).
188 : */
189 168498 : if (relation->schemaname == NULL &&
190 78648 : scanNameSpaceForENR(pstate, relation->relname))
191 6 : ereport(ERROR,
192 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
193 : errmsg("relation \"%s\" cannot be the target of a modifying statement",
194 : relation->relname)));
195 :
196 : /* Close old target; this could only happen for multi-action rules */
197 89844 : if (pstate->p_target_relation != NULL)
198 0 : table_close(pstate->p_target_relation, NoLock);
199 :
200 : /*
201 : * Open target rel and grab suitable lock (which we will hold till end of
202 : * transaction).
203 : *
204 : * free_parsestate() will eventually do the corresponding table_close(),
205 : * but *not* release the lock.
206 : */
207 89844 : pstate->p_target_relation = parserOpenTable(pstate, relation,
208 : RowExclusiveLock);
209 :
210 : /*
211 : * Now build an RTE and a ParseNamespaceItem.
212 : */
213 89824 : nsitem = addRangeTableEntryForRelation(pstate, pstate->p_target_relation,
214 : RowExclusiveLock,
215 : relation->alias, inh, false);
216 :
217 : /* remember the RTE/nsitem as being the query target */
218 89824 : pstate->p_target_nsitem = nsitem;
219 :
220 : /*
221 : * Override addRangeTableEntry's default ACL_SELECT permissions check, and
222 : * instead mark target table as requiring exactly the specified
223 : * permissions.
224 : *
225 : * If we find an explicit reference to the rel later during parse
226 : * analysis, we will add the ACL_SELECT bit back again; see
227 : * markVarForSelectPriv and its callers.
228 : */
229 89824 : nsitem->p_perminfo->requiredPerms = requiredPerms;
230 :
231 : /*
232 : * If UPDATE/DELETE, add table to joinlist and namespace.
233 : */
234 89824 : if (alsoSource)
235 17546 : addNSItemToQuery(pstate, nsitem, true, true, true);
236 :
237 89824 : return nsitem->p_rtindex;
238 : }
239 :
240 : /*
241 : * Extract all not-in-common columns from column lists of a source table
242 : *
243 : * src_nscolumns and src_colnames describe the source table.
244 : *
245 : * *src_colnos initially contains the column numbers of the already-merged
246 : * columns. We add to it the column number of each additional column.
247 : * Also append to *res_colnames the name of each additional column,
248 : * append to *res_colvars a Var for each additional column, and copy the
249 : * columns' nscolumns data into res_nscolumns[] (which is caller-allocated
250 : * space that had better be big enough).
251 : *
252 : * Returns the number of columns added.
253 : */
254 : static int
255 147044 : extractRemainingColumns(ParseState *pstate,
256 : ParseNamespaceColumn *src_nscolumns,
257 : List *src_colnames,
258 : List **src_colnos,
259 : List **res_colnames, List **res_colvars,
260 : ParseNamespaceColumn *res_nscolumns)
261 : {
262 147044 : int colcount = 0;
263 : Bitmapset *prevcols;
264 : int attnum;
265 : ListCell *lc;
266 :
267 : /*
268 : * While we could just test "list_member_int(*src_colnos, attnum)" to
269 : * detect already-merged columns in the loop below, that would be O(N^2)
270 : * for a wide input table. Instead build a bitmapset of just the merged
271 : * USING columns, which we won't add to within the main loop.
272 : */
273 147044 : prevcols = NULL;
274 150364 : foreach(lc, *src_colnos)
275 : {
276 3320 : prevcols = bms_add_member(prevcols, lfirst_int(lc));
277 : }
278 :
279 147044 : attnum = 0;
280 3316382 : foreach(lc, src_colnames)
281 : {
282 3169338 : char *colname = strVal(lfirst(lc));
283 :
284 3169338 : attnum++;
285 : /* Non-dropped and not already merged? */
286 3169338 : if (colname[0] != '\0' && !bms_is_member(attnum, prevcols))
287 : {
288 : /* Yes, so emit it as next output column */
289 3165552 : *src_colnos = lappend_int(*src_colnos, attnum);
290 3165552 : *res_colnames = lappend(*res_colnames, lfirst(lc));
291 3165552 : *res_colvars = lappend(*res_colvars,
292 3165552 : buildVarFromNSColumn(pstate,
293 3165552 : src_nscolumns + attnum - 1));
294 : /* Copy the input relation's nscolumn data for this column */
295 3165552 : res_nscolumns[colcount] = src_nscolumns[attnum - 1];
296 3165552 : colcount++;
297 : }
298 : }
299 147044 : return colcount;
300 : }
301 :
302 : /* transformJoinUsingClause()
303 : * Build a complete ON clause from a partially-transformed USING list.
304 : * We are given lists of nodes representing left and right match columns.
305 : * Result is a transformed qualification expression.
306 : */
307 : static Node *
308 1460 : transformJoinUsingClause(ParseState *pstate,
309 : List *leftVars, List *rightVars)
310 : {
311 : Node *result;
312 1460 : List *andargs = NIL;
313 : ListCell *lvars,
314 : *rvars;
315 :
316 : /*
317 : * We cheat a little bit here by building an untransformed operator tree
318 : * whose leaves are the already-transformed Vars. This requires collusion
319 : * from transformExpr(), which normally could be expected to complain
320 : * about already-transformed subnodes. However, this does mean that we
321 : * have to mark the columns as requiring SELECT privilege for ourselves;
322 : * transformExpr() won't do it.
323 : */
324 3120 : forboth(lvars, leftVars, rvars, rightVars)
325 : {
326 1660 : Var *lvar = (Var *) lfirst(lvars);
327 1660 : Var *rvar = (Var *) lfirst(rvars);
328 : A_Expr *e;
329 :
330 : /* Require read access to the join variables */
331 1660 : markVarForSelectPriv(pstate, lvar);
332 1660 : markVarForSelectPriv(pstate, rvar);
333 :
334 : /* Now create the lvar = rvar join condition */
335 1660 : e = makeSimpleA_Expr(AEXPR_OP, "=",
336 1660 : (Node *) copyObject(lvar), (Node *) copyObject(rvar),
337 : -1);
338 :
339 : /* Prepare to combine into an AND clause, if multiple join columns */
340 1660 : andargs = lappend(andargs, e);
341 : }
342 :
343 : /* Only need an AND if there's more than one join column */
344 1460 : if (list_length(andargs) == 1)
345 1282 : result = (Node *) linitial(andargs);
346 : else
347 178 : result = (Node *) makeBoolExpr(AND_EXPR, andargs, -1);
348 :
349 : /*
350 : * Since the references are already Vars, and are certainly from the input
351 : * relations, we don't have to go through the same pushups that
352 : * transformJoinOnClause() does. Just invoke transformExpr() to fix up
353 : * the operators, and we're done.
354 : */
355 1460 : result = transformExpr(pstate, result, EXPR_KIND_JOIN_USING);
356 :
357 1460 : result = coerce_to_boolean(pstate, result, "JOIN/USING");
358 :
359 1460 : return result;
360 : }
361 :
362 : /* transformJoinOnClause()
363 : * Transform the qual conditions for JOIN/ON.
364 : * Result is a transformed qualification expression.
365 : */
366 : static Node *
367 71794 : transformJoinOnClause(ParseState *pstate, JoinExpr *j, List *namespace)
368 : {
369 : Node *result;
370 : List *save_namespace;
371 :
372 : /*
373 : * The namespace that the join expression should see is just the two
374 : * subtrees of the JOIN plus any outer references from upper pstate
375 : * levels. Temporarily set this pstate's namespace accordingly. (We need
376 : * not check for refname conflicts, because transformFromClauseItem()
377 : * already did.) All namespace items are marked visible regardless of
378 : * LATERAL state.
379 : */
380 71794 : setNamespaceLateralState(namespace, false, true);
381 :
382 71794 : save_namespace = pstate->p_namespace;
383 71794 : pstate->p_namespace = namespace;
384 :
385 71794 : result = transformWhereClause(pstate, j->quals,
386 : EXPR_KIND_JOIN_ON, "JOIN/ON");
387 :
388 71776 : pstate->p_namespace = save_namespace;
389 :
390 71776 : return result;
391 : }
392 :
393 : /*
394 : * transformTableEntry --- transform a RangeVar (simple relation reference)
395 : */
396 : static ParseNamespaceItem *
397 334540 : transformTableEntry(ParseState *pstate, RangeVar *r)
398 : {
399 : /* addRangeTableEntry does all the work */
400 334540 : return addRangeTableEntry(pstate, r, r->alias, r->inh, true);
401 : }
402 :
403 : /*
404 : * transformRangeSubselect --- transform a sub-SELECT appearing in FROM
405 : */
406 : static ParseNamespaceItem *
407 13884 : transformRangeSubselect(ParseState *pstate, RangeSubselect *r)
408 : {
409 : Query *query;
410 :
411 : /*
412 : * Set p_expr_kind to show this parse level is recursing to a subselect.
413 : * We can't be nested within any expression, so don't need save-restore
414 : * logic here.
415 : */
416 : Assert(pstate->p_expr_kind == EXPR_KIND_NONE);
417 13884 : pstate->p_expr_kind = EXPR_KIND_FROM_SUBSELECT;
418 :
419 : /*
420 : * If the subselect is LATERAL, make lateral_only names of this level
421 : * visible to it. (LATERAL can't nest within a single pstate level, so we
422 : * don't need save/restore logic here.)
423 : */
424 : Assert(!pstate->p_lateral_active);
425 13884 : pstate->p_lateral_active = r->lateral;
426 :
427 : /*
428 : * Analyze and transform the subquery. Note that if the subquery doesn't
429 : * have an alias, it can't be explicitly selected for locking, but locking
430 : * might still be required (if there is an all-tables locking clause).
431 : */
432 13884 : query = parse_sub_analyze(r->subquery, pstate, NULL,
433 13884 : isLockedRefname(pstate,
434 13884 : r->alias == NULL ? NULL :
435 13746 : r->alias->aliasname),
436 : true);
437 :
438 : /* Restore state */
439 13776 : pstate->p_lateral_active = false;
440 13776 : pstate->p_expr_kind = EXPR_KIND_NONE;
441 :
442 : /*
443 : * Check that we got a SELECT. Anything else should be impossible given
444 : * restrictions of the grammar, but check anyway.
445 : */
446 13776 : if (!IsA(query, Query) ||
447 13776 : query->commandType != CMD_SELECT)
448 0 : elog(ERROR, "unexpected non-SELECT command in subquery in FROM");
449 :
450 : /*
451 : * OK, build an RTE and nsitem for the subquery.
452 : */
453 27546 : return addRangeTableEntryForSubquery(pstate,
454 : query,
455 : r->alias,
456 13776 : r->lateral,
457 : true);
458 : }
459 :
460 :
461 : /*
462 : * transformRangeFunction --- transform a function call appearing in FROM
463 : */
464 : static ParseNamespaceItem *
465 38966 : transformRangeFunction(ParseState *pstate, RangeFunction *r)
466 : {
467 38966 : List *funcexprs = NIL;
468 38966 : List *funcnames = NIL;
469 38966 : List *coldeflists = NIL;
470 : bool is_lateral;
471 : ListCell *lc;
472 :
473 : /*
474 : * We make lateral_only names of this level visible, whether or not the
475 : * RangeFunction is explicitly marked LATERAL. This is needed for SQL
476 : * spec compliance in the case of UNNEST(), and seems useful on
477 : * convenience grounds for all functions in FROM.
478 : *
479 : * (LATERAL can't nest within a single pstate level, so we don't need
480 : * save/restore logic here.)
481 : */
482 : Assert(!pstate->p_lateral_active);
483 38966 : pstate->p_lateral_active = true;
484 :
485 : /*
486 : * Transform the raw expressions.
487 : *
488 : * While transforming, also save function names for possible use as alias
489 : * and column names. We use the same transformation rules as for a SELECT
490 : * output expression. For a FuncCall node, the result will be the
491 : * function name, but it is possible for the grammar to hand back other
492 : * node types.
493 : *
494 : * We have to get this info now, because FigureColname only works on raw
495 : * parsetrees. Actually deciding what to do with the names is left up to
496 : * addRangeTableEntryForFunction.
497 : *
498 : * Likewise, collect column definition lists if there were any. But
499 : * complain if we find one here and the RangeFunction has one too.
500 : */
501 77950 : foreach(lc, r->functions)
502 : {
503 39152 : List *pair = (List *) lfirst(lc);
504 : Node *fexpr;
505 : List *coldeflist;
506 : Node *newfexpr;
507 : Node *last_srf;
508 :
509 : /* Disassemble the function-call/column-def-list pairs */
510 : Assert(list_length(pair) == 2);
511 39152 : fexpr = (Node *) linitial(pair);
512 39152 : coldeflist = (List *) lsecond(pair);
513 :
514 : /*
515 : * If we find a function call unnest() with more than one argument and
516 : * no special decoration, transform it into separate unnest() calls on
517 : * each argument. This is a kluge, for sure, but it's less nasty than
518 : * other ways of implementing the SQL-standard UNNEST() syntax.
519 : *
520 : * If there is any decoration (including a coldeflist), we don't
521 : * transform, which probably means a no-such-function error later. We
522 : * could alternatively throw an error right now, but that doesn't seem
523 : * tremendously helpful. If someone is using any such decoration,
524 : * then they're not using the SQL-standard syntax, and they're more
525 : * likely expecting an un-tweaked function call.
526 : *
527 : * Note: the transformation changes a non-schema-qualified unnest()
528 : * function name into schema-qualified pg_catalog.unnest(). This
529 : * choice is also a bit debatable, but it seems reasonable to force
530 : * use of built-in unnest() when we make this transformation.
531 : */
532 39152 : if (IsA(fexpr, FuncCall))
533 : {
534 39014 : FuncCall *fc = (FuncCall *) fexpr;
535 :
536 39014 : if (list_length(fc->funcname) == 1 &&
537 28720 : strcmp(strVal(linitial(fc->funcname)), "unnest") == 0 &&
538 2524 : list_length(fc->args) > 1 &&
539 66 : fc->agg_order == NIL &&
540 66 : fc->agg_filter == NULL &&
541 66 : fc->over == NULL &&
542 66 : !fc->agg_star &&
543 66 : !fc->agg_distinct &&
544 66 : !fc->func_variadic &&
545 : coldeflist == NIL)
546 : {
547 : ListCell *lc2;
548 :
549 240 : foreach(lc2, fc->args)
550 : {
551 174 : Node *arg = (Node *) lfirst(lc2);
552 : FuncCall *newfc;
553 :
554 174 : last_srf = pstate->p_last_srf;
555 :
556 174 : newfc = makeFuncCall(SystemFuncName("unnest"),
557 174 : list_make1(arg),
558 : COERCE_EXPLICIT_CALL,
559 : fc->location);
560 :
561 174 : newfexpr = transformExpr(pstate, (Node *) newfc,
562 : EXPR_KIND_FROM_FUNCTION);
563 :
564 : /* nodeFunctionscan.c requires SRFs to be at top level */
565 174 : if (pstate->p_last_srf != last_srf &&
566 174 : pstate->p_last_srf != newfexpr)
567 0 : ereport(ERROR,
568 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
569 : errmsg("set-returning functions must appear at top level of FROM"),
570 : parser_errposition(pstate,
571 : exprLocation(pstate->p_last_srf))));
572 :
573 174 : funcexprs = lappend(funcexprs, newfexpr);
574 :
575 174 : funcnames = lappend(funcnames,
576 174 : FigureColname((Node *) newfc));
577 :
578 : /* coldeflist is empty, so no error is possible */
579 :
580 174 : coldeflists = lappend(coldeflists, coldeflist);
581 : }
582 66 : continue; /* done with this function item */
583 : }
584 : }
585 :
586 : /* normal case ... */
587 39086 : last_srf = pstate->p_last_srf;
588 :
589 39086 : newfexpr = transformExpr(pstate, fexpr,
590 : EXPR_KIND_FROM_FUNCTION);
591 :
592 : /* nodeFunctionscan.c requires SRFs to be at top level */
593 38924 : if (pstate->p_last_srf != last_srf &&
594 33942 : pstate->p_last_srf != newfexpr)
595 6 : ereport(ERROR,
596 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
597 : errmsg("set-returning functions must appear at top level of FROM"),
598 : parser_errposition(pstate,
599 : exprLocation(pstate->p_last_srf))));
600 :
601 38918 : funcexprs = lappend(funcexprs, newfexpr);
602 :
603 38918 : funcnames = lappend(funcnames,
604 38918 : FigureColname(fexpr));
605 :
606 38918 : if (coldeflist && r->coldeflist)
607 0 : ereport(ERROR,
608 : (errcode(ERRCODE_SYNTAX_ERROR),
609 : errmsg("multiple column definition lists are not allowed for the same function"),
610 : parser_errposition(pstate,
611 : exprLocation((Node *) r->coldeflist))));
612 :
613 38918 : coldeflists = lappend(coldeflists, coldeflist);
614 : }
615 :
616 38798 : pstate->p_lateral_active = false;
617 :
618 : /*
619 : * We must assign collations now so that the RTE exposes correct collation
620 : * info for Vars created from it.
621 : */
622 38798 : assign_list_collations(pstate, funcexprs);
623 :
624 : /*
625 : * Install the top-level coldeflist if there was one (we already checked
626 : * that there was no conflicting per-function coldeflist).
627 : *
628 : * We only allow this when there's a single function (even after UNNEST
629 : * expansion) and no WITH ORDINALITY. The reason for the latter
630 : * restriction is that it's not real clear whether the ordinality column
631 : * should be in the coldeflist, and users are too likely to make mistakes
632 : * in one direction or the other. Putting the coldeflist inside ROWS
633 : * FROM() is much clearer in this case.
634 : */
635 38798 : if (r->coldeflist)
636 : {
637 738 : if (list_length(funcexprs) != 1)
638 : {
639 0 : if (r->is_rowsfrom)
640 0 : ereport(ERROR,
641 : (errcode(ERRCODE_SYNTAX_ERROR),
642 : errmsg("ROWS FROM() with multiple functions cannot have a column definition list"),
643 : errhint("Put a separate column definition list for each function inside ROWS FROM()."),
644 : parser_errposition(pstate,
645 : exprLocation((Node *) r->coldeflist))));
646 : else
647 0 : ereport(ERROR,
648 : (errcode(ERRCODE_SYNTAX_ERROR),
649 : errmsg("UNNEST() with multiple arguments cannot have a column definition list"),
650 : errhint("Use separate UNNEST() calls inside ROWS FROM(), and attach a column definition list to each one."),
651 : parser_errposition(pstate,
652 : exprLocation((Node *) r->coldeflist))));
653 : }
654 738 : if (r->ordinality)
655 0 : ereport(ERROR,
656 : (errcode(ERRCODE_SYNTAX_ERROR),
657 : errmsg("WITH ORDINALITY cannot be used with a column definition list"),
658 : errhint("Put the column definition list inside ROWS FROM()."),
659 : parser_errposition(pstate,
660 : exprLocation((Node *) r->coldeflist))));
661 :
662 738 : coldeflists = list_make1(r->coldeflist);
663 : }
664 :
665 : /*
666 : * Mark the RTE as LATERAL if the user said LATERAL explicitly, or if
667 : * there are any lateral cross-references in it.
668 : */
669 38798 : is_lateral = r->lateral || contain_vars_of_level((Node *) funcexprs, 0);
670 :
671 : /*
672 : * OK, build an RTE and nsitem for the function.
673 : */
674 38798 : return addRangeTableEntryForFunction(pstate,
675 : funcnames, funcexprs, coldeflists,
676 : r, is_lateral, true);
677 : }
678 :
679 : /*
680 : * transformRangeTableFunc -
681 : * Transform a raw RangeTableFunc into TableFunc.
682 : *
683 : * Transform the namespace clauses, the document-generating expression, the
684 : * row-generating expression, the column-generating expressions, and the
685 : * default value expressions.
686 : */
687 : static ParseNamespaceItem *
688 220 : transformRangeTableFunc(ParseState *pstate, RangeTableFunc *rtf)
689 : {
690 220 : TableFunc *tf = makeNode(TableFunc);
691 : const char *constructName;
692 : Oid docType;
693 : bool is_lateral;
694 : ListCell *col;
695 : char **names;
696 : int colno;
697 :
698 : /* Currently only XMLTABLE is supported */
699 220 : constructName = "XMLTABLE";
700 220 : docType = XMLOID;
701 :
702 : /*
703 : * We make lateral_only names of this level visible, whether or not the
704 : * RangeTableFunc is explicitly marked LATERAL. This is needed for SQL
705 : * spec compliance and seems useful on convenience grounds for all
706 : * functions in FROM.
707 : *
708 : * (LATERAL can't nest within a single pstate level, so we don't need
709 : * save/restore logic here.)
710 : */
711 : Assert(!pstate->p_lateral_active);
712 220 : pstate->p_lateral_active = true;
713 :
714 : /* Transform and apply typecast to the row-generating expression ... */
715 : Assert(rtf->rowexpr != NULL);
716 220 : tf->rowexpr = coerce_to_specific_type(pstate,
717 : transformExpr(pstate, rtf->rowexpr, EXPR_KIND_FROM_FUNCTION),
718 : TEXTOID,
719 : constructName);
720 220 : assign_expr_collations(pstate, tf->rowexpr);
721 :
722 : /* ... and to the document itself */
723 : Assert(rtf->docexpr != NULL);
724 220 : tf->docexpr = coerce_to_specific_type(pstate,
725 : transformExpr(pstate, rtf->docexpr, EXPR_KIND_FROM_FUNCTION),
726 : docType,
727 : constructName);
728 220 : assign_expr_collations(pstate, tf->docexpr);
729 :
730 : /* undef ordinality column number */
731 220 : tf->ordinalitycol = -1;
732 :
733 : /* Process column specs */
734 220 : names = palloc(sizeof(char *) * list_length(rtf->columns));
735 :
736 220 : colno = 0;
737 970 : foreach(col, rtf->columns)
738 : {
739 750 : RangeTableFuncCol *rawc = (RangeTableFuncCol *) lfirst(col);
740 : Oid typid;
741 : int32 typmod;
742 : Node *colexpr;
743 : Node *coldefexpr;
744 : int j;
745 :
746 750 : tf->colnames = lappend(tf->colnames,
747 750 : makeString(pstrdup(rawc->colname)));
748 :
749 : /*
750 : * Determine the type and typmod for the new column. FOR ORDINALITY
751 : * columns are INTEGER per spec; the others are user-specified.
752 : */
753 750 : if (rawc->for_ordinality)
754 : {
755 62 : if (tf->ordinalitycol != -1)
756 0 : ereport(ERROR,
757 : (errcode(ERRCODE_SYNTAX_ERROR),
758 : errmsg("only one FOR ORDINALITY column is allowed"),
759 : parser_errposition(pstate, rawc->location)));
760 :
761 62 : typid = INT4OID;
762 62 : typmod = -1;
763 62 : tf->ordinalitycol = colno;
764 : }
765 : else
766 : {
767 688 : if (rawc->typeName->setof)
768 0 : ereport(ERROR,
769 : (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
770 : errmsg("column \"%s\" cannot be declared SETOF",
771 : rawc->colname),
772 : parser_errposition(pstate, rawc->location)));
773 :
774 688 : typenameTypeIdAndMod(pstate, rawc->typeName,
775 : &typid, &typmod);
776 : }
777 :
778 750 : tf->coltypes = lappend_oid(tf->coltypes, typid);
779 750 : tf->coltypmods = lappend_int(tf->coltypmods, typmod);
780 750 : tf->colcollations = lappend_oid(tf->colcollations,
781 : get_typcollation(typid));
782 :
783 : /* Transform the PATH and DEFAULT expressions */
784 750 : if (rawc->colexpr)
785 : {
786 490 : colexpr = coerce_to_specific_type(pstate,
787 : transformExpr(pstate, rawc->colexpr,
788 : EXPR_KIND_FROM_FUNCTION),
789 : TEXTOID,
790 : constructName);
791 490 : assign_expr_collations(pstate, colexpr);
792 : }
793 : else
794 260 : colexpr = NULL;
795 :
796 750 : if (rawc->coldefexpr)
797 : {
798 56 : coldefexpr = coerce_to_specific_type_typmod(pstate,
799 : transformExpr(pstate, rawc->coldefexpr,
800 : EXPR_KIND_FROM_FUNCTION),
801 : typid, typmod,
802 : constructName);
803 56 : assign_expr_collations(pstate, coldefexpr);
804 : }
805 : else
806 694 : coldefexpr = NULL;
807 :
808 750 : tf->colexprs = lappend(tf->colexprs, colexpr);
809 750 : tf->coldefexprs = lappend(tf->coldefexprs, coldefexpr);
810 :
811 750 : if (rawc->is_not_null)
812 56 : tf->notnulls = bms_add_member(tf->notnulls, colno);
813 :
814 : /* make sure column names are unique */
815 2534 : for (j = 0; j < colno; j++)
816 1784 : if (strcmp(names[j], rawc->colname) == 0)
817 0 : ereport(ERROR,
818 : (errcode(ERRCODE_SYNTAX_ERROR),
819 : errmsg("column name \"%s\" is not unique",
820 : rawc->colname),
821 : parser_errposition(pstate, rawc->location)));
822 750 : names[colno] = rawc->colname;
823 :
824 750 : colno++;
825 : }
826 220 : pfree(names);
827 :
828 : /* Namespaces, if any, also need to be transformed */
829 220 : if (rtf->namespaces != NIL)
830 : {
831 : ListCell *ns;
832 : ListCell *lc2;
833 20 : List *ns_uris = NIL;
834 20 : List *ns_names = NIL;
835 20 : bool default_ns_seen = false;
836 :
837 40 : foreach(ns, rtf->namespaces)
838 : {
839 20 : ResTarget *r = (ResTarget *) lfirst(ns);
840 : Node *ns_uri;
841 :
842 : Assert(IsA(r, ResTarget));
843 20 : ns_uri = transformExpr(pstate, r->val, EXPR_KIND_FROM_FUNCTION);
844 20 : ns_uri = coerce_to_specific_type(pstate, ns_uri,
845 : TEXTOID, constructName);
846 20 : assign_expr_collations(pstate, ns_uri);
847 20 : ns_uris = lappend(ns_uris, ns_uri);
848 :
849 : /* Verify consistency of name list: no dupes, only one DEFAULT */
850 20 : if (r->name != NULL)
851 : {
852 14 : foreach(lc2, ns_names)
853 : {
854 0 : String *ns_node = lfirst_node(String, lc2);
855 :
856 0 : if (ns_node == NULL)
857 0 : continue;
858 0 : if (strcmp(strVal(ns_node), r->name) == 0)
859 0 : ereport(ERROR,
860 : (errcode(ERRCODE_SYNTAX_ERROR),
861 : errmsg("namespace name \"%s\" is not unique",
862 : r->name),
863 : parser_errposition(pstate, r->location)));
864 : }
865 : }
866 : else
867 : {
868 6 : if (default_ns_seen)
869 0 : ereport(ERROR,
870 : (errcode(ERRCODE_SYNTAX_ERROR),
871 : errmsg("only one default namespace is allowed"),
872 : parser_errposition(pstate, r->location)));
873 6 : default_ns_seen = true;
874 : }
875 :
876 : /* We represent DEFAULT by a null pointer */
877 20 : ns_names = lappend(ns_names,
878 20 : r->name ? makeString(r->name) : NULL);
879 : }
880 :
881 20 : tf->ns_uris = ns_uris;
882 20 : tf->ns_names = ns_names;
883 : }
884 :
885 220 : tf->location = rtf->location;
886 :
887 220 : pstate->p_lateral_active = false;
888 :
889 : /*
890 : * Mark the RTE as LATERAL if the user said LATERAL explicitly, or if
891 : * there are any lateral cross-references in it.
892 : */
893 220 : is_lateral = rtf->lateral || contain_vars_of_level((Node *) tf, 0);
894 :
895 220 : return addRangeTableEntryForTableFunc(pstate,
896 : tf, rtf->alias, is_lateral, true);
897 : }
898 :
899 : /*
900 : * transformRangeTableSample --- transform a TABLESAMPLE clause
901 : *
902 : * Caller has already transformed rts->relation, we just have to validate
903 : * the remaining fields and create a TableSampleClause node.
904 : */
905 : static TableSampleClause *
906 242 : transformRangeTableSample(ParseState *pstate, RangeTableSample *rts)
907 : {
908 : TableSampleClause *tablesample;
909 : Oid handlerOid;
910 : Oid funcargtypes[1];
911 : TsmRoutine *tsm;
912 : List *fargs;
913 : ListCell *larg,
914 : *ltyp;
915 :
916 : /*
917 : * To validate the sample method name, look up the handler function, which
918 : * has the same name, one dummy INTERNAL argument, and a result type of
919 : * tsm_handler. (Note: tablesample method names are not schema-qualified
920 : * in the SQL standard; but since they are just functions to us, we allow
921 : * schema qualification to resolve any potential ambiguity.)
922 : */
923 242 : funcargtypes[0] = INTERNALOID;
924 :
925 242 : handlerOid = LookupFuncName(rts->method, 1, funcargtypes, true);
926 :
927 : /* we want error to complain about no-such-method, not no-such-function */
928 242 : if (!OidIsValid(handlerOid))
929 6 : ereport(ERROR,
930 : (errcode(ERRCODE_UNDEFINED_OBJECT),
931 : errmsg("tablesample method %s does not exist",
932 : NameListToString(rts->method)),
933 : parser_errposition(pstate, rts->location)));
934 :
935 : /* check that handler has correct return type */
936 236 : if (get_func_rettype(handlerOid) != TSM_HANDLEROID)
937 0 : ereport(ERROR,
938 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
939 : errmsg("function %s must return type %s",
940 : NameListToString(rts->method), "tsm_handler"),
941 : parser_errposition(pstate, rts->location)));
942 :
943 : /* OK, run the handler to get TsmRoutine, for argument type info */
944 236 : tsm = GetTsmRoutine(handlerOid);
945 :
946 236 : tablesample = makeNode(TableSampleClause);
947 236 : tablesample->tsmhandler = handlerOid;
948 :
949 : /* check user provided the expected number of arguments */
950 236 : if (list_length(rts->args) != list_length(tsm->parameterTypes))
951 0 : ereport(ERROR,
952 : (errcode(ERRCODE_INVALID_TABLESAMPLE_ARGUMENT),
953 : errmsg_plural("tablesample method %s requires %d argument, not %d",
954 : "tablesample method %s requires %d arguments, not %d",
955 : list_length(tsm->parameterTypes),
956 : NameListToString(rts->method),
957 : list_length(tsm->parameterTypes),
958 : list_length(rts->args)),
959 : parser_errposition(pstate, rts->location)));
960 :
961 : /*
962 : * Transform the arguments, typecasting them as needed. Note we must also
963 : * assign collations now, because assign_query_collations() doesn't
964 : * examine any substructure of RTEs.
965 : */
966 236 : fargs = NIL;
967 472 : forboth(larg, rts->args, ltyp, tsm->parameterTypes)
968 : {
969 236 : Node *arg = (Node *) lfirst(larg);
970 236 : Oid argtype = lfirst_oid(ltyp);
971 :
972 236 : arg = transformExpr(pstate, arg, EXPR_KIND_FROM_FUNCTION);
973 236 : arg = coerce_to_specific_type(pstate, arg, argtype, "TABLESAMPLE");
974 236 : assign_expr_collations(pstate, arg);
975 236 : fargs = lappend(fargs, arg);
976 : }
977 236 : tablesample->args = fargs;
978 :
979 : /* Process REPEATABLE (seed) */
980 236 : if (rts->repeatable != NULL)
981 : {
982 : Node *arg;
983 :
984 102 : if (!tsm->repeatable_across_queries)
985 4 : ereport(ERROR,
986 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
987 : errmsg("tablesample method %s does not support REPEATABLE",
988 : NameListToString(rts->method)),
989 : parser_errposition(pstate, rts->location)));
990 :
991 98 : arg = transformExpr(pstate, rts->repeatable, EXPR_KIND_FROM_FUNCTION);
992 98 : arg = coerce_to_specific_type(pstate, arg, FLOAT8OID, "REPEATABLE");
993 98 : assign_expr_collations(pstate, arg);
994 98 : tablesample->repeatable = (Expr *) arg;
995 : }
996 : else
997 134 : tablesample->repeatable = NULL;
998 :
999 232 : return tablesample;
1000 : }
1001 :
1002 : /*
1003 : * getNSItemForSpecialRelationTypes
1004 : *
1005 : * If given RangeVar refers to a CTE or an EphemeralNamedRelation,
1006 : * build and return an appropriate ParseNamespaceItem, otherwise return NULL
1007 : */
1008 : static ParseNamespaceItem *
1009 340640 : getNSItemForSpecialRelationTypes(ParseState *pstate, RangeVar *rv)
1010 : {
1011 : ParseNamespaceItem *nsitem;
1012 : CommonTableExpr *cte;
1013 : Index levelsup;
1014 :
1015 : /*
1016 : * if it is a qualified name, it can't be a CTE or tuplestore reference
1017 : */
1018 340640 : if (rv->schemaname)
1019 168094 : return NULL;
1020 :
1021 172546 : cte = scanNameSpaceForCTE(pstate, rv->relname, &levelsup);
1022 172546 : if (cte)
1023 5656 : nsitem = addRangeTableEntryForCTE(pstate, cte, levelsup, rv, true);
1024 166890 : else if (scanNameSpaceForENR(pstate, rv->relname))
1025 444 : nsitem = addRangeTableEntryForENR(pstate, rv, true);
1026 : else
1027 166446 : nsitem = NULL;
1028 :
1029 172534 : return nsitem;
1030 : }
1031 :
1032 : /*
1033 : * transformFromClauseItem -
1034 : * Transform a FROM-clause item, adding any required entries to the
1035 : * range table list being built in the ParseState, and return the
1036 : * transformed item ready to include in the joinlist. Also build a
1037 : * ParseNamespaceItem list describing the names exposed by this item.
1038 : * This routine can recurse to handle SQL92 JOIN expressions.
1039 : *
1040 : * The function return value is the node to add to the jointree (a
1041 : * RangeTblRef or JoinExpr). Additional output parameters are:
1042 : *
1043 : * *top_nsitem: receives the ParseNamespaceItem directly corresponding to the
1044 : * jointree item. (This is only used during internal recursion, not by
1045 : * outside callers.)
1046 : *
1047 : * *namespace: receives a List of ParseNamespaceItems for the RTEs exposed
1048 : * as table/column names by this item. (The lateral_only flags in these items
1049 : * are indeterminate and should be explicitly set by the caller before use.)
1050 : */
1051 : static Node *
1052 467540 : transformFromClauseItem(ParseState *pstate, Node *n,
1053 : ParseNamespaceItem **top_nsitem,
1054 : List **namespace)
1055 : {
1056 : /* Guard against stack overflow due to overly deep subtree */
1057 467540 : check_stack_depth();
1058 :
1059 467540 : if (IsA(n, RangeVar))
1060 : {
1061 : /* Plain relation reference, or perhaps a CTE reference */
1062 340640 : RangeVar *rv = (RangeVar *) n;
1063 : RangeTblRef *rtr;
1064 : ParseNamespaceItem *nsitem;
1065 :
1066 : /* Check if it's a CTE or tuplestore reference */
1067 340640 : nsitem = getNSItemForSpecialRelationTypes(pstate, rv);
1068 :
1069 : /* if not found above, must be a table reference */
1070 340628 : if (!nsitem)
1071 334540 : nsitem = transformTableEntry(pstate, rv);
1072 :
1073 340448 : *top_nsitem = nsitem;
1074 340448 : *namespace = list_make1(nsitem);
1075 340448 : rtr = makeNode(RangeTblRef);
1076 340448 : rtr->rtindex = nsitem->p_rtindex;
1077 340448 : return (Node *) rtr;
1078 : }
1079 126900 : else if (IsA(n, RangeSubselect))
1080 : {
1081 : /* sub-SELECT is like a plain relation */
1082 : RangeTblRef *rtr;
1083 : ParseNamespaceItem *nsitem;
1084 :
1085 13884 : nsitem = transformRangeSubselect(pstate, (RangeSubselect *) n);
1086 13770 : *top_nsitem = nsitem;
1087 13770 : *namespace = list_make1(nsitem);
1088 13770 : rtr = makeNode(RangeTblRef);
1089 13770 : rtr->rtindex = nsitem->p_rtindex;
1090 13770 : return (Node *) rtr;
1091 : }
1092 113016 : else if (IsA(n, RangeFunction))
1093 : {
1094 : /* function is like a plain relation */
1095 : RangeTblRef *rtr;
1096 : ParseNamespaceItem *nsitem;
1097 :
1098 38966 : nsitem = transformRangeFunction(pstate, (RangeFunction *) n);
1099 38744 : *top_nsitem = nsitem;
1100 38744 : *namespace = list_make1(nsitem);
1101 38744 : rtr = makeNode(RangeTblRef);
1102 38744 : rtr->rtindex = nsitem->p_rtindex;
1103 38744 : return (Node *) rtr;
1104 : }
1105 74050 : else if (IsA(n, RangeTableFunc))
1106 : {
1107 : /* table function is like a plain relation */
1108 : RangeTblRef *rtr;
1109 : ParseNamespaceItem *nsitem;
1110 :
1111 220 : nsitem = transformRangeTableFunc(pstate, (RangeTableFunc *) n);
1112 214 : *top_nsitem = nsitem;
1113 214 : *namespace = list_make1(nsitem);
1114 214 : rtr = makeNode(RangeTblRef);
1115 214 : rtr->rtindex = nsitem->p_rtindex;
1116 214 : return (Node *) rtr;
1117 : }
1118 73830 : else if (IsA(n, RangeTableSample))
1119 : {
1120 : /* TABLESAMPLE clause (wrapping some other valid FROM node) */
1121 254 : RangeTableSample *rts = (RangeTableSample *) n;
1122 : Node *rel;
1123 : RangeTblEntry *rte;
1124 :
1125 : /* Recursively transform the contained relation */
1126 254 : rel = transformFromClauseItem(pstate, rts->relation,
1127 : top_nsitem, namespace);
1128 254 : rte = (*top_nsitem)->p_rte;
1129 : /* We only support this on plain relations and matviews */
1130 254 : if (rte->rtekind != RTE_RELATION ||
1131 248 : (rte->relkind != RELKIND_RELATION &&
1132 24 : rte->relkind != RELKIND_MATVIEW &&
1133 24 : rte->relkind != RELKIND_PARTITIONED_TABLE))
1134 12 : ereport(ERROR,
1135 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1136 : errmsg("TABLESAMPLE clause can only be applied to tables and materialized views"),
1137 : parser_errposition(pstate, exprLocation(rts->relation))));
1138 :
1139 : /* Transform TABLESAMPLE details and attach to the RTE */
1140 242 : rte->tablesample = transformRangeTableSample(pstate, rts);
1141 232 : return rel;
1142 : }
1143 73576 : else if (IsA(n, JoinExpr))
1144 : {
1145 : /* A newfangled join expression */
1146 73576 : JoinExpr *j = (JoinExpr *) n;
1147 : ParseNamespaceItem *nsitem;
1148 : ParseNamespaceItem *l_nsitem;
1149 : ParseNamespaceItem *r_nsitem;
1150 : List *l_namespace,
1151 : *r_namespace,
1152 : *my_namespace,
1153 : *l_colnames,
1154 : *r_colnames,
1155 : *res_colnames,
1156 : *l_colnos,
1157 : *r_colnos,
1158 : *res_colvars;
1159 : ParseNamespaceColumn *l_nscolumns,
1160 : *r_nscolumns,
1161 : *res_nscolumns;
1162 : int res_colindex;
1163 : bool lateral_ok;
1164 : int sv_namespace_length;
1165 : int k;
1166 :
1167 : /*
1168 : * Recursively process the left subtree, then the right. We must do
1169 : * it in this order for correct visibility of LATERAL references.
1170 : */
1171 73576 : j->larg = transformFromClauseItem(pstate, j->larg,
1172 : &l_nsitem,
1173 : &l_namespace);
1174 :
1175 : /*
1176 : * Make the left-side RTEs available for LATERAL access within the
1177 : * right side, by temporarily adding them to the pstate's namespace
1178 : * list. Per SQL:2008, if the join type is not INNER or LEFT then the
1179 : * left-side names must still be exposed, but it's an error to
1180 : * reference them. (Stupid design, but that's what it says.) Hence,
1181 : * we always push them into the namespace, but mark them as not
1182 : * lateral_ok if the jointype is wrong.
1183 : *
1184 : * Notice that we don't require the merged namespace list to be
1185 : * conflict-free. See the comments for scanNameSpaceForRefname().
1186 : */
1187 73576 : lateral_ok = (j->jointype == JOIN_INNER || j->jointype == JOIN_LEFT);
1188 73576 : setNamespaceLateralState(l_namespace, true, lateral_ok);
1189 :
1190 73576 : sv_namespace_length = list_length(pstate->p_namespace);
1191 73576 : pstate->p_namespace = list_concat(pstate->p_namespace, l_namespace);
1192 :
1193 : /* And now we can process the RHS */
1194 73576 : j->rarg = transformFromClauseItem(pstate, j->rarg,
1195 : &r_nsitem,
1196 : &r_namespace);
1197 :
1198 : /* Remove the left-side RTEs from the namespace list again */
1199 73540 : pstate->p_namespace = list_truncate(pstate->p_namespace,
1200 : sv_namespace_length);
1201 :
1202 : /*
1203 : * Check for conflicting refnames in left and right subtrees. Must do
1204 : * this because higher levels will assume I hand back a self-
1205 : * consistent namespace list.
1206 : */
1207 73540 : checkNameSpaceConflicts(pstate, l_namespace, r_namespace);
1208 :
1209 : /*
1210 : * Generate combined namespace info for possible use below.
1211 : */
1212 73540 : my_namespace = list_concat(l_namespace, r_namespace);
1213 :
1214 : /*
1215 : * We'll work from the nscolumns data and eref alias column names for
1216 : * each of the input nsitems. Note that these include dropped
1217 : * columns, which is helpful because we can keep track of physical
1218 : * input column numbers more easily.
1219 : */
1220 73540 : l_nscolumns = l_nsitem->p_nscolumns;
1221 73540 : l_colnames = l_nsitem->p_names->colnames;
1222 73540 : r_nscolumns = r_nsitem->p_nscolumns;
1223 73540 : r_colnames = r_nsitem->p_names->colnames;
1224 :
1225 : /*
1226 : * Natural join does not explicitly specify columns; must generate
1227 : * columns to join. Need to run through the list of columns from each
1228 : * table or join result and match up the column names. Use the first
1229 : * table, and check every column in the second table for a match.
1230 : * (We'll check that the matches were unique later on.) The result of
1231 : * this step is a list of column names just like an explicitly-written
1232 : * USING list.
1233 : */
1234 73540 : if (j->isNatural)
1235 : {
1236 258 : List *rlist = NIL;
1237 : ListCell *lx,
1238 : *rx;
1239 :
1240 : Assert(j->usingClause == NIL); /* shouldn't have USING() too */
1241 :
1242 1140 : foreach(lx, l_colnames)
1243 : {
1244 882 : char *l_colname = strVal(lfirst(lx));
1245 882 : String *m_name = NULL;
1246 :
1247 882 : if (l_colname[0] == '\0')
1248 12 : continue; /* ignore dropped columns */
1249 :
1250 2412 : foreach(rx, r_colnames)
1251 : {
1252 1848 : char *r_colname = strVal(lfirst(rx));
1253 :
1254 1848 : if (strcmp(l_colname, r_colname) == 0)
1255 : {
1256 306 : m_name = makeString(l_colname);
1257 306 : break;
1258 : }
1259 : }
1260 :
1261 : /* matched a right column? then keep as join column... */
1262 870 : if (m_name != NULL)
1263 306 : rlist = lappend(rlist, m_name);
1264 : }
1265 :
1266 258 : j->usingClause = rlist;
1267 : }
1268 :
1269 : /*
1270 : * If a USING clause alias was specified, save the USING columns as
1271 : * its column list.
1272 : */
1273 73540 : if (j->join_using_alias)
1274 84 : j->join_using_alias->colnames = j->usingClause;
1275 :
1276 : /*
1277 : * Now transform the join qualifications, if any.
1278 : */
1279 73540 : l_colnos = NIL;
1280 73540 : r_colnos = NIL;
1281 73540 : res_colnames = NIL;
1282 73540 : res_colvars = NIL;
1283 :
1284 : /* this may be larger than needed, but it's not worth being exact */
1285 : res_nscolumns = (ParseNamespaceColumn *)
1286 73540 : palloc0((list_length(l_colnames) + list_length(r_colnames)) *
1287 : sizeof(ParseNamespaceColumn));
1288 73540 : res_colindex = 0;
1289 :
1290 73540 : if (j->usingClause)
1291 : {
1292 : /*
1293 : * JOIN/USING (or NATURAL JOIN, as transformed above). Transform
1294 : * the list into an explicit ON-condition.
1295 : */
1296 1460 : List *ucols = j->usingClause;
1297 1460 : List *l_usingvars = NIL;
1298 1460 : List *r_usingvars = NIL;
1299 : ListCell *ucol;
1300 :
1301 : Assert(j->quals == NULL); /* shouldn't have ON() too */
1302 :
1303 3120 : foreach(ucol, ucols)
1304 : {
1305 1660 : char *u_colname = strVal(lfirst(ucol));
1306 : ListCell *col;
1307 : int ndx;
1308 1660 : int l_index = -1;
1309 1660 : int r_index = -1;
1310 : Var *l_colvar,
1311 : *r_colvar;
1312 :
1313 : Assert(u_colname[0] != '\0');
1314 :
1315 : /* Check for USING(foo,foo) */
1316 1888 : foreach(col, res_colnames)
1317 : {
1318 228 : char *res_colname = strVal(lfirst(col));
1319 :
1320 228 : if (strcmp(res_colname, u_colname) == 0)
1321 0 : ereport(ERROR,
1322 : (errcode(ERRCODE_DUPLICATE_COLUMN),
1323 : errmsg("column name \"%s\" appears more than once in USING clause",
1324 : u_colname)));
1325 : }
1326 :
1327 : /* Find it in left input */
1328 1660 : ndx = 0;
1329 8134 : foreach(col, l_colnames)
1330 : {
1331 6474 : char *l_colname = strVal(lfirst(col));
1332 :
1333 6474 : if (strcmp(l_colname, u_colname) == 0)
1334 : {
1335 1660 : if (l_index >= 0)
1336 0 : ereport(ERROR,
1337 : (errcode(ERRCODE_AMBIGUOUS_COLUMN),
1338 : errmsg("common column name \"%s\" appears more than once in left table",
1339 : u_colname)));
1340 1660 : l_index = ndx;
1341 : }
1342 6474 : ndx++;
1343 : }
1344 1660 : if (l_index < 0)
1345 0 : ereport(ERROR,
1346 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1347 : errmsg("column \"%s\" specified in USING clause does not exist in left table",
1348 : u_colname)));
1349 1660 : l_colnos = lappend_int(l_colnos, l_index + 1);
1350 :
1351 : /* Find it in right input */
1352 1660 : ndx = 0;
1353 8034 : foreach(col, r_colnames)
1354 : {
1355 6374 : char *r_colname = strVal(lfirst(col));
1356 :
1357 6374 : if (strcmp(r_colname, u_colname) == 0)
1358 : {
1359 1660 : if (r_index >= 0)
1360 0 : ereport(ERROR,
1361 : (errcode(ERRCODE_AMBIGUOUS_COLUMN),
1362 : errmsg("common column name \"%s\" appears more than once in right table",
1363 : u_colname)));
1364 1660 : r_index = ndx;
1365 : }
1366 6374 : ndx++;
1367 : }
1368 1660 : if (r_index < 0)
1369 0 : ereport(ERROR,
1370 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1371 : errmsg("column \"%s\" specified in USING clause does not exist in right table",
1372 : u_colname)));
1373 1660 : r_colnos = lappend_int(r_colnos, r_index + 1);
1374 :
1375 : /* Build Vars to use in the generated JOIN ON clause */
1376 1660 : l_colvar = buildVarFromNSColumn(pstate, l_nscolumns + l_index);
1377 1660 : l_usingvars = lappend(l_usingvars, l_colvar);
1378 1660 : r_colvar = buildVarFromNSColumn(pstate, r_nscolumns + r_index);
1379 1660 : r_usingvars = lappend(r_usingvars, r_colvar);
1380 :
1381 : /*
1382 : * While we're here, add column names to the res_colnames
1383 : * list. It's a bit ugly to do this here while the
1384 : * corresponding res_colvars entries are not made till later,
1385 : * but doing this later would require an additional traversal
1386 : * of the usingClause list.
1387 : */
1388 1660 : res_colnames = lappend(res_colnames, lfirst(ucol));
1389 : }
1390 :
1391 : /* Construct the generated JOIN ON clause */
1392 1460 : j->quals = transformJoinUsingClause(pstate,
1393 : l_usingvars,
1394 : r_usingvars);
1395 : }
1396 72080 : else if (j->quals)
1397 : {
1398 : /* User-written ON-condition; transform it */
1399 71794 : j->quals = transformJoinOnClause(pstate, j, my_namespace);
1400 : }
1401 : else
1402 : {
1403 : /* CROSS JOIN: no quals */
1404 : }
1405 :
1406 : /*
1407 : * If this is an outer join, now mark the appropriate child RTEs as
1408 : * being nulled by this join. We have finished processing the child
1409 : * join expressions as well as the current join's quals, which deal in
1410 : * non-nulled input columns. All future references to those RTEs will
1411 : * see possibly-nulled values, and we should mark generated Vars to
1412 : * account for that. In particular, the join alias Vars that we're
1413 : * about to build should reflect the nulling effects of this join.
1414 : *
1415 : * A difficulty with doing this is that we need the join's RT index,
1416 : * which we don't officially have yet. However, no other RTE can get
1417 : * made between here and the addRangeTableEntryForJoin call, so we can
1418 : * predict what the assignment will be. (Alternatively, we could call
1419 : * addRangeTableEntryForJoin before we have all the data computed, but
1420 : * this seems less ugly.)
1421 : */
1422 73522 : j->rtindex = list_length(pstate->p_rtable) + 1;
1423 :
1424 73522 : switch (j->jointype)
1425 : {
1426 32892 : case JOIN_INNER:
1427 32892 : break;
1428 39296 : case JOIN_LEFT:
1429 39296 : markRelsAsNulledBy(pstate, j->rarg, j->rtindex);
1430 39296 : break;
1431 1008 : case JOIN_FULL:
1432 1008 : markRelsAsNulledBy(pstate, j->larg, j->rtindex);
1433 1008 : markRelsAsNulledBy(pstate, j->rarg, j->rtindex);
1434 1008 : break;
1435 326 : case JOIN_RIGHT:
1436 326 : markRelsAsNulledBy(pstate, j->larg, j->rtindex);
1437 326 : break;
1438 0 : default:
1439 : /* shouldn't see any other types here */
1440 0 : elog(ERROR, "unrecognized join type: %d",
1441 : (int) j->jointype);
1442 : break;
1443 : }
1444 :
1445 : /*
1446 : * Now we can construct join alias expressions for the USING columns.
1447 : */
1448 73522 : if (j->usingClause)
1449 : {
1450 : ListCell *lc1,
1451 : *lc2;
1452 :
1453 : /* Scan the colnos lists to recover info from the previous loop */
1454 3120 : forboth(lc1, l_colnos, lc2, r_colnos)
1455 : {
1456 1660 : int l_index = lfirst_int(lc1) - 1;
1457 1660 : int r_index = lfirst_int(lc2) - 1;
1458 : Var *l_colvar,
1459 : *r_colvar;
1460 : Node *u_colvar;
1461 : ParseNamespaceColumn *res_nscolumn;
1462 :
1463 : /*
1464 : * Note we re-build these Vars: they might have different
1465 : * varnullingrels than the ones made in the previous loop.
1466 : */
1467 1660 : l_colvar = buildVarFromNSColumn(pstate, l_nscolumns + l_index);
1468 1660 : r_colvar = buildVarFromNSColumn(pstate, r_nscolumns + r_index);
1469 :
1470 : /* Construct the join alias Var for this column */
1471 1660 : u_colvar = buildMergedJoinVar(pstate,
1472 : j->jointype,
1473 : l_colvar,
1474 : r_colvar);
1475 1660 : res_colvars = lappend(res_colvars, u_colvar);
1476 :
1477 : /* Construct column's res_nscolumns[] entry */
1478 1660 : res_nscolumn = res_nscolumns + res_colindex;
1479 1660 : res_colindex++;
1480 1660 : if (u_colvar == (Node *) l_colvar)
1481 : {
1482 : /* Merged column is equivalent to left input */
1483 1214 : *res_nscolumn = l_nscolumns[l_index];
1484 : }
1485 446 : else if (u_colvar == (Node *) r_colvar)
1486 : {
1487 : /* Merged column is equivalent to right input */
1488 42 : *res_nscolumn = r_nscolumns[r_index];
1489 : }
1490 : else
1491 : {
1492 : /*
1493 : * Merged column is not semantically equivalent to either
1494 : * input, so it needs to be referenced as the join output
1495 : * column.
1496 : */
1497 404 : res_nscolumn->p_varno = j->rtindex;
1498 404 : res_nscolumn->p_varattno = res_colindex;
1499 404 : res_nscolumn->p_vartype = exprType(u_colvar);
1500 404 : res_nscolumn->p_vartypmod = exprTypmod(u_colvar);
1501 404 : res_nscolumn->p_varcollid = exprCollation(u_colvar);
1502 404 : res_nscolumn->p_varnosyn = j->rtindex;
1503 404 : res_nscolumn->p_varattnosyn = res_colindex;
1504 : }
1505 : }
1506 : }
1507 :
1508 : /* Add remaining columns from each side to the output columns */
1509 73522 : res_colindex +=
1510 73522 : extractRemainingColumns(pstate,
1511 : l_nscolumns, l_colnames, &l_colnos,
1512 : &res_colnames, &res_colvars,
1513 73522 : res_nscolumns + res_colindex);
1514 73522 : res_colindex +=
1515 73522 : extractRemainingColumns(pstate,
1516 : r_nscolumns, r_colnames, &r_colnos,
1517 : &res_colnames, &res_colvars,
1518 73522 : res_nscolumns + res_colindex);
1519 :
1520 : /* If join has an alias, it syntactically hides all inputs */
1521 73522 : if (j->alias)
1522 : {
1523 978 : for (k = 0; k < res_colindex; k++)
1524 : {
1525 804 : ParseNamespaceColumn *nscol = res_nscolumns + k;
1526 :
1527 804 : nscol->p_varnosyn = j->rtindex;
1528 804 : nscol->p_varattnosyn = k + 1;
1529 : }
1530 : }
1531 :
1532 : /*
1533 : * Now build an RTE and nsitem for the result of the join.
1534 : */
1535 73522 : nsitem = addRangeTableEntryForJoin(pstate,
1536 : res_colnames,
1537 : res_nscolumns,
1538 : j->jointype,
1539 73522 : list_length(j->usingClause),
1540 : res_colvars,
1541 : l_colnos,
1542 : r_colnos,
1543 : j->join_using_alias,
1544 : j->alias,
1545 : true);
1546 :
1547 : /* Verify that we correctly predicted the join's RT index */
1548 : Assert(j->rtindex == nsitem->p_rtindex);
1549 : /* Cross-check number of columns, too */
1550 : Assert(res_colindex == list_length(nsitem->p_names->colnames));
1551 :
1552 : /*
1553 : * Save a link to the JoinExpr in the proper element of p_joinexprs.
1554 : * Since we maintain that list lazily, it may be necessary to fill in
1555 : * empty entries before we can add the JoinExpr in the right place.
1556 : */
1557 191284 : for (k = list_length(pstate->p_joinexprs) + 1; k < j->rtindex; k++)
1558 117768 : pstate->p_joinexprs = lappend(pstate->p_joinexprs, NULL);
1559 73516 : pstate->p_joinexprs = lappend(pstate->p_joinexprs, j);
1560 : Assert(list_length(pstate->p_joinexprs) == j->rtindex);
1561 :
1562 : /*
1563 : * If the join has a USING alias, build a ParseNamespaceItem for that
1564 : * and add it to the list of nsitems in the join's input.
1565 : */
1566 73516 : if (j->join_using_alias)
1567 : {
1568 : ParseNamespaceItem *jnsitem;
1569 :
1570 84 : jnsitem = (ParseNamespaceItem *) palloc(sizeof(ParseNamespaceItem));
1571 84 : jnsitem->p_names = j->join_using_alias;
1572 84 : jnsitem->p_rte = nsitem->p_rte;
1573 84 : jnsitem->p_rtindex = nsitem->p_rtindex;
1574 84 : jnsitem->p_perminfo = NULL;
1575 : /* no need to copy the first N columns, just use res_nscolumns */
1576 84 : jnsitem->p_nscolumns = res_nscolumns;
1577 : /* set default visibility flags; might get changed later */
1578 84 : jnsitem->p_rel_visible = true;
1579 84 : jnsitem->p_cols_visible = true;
1580 84 : jnsitem->p_lateral_only = false;
1581 84 : jnsitem->p_lateral_ok = true;
1582 : /* Per SQL, we must check for alias conflicts */
1583 84 : checkNameSpaceConflicts(pstate, list_make1(jnsitem), my_namespace);
1584 78 : my_namespace = lappend(my_namespace, jnsitem);
1585 : }
1586 :
1587 : /*
1588 : * Prepare returned namespace list. If the JOIN has an alias then it
1589 : * hides the contained RTEs completely; otherwise, the contained RTEs
1590 : * are still visible as table names, but are not visible for
1591 : * unqualified column-name access.
1592 : *
1593 : * Note: if there are nested alias-less JOINs, the lower-level ones
1594 : * will remain in the list although they have neither p_rel_visible
1595 : * nor p_cols_visible set. We could delete such list items, but it's
1596 : * unclear that it's worth expending cycles to do so.
1597 : */
1598 73510 : if (j->alias != NULL)
1599 168 : my_namespace = NIL;
1600 : else
1601 73342 : setNamespaceColumnVisibility(my_namespace, false);
1602 :
1603 : /*
1604 : * The join RTE itself is always made visible for unqualified column
1605 : * names. It's visible as a relation name only if it has an alias.
1606 : */
1607 73510 : nsitem->p_rel_visible = (j->alias != NULL);
1608 73510 : nsitem->p_cols_visible = true;
1609 73510 : nsitem->p_lateral_only = false;
1610 73510 : nsitem->p_lateral_ok = true;
1611 :
1612 73510 : *top_nsitem = nsitem;
1613 73510 : *namespace = lappend(my_namespace, nsitem);
1614 :
1615 73510 : return (Node *) j;
1616 : }
1617 : else
1618 0 : elog(ERROR, "unrecognized node type: %d", (int) nodeTag(n));
1619 : return NULL; /* can't get here, keep compiler quiet */
1620 : }
1621 :
1622 : /*
1623 : * buildVarFromNSColumn -
1624 : * build a Var node using ParseNamespaceColumn data
1625 : *
1626 : * This is used to construct joinaliasvars entries.
1627 : * We can assume varlevelsup should be 0, and no location is specified.
1628 : * Note also that no column SELECT privilege is requested here; that would
1629 : * happen only if the column is actually referenced in the query.
1630 : */
1631 : static Var *
1632 3172192 : buildVarFromNSColumn(ParseState *pstate, ParseNamespaceColumn *nscol)
1633 : {
1634 : Var *var;
1635 :
1636 : Assert(nscol->p_varno > 0); /* i.e., not deleted column */
1637 3172192 : var = makeVar(nscol->p_varno,
1638 3172192 : nscol->p_varattno,
1639 : nscol->p_vartype,
1640 : nscol->p_vartypmod,
1641 : nscol->p_varcollid,
1642 : 0);
1643 : /* makeVar doesn't offer parameters for these, so set by hand: */
1644 3172192 : var->varnosyn = nscol->p_varnosyn;
1645 3172192 : var->varattnosyn = nscol->p_varattnosyn;
1646 :
1647 : /* ... and update varnullingrels */
1648 3172192 : markNullableIfNeeded(pstate, var);
1649 :
1650 3172192 : return var;
1651 : }
1652 :
1653 : /*
1654 : * buildMergedJoinVar -
1655 : * generate a suitable replacement expression for a merged join column
1656 : */
1657 : static Node *
1658 1660 : buildMergedJoinVar(ParseState *pstate, JoinType jointype,
1659 : Var *l_colvar, Var *r_colvar)
1660 : {
1661 : Oid outcoltype;
1662 : int32 outcoltypmod;
1663 : Node *l_node,
1664 : *r_node,
1665 : *res_node;
1666 :
1667 1660 : outcoltype = select_common_type(pstate,
1668 1660 : list_make2(l_colvar, r_colvar),
1669 : "JOIN/USING",
1670 : NULL);
1671 1660 : outcoltypmod = select_common_typmod(pstate,
1672 1660 : list_make2(l_colvar, r_colvar),
1673 : outcoltype);
1674 :
1675 : /*
1676 : * Insert coercion functions if needed. Note that a difference in typmod
1677 : * can only happen if input has typmod but outcoltypmod is -1. In that
1678 : * case we insert a RelabelType to clearly mark that result's typmod is
1679 : * not same as input. We never need coerce_type_typmod.
1680 : */
1681 1660 : if (l_colvar->vartype != outcoltype)
1682 84 : l_node = coerce_type(pstate, (Node *) l_colvar, l_colvar->vartype,
1683 : outcoltype, outcoltypmod,
1684 : COERCION_IMPLICIT, COERCE_IMPLICIT_CAST, -1);
1685 1576 : else if (l_colvar->vartypmod != outcoltypmod)
1686 0 : l_node = (Node *) makeRelabelType((Expr *) l_colvar,
1687 : outcoltype, outcoltypmod,
1688 : InvalidOid, /* fixed below */
1689 : COERCE_IMPLICIT_CAST);
1690 : else
1691 1576 : l_node = (Node *) l_colvar;
1692 :
1693 1660 : if (r_colvar->vartype != outcoltype)
1694 30 : r_node = coerce_type(pstate, (Node *) r_colvar, r_colvar->vartype,
1695 : outcoltype, outcoltypmod,
1696 : COERCION_IMPLICIT, COERCE_IMPLICIT_CAST, -1);
1697 1630 : else if (r_colvar->vartypmod != outcoltypmod)
1698 0 : r_node = (Node *) makeRelabelType((Expr *) r_colvar,
1699 : outcoltype, outcoltypmod,
1700 : InvalidOid, /* fixed below */
1701 : COERCE_IMPLICIT_CAST);
1702 : else
1703 1630 : r_node = (Node *) r_colvar;
1704 :
1705 : /*
1706 : * Choose what to emit
1707 : */
1708 1660 : switch (jointype)
1709 : {
1710 1088 : case JOIN_INNER:
1711 :
1712 : /*
1713 : * We can use either var; prefer non-coerced one if available.
1714 : */
1715 1088 : if (IsA(l_node, Var))
1716 1058 : res_node = l_node;
1717 30 : else if (IsA(r_node, Var))
1718 30 : res_node = r_node;
1719 : else
1720 0 : res_node = l_node;
1721 1088 : break;
1722 210 : case JOIN_LEFT:
1723 : /* Always use left var */
1724 210 : res_node = l_node;
1725 210 : break;
1726 12 : case JOIN_RIGHT:
1727 : /* Always use right var */
1728 12 : res_node = r_node;
1729 12 : break;
1730 350 : case JOIN_FULL:
1731 : {
1732 : /*
1733 : * Here we must build a COALESCE expression to ensure that the
1734 : * join output is non-null if either input is.
1735 : */
1736 350 : CoalesceExpr *c = makeNode(CoalesceExpr);
1737 :
1738 350 : c->coalescetype = outcoltype;
1739 : /* coalescecollid will get set below */
1740 350 : c->args = list_make2(l_node, r_node);
1741 350 : c->location = -1;
1742 350 : res_node = (Node *) c;
1743 350 : break;
1744 : }
1745 0 : default:
1746 0 : elog(ERROR, "unrecognized join type: %d", (int) jointype);
1747 : res_node = NULL; /* keep compiler quiet */
1748 : break;
1749 : }
1750 :
1751 : /*
1752 : * Apply assign_expr_collations to fix up the collation info in the
1753 : * coercion and CoalesceExpr nodes, if we made any. This must be done now
1754 : * so that the join node's alias vars show correct collation info.
1755 : */
1756 1660 : assign_expr_collations(pstate, res_node);
1757 :
1758 1660 : return res_node;
1759 : }
1760 :
1761 : /*
1762 : * markRelsAsNulledBy -
1763 : * Mark the given jointree node and its children as nulled by join jindex
1764 : */
1765 : static void
1766 44514 : markRelsAsNulledBy(ParseState *pstate, Node *n, int jindex)
1767 : {
1768 : int varno;
1769 : ListCell *lc;
1770 :
1771 : /* Note: we can't see FromExpr here */
1772 44514 : if (IsA(n, RangeTblRef))
1773 : {
1774 43076 : varno = ((RangeTblRef *) n)->rtindex;
1775 : }
1776 1438 : else if (IsA(n, JoinExpr))
1777 : {
1778 1438 : JoinExpr *j = (JoinExpr *) n;
1779 :
1780 : /* recurse to children */
1781 1438 : markRelsAsNulledBy(pstate, j->larg, jindex);
1782 1438 : markRelsAsNulledBy(pstate, j->rarg, jindex);
1783 1438 : varno = j->rtindex;
1784 : }
1785 : else
1786 : {
1787 0 : elog(ERROR, "unrecognized node type: %d", (int) nodeTag(n));
1788 : varno = 0; /* keep compiler quiet */
1789 : }
1790 :
1791 : /*
1792 : * Now add jindex to the p_nullingrels set for relation varno. Since we
1793 : * maintain the p_nullingrels list lazily, we might need to extend it to
1794 : * make the varno'th entry exist.
1795 : */
1796 141780 : while (list_length(pstate->p_nullingrels) < varno)
1797 97266 : pstate->p_nullingrels = lappend(pstate->p_nullingrels, NULL);
1798 44514 : lc = list_nth_cell(pstate->p_nullingrels, varno - 1);
1799 44514 : lfirst(lc) = bms_add_member((Bitmapset *) lfirst(lc), jindex);
1800 44514 : }
1801 :
1802 : /*
1803 : * setNamespaceColumnVisibility -
1804 : * Convenience subroutine to update cols_visible flags in a namespace list.
1805 : */
1806 : static void
1807 73342 : setNamespaceColumnVisibility(List *namespace, bool cols_visible)
1808 : {
1809 : ListCell *lc;
1810 :
1811 315168 : foreach(lc, namespace)
1812 : {
1813 241826 : ParseNamespaceItem *nsitem = (ParseNamespaceItem *) lfirst(lc);
1814 :
1815 241826 : nsitem->p_cols_visible = cols_visible;
1816 : }
1817 73342 : }
1818 :
1819 : /*
1820 : * setNamespaceLateralState -
1821 : * Convenience subroutine to update LATERAL flags in a namespace list.
1822 : */
1823 : static void
1824 923114 : setNamespaceLateralState(List *namespace, bool lateral_only, bool lateral_ok)
1825 : {
1826 : ListCell *lc;
1827 :
1828 2277870 : foreach(lc, namespace)
1829 : {
1830 1354756 : ParseNamespaceItem *nsitem = (ParseNamespaceItem *) lfirst(lc);
1831 :
1832 1354756 : nsitem->p_lateral_only = lateral_only;
1833 1354756 : nsitem->p_lateral_ok = lateral_ok;
1834 : }
1835 923114 : }
1836 :
1837 :
1838 : /*
1839 : * transformWhereClause -
1840 : * Transform the qualification and make sure it is of type boolean.
1841 : * Used for WHERE and allied clauses.
1842 : *
1843 : * constructName does not affect the semantics, but is used in error messages
1844 : */
1845 : Node *
1846 965490 : transformWhereClause(ParseState *pstate, Node *clause,
1847 : ParseExprKind exprKind, const char *constructName)
1848 : {
1849 : Node *qual;
1850 :
1851 965490 : if (clause == NULL)
1852 699218 : return NULL;
1853 :
1854 266272 : qual = transformExpr(pstate, clause, exprKind);
1855 :
1856 266068 : qual = coerce_to_boolean(pstate, qual, constructName);
1857 :
1858 266062 : return qual;
1859 : }
1860 :
1861 :
1862 : /*
1863 : * transformLimitClause -
1864 : * Transform the expression and make sure it is of type bigint.
1865 : * Used for LIMIT and allied clauses.
1866 : *
1867 : * Note: as of Postgres 8.2, LIMIT expressions are expected to yield int8,
1868 : * rather than int4 as before.
1869 : *
1870 : * constructName does not affect the semantics, but is used in error messages
1871 : */
1872 : Node *
1873 895480 : transformLimitClause(ParseState *pstate, Node *clause,
1874 : ParseExprKind exprKind, const char *constructName,
1875 : LimitOption limitOption)
1876 : {
1877 : Node *qual;
1878 :
1879 895480 : if (clause == NULL)
1880 890268 : return NULL;
1881 :
1882 5212 : qual = transformExpr(pstate, clause, exprKind);
1883 :
1884 5206 : qual = coerce_to_specific_type(pstate, qual, INT8OID, constructName);
1885 :
1886 : /* LIMIT can't refer to any variables of the current query */
1887 5206 : checkExprIsVarFree(pstate, qual, constructName);
1888 :
1889 : /*
1890 : * Don't allow NULLs in FETCH FIRST .. WITH TIES. This test is ugly and
1891 : * extremely simplistic, in that you can pass a NULL anyway by hiding it
1892 : * inside an expression -- but this protects ruleutils against emitting an
1893 : * unadorned NULL that's not accepted back by the grammar.
1894 : */
1895 5206 : if (exprKind == EXPR_KIND_LIMIT && limitOption == LIMIT_OPTION_WITH_TIES &&
1896 48 : IsA(clause, A_Const) && castNode(A_Const, clause)->isnull)
1897 6 : ereport(ERROR,
1898 : (errcode(ERRCODE_INVALID_ROW_COUNT_IN_LIMIT_CLAUSE),
1899 : errmsg("row count cannot be null in FETCH FIRST ... WITH TIES clause")));
1900 :
1901 5200 : return qual;
1902 : }
1903 :
1904 : /*
1905 : * checkExprIsVarFree
1906 : * Check that given expr has no Vars of the current query level
1907 : * (aggregates and window functions should have been rejected already).
1908 : *
1909 : * This is used to check expressions that have to have a consistent value
1910 : * across all rows of the query, such as a LIMIT. Arguably it should reject
1911 : * volatile functions, too, but we don't do that --- whatever value the
1912 : * function gives on first execution is what you get.
1913 : *
1914 : * constructName does not affect the semantics, but is used in error messages
1915 : */
1916 : static void
1917 7012 : checkExprIsVarFree(ParseState *pstate, Node *n, const char *constructName)
1918 : {
1919 7012 : if (contain_vars_of_level(n, 0))
1920 : {
1921 6 : ereport(ERROR,
1922 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
1923 : /* translator: %s is name of a SQL construct, eg LIMIT */
1924 : errmsg("argument of %s must not contain variables",
1925 : constructName),
1926 : parser_errposition(pstate,
1927 : locate_var_of_level(n, 0))));
1928 : }
1929 7006 : }
1930 :
1931 :
1932 : /*
1933 : * checkTargetlistEntrySQL92 -
1934 : * Validate a targetlist entry found by findTargetlistEntrySQL92
1935 : *
1936 : * When we select a pre-existing tlist entry as a result of syntax such
1937 : * as "GROUP BY 1", we have to make sure it is acceptable for use in the
1938 : * indicated clause type; transformExpr() will have treated it as a regular
1939 : * targetlist item.
1940 : */
1941 : static void
1942 54498 : checkTargetlistEntrySQL92(ParseState *pstate, TargetEntry *tle,
1943 : ParseExprKind exprKind)
1944 : {
1945 54498 : switch (exprKind)
1946 : {
1947 698 : case EXPR_KIND_GROUP_BY:
1948 : /* reject aggregates and window functions */
1949 1274 : if (pstate->p_hasAggs &&
1950 576 : contain_aggs_of_level((Node *) tle->expr, 0))
1951 0 : ereport(ERROR,
1952 : (errcode(ERRCODE_GROUPING_ERROR),
1953 : /* translator: %s is name of a SQL construct, eg GROUP BY */
1954 : errmsg("aggregate functions are not allowed in %s",
1955 : ParseExprKindName(exprKind)),
1956 : parser_errposition(pstate,
1957 : locate_agg_of_level((Node *) tle->expr, 0))));
1958 710 : if (pstate->p_hasWindowFuncs &&
1959 12 : contain_windowfuncs((Node *) tle->expr))
1960 6 : ereport(ERROR,
1961 : (errcode(ERRCODE_WINDOWING_ERROR),
1962 : /* translator: %s is name of a SQL construct, eg GROUP BY */
1963 : errmsg("window functions are not allowed in %s",
1964 : ParseExprKindName(exprKind)),
1965 : parser_errposition(pstate,
1966 : locate_windowfunc((Node *) tle->expr))));
1967 692 : break;
1968 53692 : case EXPR_KIND_ORDER_BY:
1969 : /* no extra checks needed */
1970 53692 : break;
1971 108 : case EXPR_KIND_DISTINCT_ON:
1972 : /* no extra checks needed */
1973 108 : break;
1974 0 : default:
1975 0 : elog(ERROR, "unexpected exprKind in checkTargetlistEntrySQL92");
1976 : break;
1977 : }
1978 54492 : }
1979 :
1980 : /*
1981 : * findTargetlistEntrySQL92 -
1982 : * Returns the targetlist entry matching the given (untransformed) node.
1983 : * If no matching entry exists, one is created and appended to the target
1984 : * list as a "resjunk" node.
1985 : *
1986 : * This function supports the old SQL92 ORDER BY interpretation, where the
1987 : * expression is an output column name or number. If we fail to find a
1988 : * match of that sort, we fall through to the SQL99 rules. For historical
1989 : * reasons, Postgres also allows this interpretation for GROUP BY, though
1990 : * the standard never did. However, for GROUP BY we prefer a SQL99 match.
1991 : * This function is *not* used for WINDOW definitions.
1992 : *
1993 : * node the ORDER BY, GROUP BY, or DISTINCT ON expression to be matched
1994 : * tlist the target list (passed by reference so we can append to it)
1995 : * exprKind identifies clause type being processed
1996 : */
1997 : static TargetEntry *
1998 87678 : findTargetlistEntrySQL92(ParseState *pstate, Node *node, List **tlist,
1999 : ParseExprKind exprKind)
2000 : {
2001 : ListCell *tl;
2002 :
2003 : /*----------
2004 : * Handle two special cases as mandated by the SQL92 spec:
2005 : *
2006 : * 1. Bare ColumnName (no qualifier or subscripts)
2007 : * For a bare identifier, we search for a matching column name
2008 : * in the existing target list. Multiple matches are an error
2009 : * unless they refer to identical values; for example,
2010 : * we allow SELECT a, a FROM table ORDER BY a
2011 : * but not SELECT a AS b, b FROM table ORDER BY b
2012 : * If no match is found, we fall through and treat the identifier
2013 : * as an expression.
2014 : * For GROUP BY, it is incorrect to match the grouping item against
2015 : * targetlist entries: according to SQL92, an identifier in GROUP BY
2016 : * is a reference to a column name exposed by FROM, not to a target
2017 : * list column. However, many implementations (including pre-7.0
2018 : * PostgreSQL) accept this anyway. So for GROUP BY, we look first
2019 : * to see if the identifier matches any FROM column name, and only
2020 : * try for a targetlist name if it doesn't. This ensures that we
2021 : * adhere to the spec in the case where the name could be both.
2022 : * DISTINCT ON isn't in the standard, so we can do what we like there;
2023 : * we choose to make it work like ORDER BY, on the rather flimsy
2024 : * grounds that ordinary DISTINCT works on targetlist entries.
2025 : *
2026 : * 2. IntegerConstant
2027 : * This means to use the n'th item in the existing target list.
2028 : * Note that it would make no sense to order/group/distinct by an
2029 : * actual constant, so this does not create a conflict with SQL99.
2030 : * GROUP BY column-number is not allowed by SQL92, but since
2031 : * the standard has no other behavior defined for this syntax,
2032 : * we may as well accept this common extension.
2033 : *
2034 : * Note that pre-existing resjunk targets must not be used in either case,
2035 : * since the user didn't write them in his SELECT list.
2036 : *
2037 : * If neither special case applies, fall through to treat the item as
2038 : * an expression per SQL99.
2039 : *----------
2040 : */
2041 136466 : if (IsA(node, ColumnRef) &&
2042 48788 : list_length(((ColumnRef *) node)->fields) == 1 &&
2043 33092 : IsA(linitial(((ColumnRef *) node)->fields), String))
2044 : {
2045 33092 : char *name = strVal(linitial(((ColumnRef *) node)->fields));
2046 33092 : int location = ((ColumnRef *) node)->location;
2047 :
2048 33092 : if (exprKind == EXPR_KIND_GROUP_BY)
2049 : {
2050 : /*
2051 : * In GROUP BY, we must prefer a match against a FROM-clause
2052 : * column to one against the targetlist. Look to see if there is
2053 : * a matching column. If so, fall through to use SQL99 rules.
2054 : * NOTE: if name could refer ambiguously to more than one column
2055 : * name exposed by FROM, colNameToVar will ereport(ERROR). That's
2056 : * just what we want here.
2057 : *
2058 : * Small tweak for 7.4.3: ignore matches in upper query levels.
2059 : * This effectively changes the search order for bare names to (1)
2060 : * local FROM variables, (2) local targetlist aliases, (3) outer
2061 : * FROM variables, whereas before it was (1) (3) (2). SQL92 and
2062 : * SQL99 do not allow GROUPing BY an outer reference, so this
2063 : * breaks no cases that are legal per spec, and it seems a more
2064 : * self-consistent behavior.
2065 : */
2066 4346 : if (colNameToVar(pstate, name, true, location) != NULL)
2067 4272 : name = NULL;
2068 : }
2069 :
2070 33092 : if (name != NULL)
2071 : {
2072 28820 : TargetEntry *target_result = NULL;
2073 :
2074 160868 : foreach(tl, *tlist)
2075 : {
2076 132048 : TargetEntry *tle = (TargetEntry *) lfirst(tl);
2077 :
2078 132048 : if (!tle->resjunk &&
2079 131748 : strcmp(tle->resname, name) == 0)
2080 : {
2081 23962 : if (target_result != NULL)
2082 : {
2083 6 : if (!equal(target_result->expr, tle->expr))
2084 0 : ereport(ERROR,
2085 : (errcode(ERRCODE_AMBIGUOUS_COLUMN),
2086 :
2087 : /*------
2088 : translator: first %s is name of a SQL construct, eg ORDER BY */
2089 : errmsg("%s \"%s\" is ambiguous",
2090 : ParseExprKindName(exprKind),
2091 : name),
2092 : parser_errposition(pstate, location)));
2093 : }
2094 : else
2095 23956 : target_result = tle;
2096 : /* Stay in loop to check for ambiguity */
2097 : }
2098 : }
2099 28820 : if (target_result != NULL)
2100 : {
2101 : /* return the first match, after suitable validation */
2102 23956 : checkTargetlistEntrySQL92(pstate, target_result, exprKind);
2103 23956 : return target_result;
2104 : }
2105 : }
2106 : }
2107 63722 : if (IsA(node, A_Const))
2108 : {
2109 30548 : A_Const *aconst = castNode(A_Const, node);
2110 30548 : int targetlist_pos = 0;
2111 : int target_pos;
2112 :
2113 30548 : if (!IsA(&aconst->val, Integer))
2114 0 : ereport(ERROR,
2115 : (errcode(ERRCODE_SYNTAX_ERROR),
2116 : /* translator: %s is name of a SQL construct, eg ORDER BY */
2117 : errmsg("non-integer constant in %s",
2118 : ParseExprKindName(exprKind)),
2119 : parser_errposition(pstate, aconst->location)));
2120 :
2121 30548 : target_pos = intVal(&aconst->val);
2122 54152 : foreach(tl, *tlist)
2123 : {
2124 54146 : TargetEntry *tle = (TargetEntry *) lfirst(tl);
2125 :
2126 54146 : if (!tle->resjunk)
2127 : {
2128 54146 : if (++targetlist_pos == target_pos)
2129 : {
2130 : /* return the unique match, after suitable validation */
2131 30542 : checkTargetlistEntrySQL92(pstate, tle, exprKind);
2132 30536 : return tle;
2133 : }
2134 : }
2135 : }
2136 6 : ereport(ERROR,
2137 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
2138 : /* translator: %s is name of a SQL construct, eg ORDER BY */
2139 : errmsg("%s position %d is not in select list",
2140 : ParseExprKindName(exprKind), target_pos),
2141 : parser_errposition(pstate, aconst->location)));
2142 : }
2143 :
2144 : /*
2145 : * Otherwise, we have an expression, so process it per SQL99 rules.
2146 : */
2147 33174 : return findTargetlistEntrySQL99(pstate, node, tlist, exprKind);
2148 : }
2149 :
2150 : /*
2151 : * findTargetlistEntrySQL99 -
2152 : * Returns the targetlist entry matching the given (untransformed) node.
2153 : * If no matching entry exists, one is created and appended to the target
2154 : * list as a "resjunk" node.
2155 : *
2156 : * This function supports the SQL99 interpretation, wherein the expression
2157 : * is just an ordinary expression referencing input column names.
2158 : *
2159 : * node the ORDER BY, GROUP BY, etc expression to be matched
2160 : * tlist the target list (passed by reference so we can append to it)
2161 : * exprKind identifies clause type being processed
2162 : */
2163 : static TargetEntry *
2164 38398 : findTargetlistEntrySQL99(ParseState *pstate, Node *node, List **tlist,
2165 : ParseExprKind exprKind)
2166 : {
2167 : TargetEntry *target_result;
2168 : ListCell *tl;
2169 : Node *expr;
2170 :
2171 : /*
2172 : * Convert the untransformed node to a transformed expression, and search
2173 : * for a match in the tlist. NOTE: it doesn't really matter whether there
2174 : * is more than one match. Also, we are willing to match an existing
2175 : * resjunk target here, though the SQL92 cases above must ignore resjunk
2176 : * targets.
2177 : */
2178 38398 : expr = transformExpr(pstate, node, exprKind);
2179 :
2180 141072 : foreach(tl, *tlist)
2181 : {
2182 118490 : TargetEntry *tle = (TargetEntry *) lfirst(tl);
2183 : Node *texpr;
2184 :
2185 : /*
2186 : * Ignore any implicit cast on the existing tlist expression.
2187 : *
2188 : * This essentially allows the ORDER/GROUP/etc item to adopt the same
2189 : * datatype previously selected for a textually-equivalent tlist item.
2190 : * There can't be any implicit cast at top level in an ordinary SELECT
2191 : * tlist at this stage, but the case does arise with ORDER BY in an
2192 : * aggregate function.
2193 : */
2194 118490 : texpr = strip_implicit_coercions((Node *) tle->expr);
2195 :
2196 118490 : if (equal(expr, texpr))
2197 15762 : return tle;
2198 : }
2199 :
2200 : /*
2201 : * If no matches, construct a new target entry which is appended to the
2202 : * end of the target list. This target is given resjunk = true so that it
2203 : * will not be projected into the final tuple.
2204 : */
2205 22582 : target_result = transformTargetEntry(pstate, node, expr, exprKind,
2206 : NULL, true);
2207 :
2208 22582 : *tlist = lappend(*tlist, target_result);
2209 :
2210 22582 : return target_result;
2211 : }
2212 :
2213 : /*-------------------------------------------------------------------------
2214 : * Flatten out parenthesized sublists in grouping lists, and some cases
2215 : * of nested grouping sets.
2216 : *
2217 : * Inside a grouping set (ROLLUP, CUBE, or GROUPING SETS), we expect the
2218 : * content to be nested no more than 2 deep: i.e. ROLLUP((a,b),(c,d)) is
2219 : * ok, but ROLLUP((a,(b,c)),d) is flattened to ((a,b,c),d), which we then
2220 : * (later) normalize to ((a,b,c),(d)).
2221 : *
2222 : * CUBE or ROLLUP can be nested inside GROUPING SETS (but not the reverse),
2223 : * and we leave that alone if we find it. But if we see GROUPING SETS inside
2224 : * GROUPING SETS, we can flatten and normalize as follows:
2225 : * GROUPING SETS (a, (b,c), GROUPING SETS ((c,d),(e)), (f,g))
2226 : * becomes
2227 : * GROUPING SETS ((a), (b,c), (c,d), (e), (f,g))
2228 : *
2229 : * This is per the spec's syntax transformations, but these are the only such
2230 : * transformations we do in parse analysis, so that queries retain the
2231 : * originally specified grouping set syntax for CUBE and ROLLUP as much as
2232 : * possible when deparsed. (Full expansion of the result into a list of
2233 : * grouping sets is left to the planner.)
2234 : *
2235 : * When we're done, the resulting list should contain only these possible
2236 : * elements:
2237 : * - an expression
2238 : * - a CUBE or ROLLUP with a list of expressions nested 2 deep
2239 : * - a GROUPING SET containing any of:
2240 : * - expression lists
2241 : * - empty grouping sets
2242 : * - CUBE or ROLLUP nodes with lists nested 2 deep
2243 : * The return is a new list, but doesn't deep-copy the old nodes except for
2244 : * GroupingSet nodes.
2245 : *
2246 : * As a side effect, flag whether the list has any GroupingSet nodes.
2247 : *-------------------------------------------------------------------------
2248 : */
2249 : static Node *
2250 446798 : flatten_grouping_sets(Node *expr, bool toplevel, bool *hasGroupingSets)
2251 : {
2252 : /* just in case of pathological input */
2253 446798 : check_stack_depth();
2254 :
2255 446798 : if (expr == (Node *) NIL)
2256 431858 : return (Node *) NIL;
2257 :
2258 14940 : switch (expr->type)
2259 : {
2260 292 : case T_RowExpr:
2261 : {
2262 292 : RowExpr *r = (RowExpr *) expr;
2263 :
2264 292 : if (r->row_format == COERCE_IMPLICIT_CAST)
2265 292 : return flatten_grouping_sets((Node *) r->args,
2266 : false, NULL);
2267 : }
2268 0 : break;
2269 1128 : case T_GroupingSet:
2270 : {
2271 1128 : GroupingSet *gset = (GroupingSet *) expr;
2272 : ListCell *l2;
2273 1128 : List *result_set = NIL;
2274 :
2275 1128 : if (hasGroupingSets)
2276 804 : *hasGroupingSets = true;
2277 :
2278 : /*
2279 : * at the top level, we skip over all empty grouping sets; the
2280 : * caller can supply the canonical GROUP BY () if nothing is
2281 : * left.
2282 : */
2283 :
2284 1128 : if (toplevel && gset->kind == GROUPING_SET_EMPTY)
2285 18 : return (Node *) NIL;
2286 :
2287 2968 : foreach(l2, gset->content)
2288 : {
2289 1858 : Node *n1 = lfirst(l2);
2290 1858 : Node *n2 = flatten_grouping_sets(n1, false, NULL);
2291 :
2292 1858 : if (IsA(n1, GroupingSet) &&
2293 324 : ((GroupingSet *) n1)->kind == GROUPING_SET_SETS)
2294 96 : result_set = list_concat(result_set, (List *) n2);
2295 : else
2296 1762 : result_set = lappend(result_set, n2);
2297 : }
2298 :
2299 : /*
2300 : * At top level, keep the grouping set node; but if we're in a
2301 : * nested grouping set, then we need to concat the flattened
2302 : * result into the outer list if it's simply nested.
2303 : */
2304 :
2305 1110 : if (toplevel || (gset->kind != GROUPING_SET_SETS))
2306 : {
2307 1014 : return (Node *) makeGroupingSet(gset->kind, result_set, gset->location);
2308 : }
2309 : else
2310 96 : return (Node *) result_set;
2311 : }
2312 5152 : case T_List:
2313 : {
2314 5152 : List *result = NIL;
2315 : ListCell *l;
2316 :
2317 13082 : foreach(l, (List *) expr)
2318 : {
2319 7930 : Node *n = flatten_grouping_sets(lfirst(l), toplevel, hasGroupingSets);
2320 :
2321 7930 : if (n != (Node *) NIL)
2322 : {
2323 7912 : if (IsA(n, List))
2324 46 : result = list_concat(result, (List *) n);
2325 : else
2326 7866 : result = lappend(result, n);
2327 : }
2328 : }
2329 :
2330 5152 : return (Node *) result;
2331 : }
2332 8368 : default:
2333 8368 : break;
2334 : }
2335 :
2336 8368 : return expr;
2337 : }
2338 :
2339 : /*
2340 : * Transform a single expression within a GROUP BY clause or grouping set.
2341 : *
2342 : * The expression is added to the targetlist if not already present, and to the
2343 : * flatresult list (which will become the groupClause) if not already present
2344 : * there. The sortClause is consulted for operator and sort order hints.
2345 : *
2346 : * Returns the ressortgroupref of the expression.
2347 : *
2348 : * flatresult reference to flat list of SortGroupClause nodes
2349 : * seen_local bitmapset of sortgrouprefs already seen at the local level
2350 : * pstate ParseState
2351 : * gexpr node to transform
2352 : * targetlist reference to TargetEntry list
2353 : * sortClause ORDER BY clause (SortGroupClause nodes)
2354 : * exprKind expression kind
2355 : * useSQL99 SQL99 rather than SQL92 syntax
2356 : * toplevel false if within any grouping set
2357 : */
2358 : static Index
2359 8368 : transformGroupClauseExpr(List **flatresult, Bitmapset *seen_local,
2360 : ParseState *pstate, Node *gexpr,
2361 : List **targetlist, List *sortClause,
2362 : ParseExprKind exprKind, bool useSQL99, bool toplevel)
2363 : {
2364 : TargetEntry *tle;
2365 8368 : bool found = false;
2366 :
2367 8368 : if (useSQL99)
2368 1092 : tle = findTargetlistEntrySQL99(pstate, gexpr,
2369 : targetlist, exprKind);
2370 : else
2371 7276 : tle = findTargetlistEntrySQL92(pstate, gexpr,
2372 : targetlist, exprKind);
2373 :
2374 8344 : if (tle->ressortgroupref > 0)
2375 : {
2376 : ListCell *sl;
2377 :
2378 : /*
2379 : * Eliminate duplicates (GROUP BY x, x) but only at local level.
2380 : * (Duplicates in grouping sets can affect the number of returned
2381 : * rows, so can't be dropped indiscriminately.)
2382 : *
2383 : * Since we don't care about anything except the sortgroupref, we can
2384 : * use a bitmapset rather than scanning lists.
2385 : */
2386 2446 : if (bms_is_member(tle->ressortgroupref, seen_local))
2387 24 : return 0;
2388 :
2389 : /*
2390 : * If we're already in the flat clause list, we don't need to consider
2391 : * adding ourselves again.
2392 : */
2393 2422 : found = targetIsInSortList(tle, InvalidOid, *flatresult);
2394 2422 : if (found)
2395 202 : return tle->ressortgroupref;
2396 :
2397 : /*
2398 : * If the GROUP BY tlist entry also appears in ORDER BY, copy operator
2399 : * info from the (first) matching ORDER BY item. This means that if
2400 : * you write something like "GROUP BY foo ORDER BY foo USING <<<", the
2401 : * GROUP BY operation silently takes on the equality semantics implied
2402 : * by the ORDER BY. There are two reasons to do this: it improves the
2403 : * odds that we can implement both GROUP BY and ORDER BY with a single
2404 : * sort step, and it allows the user to choose the equality semantics
2405 : * used by GROUP BY, should she be working with a datatype that has
2406 : * more than one equality operator.
2407 : *
2408 : * If we're in a grouping set, though, we force our requested ordering
2409 : * to be NULLS LAST, because if we have any hope of using a sorted agg
2410 : * for the job, we're going to be tacking on generated NULL values
2411 : * after the corresponding groups. If the user demands nulls first,
2412 : * another sort step is going to be inevitable, but that's the
2413 : * planner's problem.
2414 : */
2415 :
2416 3118 : foreach(sl, sortClause)
2417 : {
2418 2928 : SortGroupClause *sc = (SortGroupClause *) lfirst(sl);
2419 :
2420 2928 : if (sc->tleSortGroupRef == tle->ressortgroupref)
2421 : {
2422 2030 : SortGroupClause *grpc = copyObject(sc);
2423 :
2424 2030 : if (!toplevel)
2425 508 : grpc->nulls_first = false;
2426 2030 : *flatresult = lappend(*flatresult, grpc);
2427 2030 : found = true;
2428 2030 : break;
2429 : }
2430 : }
2431 : }
2432 :
2433 : /*
2434 : * If no match in ORDER BY, just add it to the result using default
2435 : * sort/group semantics.
2436 : */
2437 8118 : if (!found)
2438 6088 : *flatresult = addTargetToGroupList(pstate, tle,
2439 : *flatresult, *targetlist,
2440 : exprLocation(gexpr));
2441 :
2442 : /*
2443 : * _something_ must have assigned us a sortgroupref by now...
2444 : */
2445 :
2446 8118 : return tle->ressortgroupref;
2447 : }
2448 :
2449 : /*
2450 : * Transform a list of expressions within a GROUP BY clause or grouping set.
2451 : *
2452 : * The list of expressions belongs to a single clause within which duplicates
2453 : * can be safely eliminated.
2454 : *
2455 : * Returns an integer list of ressortgroupref values.
2456 : *
2457 : * flatresult reference to flat list of SortGroupClause nodes
2458 : * pstate ParseState
2459 : * list nodes to transform
2460 : * targetlist reference to TargetEntry list
2461 : * sortClause ORDER BY clause (SortGroupClause nodes)
2462 : * exprKind expression kind
2463 : * useSQL99 SQL99 rather than SQL92 syntax
2464 : * toplevel false if within any grouping set
2465 : */
2466 : static List *
2467 246 : transformGroupClauseList(List **flatresult,
2468 : ParseState *pstate, List *list,
2469 : List **targetlist, List *sortClause,
2470 : ParseExprKind exprKind, bool useSQL99, bool toplevel)
2471 : {
2472 246 : Bitmapset *seen_local = NULL;
2473 246 : List *result = NIL;
2474 : ListCell *gl;
2475 :
2476 762 : foreach(gl, list)
2477 : {
2478 516 : Node *gexpr = (Node *) lfirst(gl);
2479 :
2480 516 : Index ref = transformGroupClauseExpr(flatresult,
2481 : seen_local,
2482 : pstate,
2483 : gexpr,
2484 : targetlist,
2485 : sortClause,
2486 : exprKind,
2487 : useSQL99,
2488 : toplevel);
2489 :
2490 516 : if (ref > 0)
2491 : {
2492 504 : seen_local = bms_add_member(seen_local, ref);
2493 504 : result = lappend_int(result, ref);
2494 : }
2495 : }
2496 :
2497 246 : return result;
2498 : }
2499 :
2500 : /*
2501 : * Transform a grouping set and (recursively) its content.
2502 : *
2503 : * The grouping set might be a GROUPING SETS node with other grouping sets
2504 : * inside it, but SETS within SETS have already been flattened out before
2505 : * reaching here.
2506 : *
2507 : * Returns the transformed node, which now contains SIMPLE nodes with lists
2508 : * of ressortgrouprefs rather than expressions.
2509 : *
2510 : * flatresult reference to flat list of SortGroupClause nodes
2511 : * pstate ParseState
2512 : * gset grouping set to transform
2513 : * targetlist reference to TargetEntry list
2514 : * sortClause ORDER BY clause (SortGroupClause nodes)
2515 : * exprKind expression kind
2516 : * useSQL99 SQL99 rather than SQL92 syntax
2517 : * toplevel false if within any grouping set
2518 : */
2519 : static Node *
2520 1014 : transformGroupingSet(List **flatresult,
2521 : ParseState *pstate, GroupingSet *gset,
2522 : List **targetlist, List *sortClause,
2523 : ParseExprKind exprKind, bool useSQL99, bool toplevel)
2524 : {
2525 : ListCell *gl;
2526 1014 : List *content = NIL;
2527 :
2528 : Assert(toplevel || gset->kind != GROUPING_SET_SETS);
2529 :
2530 2776 : foreach(gl, gset->content)
2531 : {
2532 1762 : Node *n = lfirst(gl);
2533 :
2534 1762 : if (IsA(n, List))
2535 : {
2536 246 : List *l = transformGroupClauseList(flatresult,
2537 : pstate, (List *) n,
2538 : targetlist, sortClause,
2539 : exprKind, useSQL99, false);
2540 :
2541 246 : content = lappend(content, makeGroupingSet(GROUPING_SET_SIMPLE,
2542 : l,
2543 : exprLocation(n)));
2544 : }
2545 1516 : else if (IsA(n, GroupingSet))
2546 : {
2547 228 : GroupingSet *gset2 = (GroupingSet *) lfirst(gl);
2548 :
2549 228 : content = lappend(content, transformGroupingSet(flatresult,
2550 : pstate, gset2,
2551 : targetlist, sortClause,
2552 : exprKind, useSQL99, false));
2553 : }
2554 : else
2555 : {
2556 1288 : Index ref = transformGroupClauseExpr(flatresult,
2557 : NULL,
2558 : pstate,
2559 : n,
2560 : targetlist,
2561 : sortClause,
2562 : exprKind,
2563 : useSQL99,
2564 : false);
2565 :
2566 2576 : content = lappend(content, makeGroupingSet(GROUPING_SET_SIMPLE,
2567 1288 : list_make1_int(ref),
2568 : exprLocation(n)));
2569 : }
2570 : }
2571 :
2572 : /* Arbitrarily cap the size of CUBE, which has exponential growth */
2573 1014 : if (gset->kind == GROUPING_SET_CUBE)
2574 : {
2575 184 : if (list_length(content) > 12)
2576 0 : ereport(ERROR,
2577 : (errcode(ERRCODE_TOO_MANY_COLUMNS),
2578 : errmsg("CUBE is limited to 12 elements"),
2579 : parser_errposition(pstate, gset->location)));
2580 : }
2581 :
2582 1014 : return (Node *) makeGroupingSet(gset->kind, content, gset->location);
2583 : }
2584 :
2585 :
2586 : /*
2587 : * transformGroupClause -
2588 : * transform a GROUP BY clause
2589 : *
2590 : * GROUP BY items will be added to the targetlist (as resjunk columns)
2591 : * if not already present, so the targetlist must be passed by reference.
2592 : *
2593 : * This is also used for window PARTITION BY clauses (which act almost the
2594 : * same, but are always interpreted per SQL99 rules).
2595 : *
2596 : * Grouping sets make this a lot more complex than it was. Our goal here is
2597 : * twofold: we make a flat list of SortGroupClause nodes referencing each
2598 : * distinct expression used for grouping, with those expressions added to the
2599 : * targetlist if needed. At the same time, we build the groupingSets tree,
2600 : * which stores only ressortgrouprefs as integer lists inside GroupingSet nodes
2601 : * (possibly nested, but limited in depth: a GROUPING_SET_SETS node can contain
2602 : * nested SIMPLE, CUBE or ROLLUP nodes, but not more sets - we flatten that
2603 : * out; while CUBE and ROLLUP can contain only SIMPLE nodes).
2604 : *
2605 : * We skip much of the hard work if there are no grouping sets.
2606 : *
2607 : * One subtlety is that the groupClause list can end up empty while the
2608 : * groupingSets list is not; this happens if there are only empty grouping
2609 : * sets, or an explicit GROUP BY (). This has the same effect as specifying
2610 : * aggregates or a HAVING clause with no GROUP BY; the output is one row per
2611 : * grouping set even if the input is empty.
2612 : *
2613 : * Returns the transformed (flat) groupClause.
2614 : *
2615 : * pstate ParseState
2616 : * grouplist clause to transform
2617 : * groupingSets reference to list to contain the grouping set tree
2618 : * targetlist reference to TargetEntry list
2619 : * sortClause ORDER BY clause (SortGroupClause nodes)
2620 : * exprKind expression kind
2621 : * useSQL99 SQL99 rather than SQL92 syntax
2622 : */
2623 : List *
2624 436718 : transformGroupClause(ParseState *pstate, List *grouplist, List **groupingSets,
2625 : List **targetlist, List *sortClause,
2626 : ParseExprKind exprKind, bool useSQL99)
2627 : {
2628 436718 : List *result = NIL;
2629 : List *flat_grouplist;
2630 436718 : List *gsets = NIL;
2631 : ListCell *gl;
2632 436718 : bool hasGroupingSets = false;
2633 436718 : Bitmapset *seen_local = NULL;
2634 :
2635 : /*
2636 : * Recursively flatten implicit RowExprs. (Technically this is only needed
2637 : * for GROUP BY, per the syntax rules for grouping sets, but we do it
2638 : * anyway.)
2639 : */
2640 436718 : flat_grouplist = (List *) flatten_grouping_sets((Node *) grouplist,
2641 : true,
2642 : &hasGroupingSets);
2643 :
2644 : /*
2645 : * If the list is now empty, but hasGroupingSets is true, it's because we
2646 : * elided redundant empty grouping sets. Restore a single empty grouping
2647 : * set to leave a canonical form: GROUP BY ()
2648 : */
2649 :
2650 436718 : if (flat_grouplist == NIL && hasGroupingSets)
2651 : {
2652 18 : flat_grouplist = list_make1(makeGroupingSet(GROUPING_SET_EMPTY,
2653 : NIL,
2654 : exprLocation((Node *) grouplist)));
2655 : }
2656 :
2657 444062 : foreach(gl, flat_grouplist)
2658 : {
2659 7368 : Node *gexpr = (Node *) lfirst(gl);
2660 :
2661 7368 : if (IsA(gexpr, GroupingSet))
2662 : {
2663 804 : GroupingSet *gset = (GroupingSet *) gexpr;
2664 :
2665 804 : switch (gset->kind)
2666 : {
2667 18 : case GROUPING_SET_EMPTY:
2668 18 : gsets = lappend(gsets, gset);
2669 18 : break;
2670 0 : case GROUPING_SET_SIMPLE:
2671 : /* can't happen */
2672 : Assert(false);
2673 0 : break;
2674 786 : case GROUPING_SET_SETS:
2675 : case GROUPING_SET_CUBE:
2676 : case GROUPING_SET_ROLLUP:
2677 786 : gsets = lappend(gsets,
2678 786 : transformGroupingSet(&result,
2679 : pstate, gset,
2680 : targetlist, sortClause,
2681 : exprKind, useSQL99, true));
2682 786 : break;
2683 : }
2684 804 : }
2685 : else
2686 : {
2687 6564 : Index ref = transformGroupClauseExpr(&result, seen_local,
2688 : pstate, gexpr,
2689 : targetlist, sortClause,
2690 : exprKind, useSQL99, true);
2691 :
2692 6540 : if (ref > 0)
2693 : {
2694 6528 : seen_local = bms_add_member(seen_local, ref);
2695 6528 : if (hasGroupingSets)
2696 36 : gsets = lappend(gsets,
2697 72 : makeGroupingSet(GROUPING_SET_SIMPLE,
2698 36 : list_make1_int(ref),
2699 : exprLocation(gexpr)));
2700 : }
2701 : }
2702 : }
2703 :
2704 : /* parser should prevent this */
2705 : Assert(gsets == NIL || groupingSets != NULL);
2706 :
2707 436694 : if (groupingSets)
2708 434030 : *groupingSets = gsets;
2709 :
2710 436694 : return result;
2711 : }
2712 :
2713 : /*
2714 : * transformSortClause -
2715 : * transform an ORDER BY clause
2716 : *
2717 : * ORDER BY items will be added to the targetlist (as resjunk columns)
2718 : * if not already present, so the targetlist must be passed by reference.
2719 : *
2720 : * This is also used for window and aggregate ORDER BY clauses (which act
2721 : * almost the same, but are always interpreted per SQL99 rules).
2722 : */
2723 : List *
2724 489158 : transformSortClause(ParseState *pstate,
2725 : List *orderlist,
2726 : List **targetlist,
2727 : ParseExprKind exprKind,
2728 : bool useSQL99)
2729 : {
2730 489158 : List *sortlist = NIL;
2731 : ListCell *olitem;
2732 :
2733 573406 : foreach(olitem, orderlist)
2734 : {
2735 84290 : SortBy *sortby = (SortBy *) lfirst(olitem);
2736 : TargetEntry *tle;
2737 :
2738 84290 : if (useSQL99)
2739 4132 : tle = findTargetlistEntrySQL99(pstate, sortby->node,
2740 : targetlist, exprKind);
2741 : else
2742 80158 : tle = findTargetlistEntrySQL92(pstate, sortby->node,
2743 : targetlist, exprKind);
2744 :
2745 84254 : sortlist = addTargetToSortList(pstate, tle,
2746 : sortlist, *targetlist, sortby);
2747 : }
2748 :
2749 489116 : return sortlist;
2750 : }
2751 :
2752 : /*
2753 : * transformWindowDefinitions -
2754 : * transform window definitions (WindowDef to WindowClause)
2755 : */
2756 : List *
2757 434006 : transformWindowDefinitions(ParseState *pstate,
2758 : List *windowdefs,
2759 : List **targetlist)
2760 : {
2761 434006 : List *result = NIL;
2762 434006 : Index winref = 0;
2763 : ListCell *lc;
2764 :
2765 436616 : foreach(lc, windowdefs)
2766 : {
2767 2676 : WindowDef *windef = (WindowDef *) lfirst(lc);
2768 2676 : WindowClause *refwc = NULL;
2769 : List *partitionClause;
2770 : List *orderClause;
2771 2676 : Oid rangeopfamily = InvalidOid;
2772 2676 : Oid rangeopcintype = InvalidOid;
2773 : WindowClause *wc;
2774 :
2775 2676 : winref++;
2776 :
2777 : /*
2778 : * Check for duplicate window names.
2779 : */
2780 3216 : if (windef->name &&
2781 540 : findWindowClause(result, windef->name) != NULL)
2782 6 : ereport(ERROR,
2783 : (errcode(ERRCODE_WINDOWING_ERROR),
2784 : errmsg("window \"%s\" is already defined", windef->name),
2785 : parser_errposition(pstate, windef->location)));
2786 :
2787 : /*
2788 : * If it references a previous window, look that up.
2789 : */
2790 2670 : if (windef->refname)
2791 : {
2792 24 : refwc = findWindowClause(result, windef->refname);
2793 24 : if (refwc == NULL)
2794 0 : ereport(ERROR,
2795 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2796 : errmsg("window \"%s\" does not exist",
2797 : windef->refname),
2798 : parser_errposition(pstate, windef->location)));
2799 : }
2800 :
2801 : /*
2802 : * Transform PARTITION and ORDER specs, if any. These are treated
2803 : * almost exactly like top-level GROUP BY and ORDER BY clauses,
2804 : * including the special handling of nondefault operator semantics.
2805 : */
2806 2670 : orderClause = transformSortClause(pstate,
2807 : windef->orderClause,
2808 : targetlist,
2809 : EXPR_KIND_WINDOW_ORDER,
2810 : true /* force SQL99 rules */ );
2811 2664 : partitionClause = transformGroupClause(pstate,
2812 : windef->partitionClause,
2813 : NULL,
2814 : targetlist,
2815 : orderClause,
2816 : EXPR_KIND_WINDOW_PARTITION,
2817 : true /* force SQL99 rules */ );
2818 :
2819 : /*
2820 : * And prepare the new WindowClause.
2821 : */
2822 2664 : wc = makeNode(WindowClause);
2823 2664 : wc->name = windef->name;
2824 2664 : wc->refname = windef->refname;
2825 :
2826 : /*
2827 : * Per spec, a windowdef that references a previous one copies the
2828 : * previous partition clause (and mustn't specify its own). It can
2829 : * specify its own ordering clause, but only if the previous one had
2830 : * none. It always specifies its own frame clause, and the previous
2831 : * one must not have a frame clause. Yeah, it's bizarre that each of
2832 : * these cases works differently, but SQL:2008 says so; see 7.11
2833 : * <window clause> syntax rule 10 and general rule 1. The frame
2834 : * clause rule is especially bizarre because it makes "OVER foo"
2835 : * different from "OVER (foo)", and requires the latter to throw an
2836 : * error if foo has a nondefault frame clause. Well, ours not to
2837 : * reason why, but we do go out of our way to throw a useful error
2838 : * message for such cases.
2839 : */
2840 2664 : if (refwc)
2841 : {
2842 24 : if (partitionClause)
2843 0 : ereport(ERROR,
2844 : (errcode(ERRCODE_WINDOWING_ERROR),
2845 : errmsg("cannot override PARTITION BY clause of window \"%s\"",
2846 : windef->refname),
2847 : parser_errposition(pstate, windef->location)));
2848 24 : wc->partitionClause = copyObject(refwc->partitionClause);
2849 : }
2850 : else
2851 2640 : wc->partitionClause = partitionClause;
2852 2664 : if (refwc)
2853 : {
2854 24 : if (orderClause && refwc->orderClause)
2855 0 : ereport(ERROR,
2856 : (errcode(ERRCODE_WINDOWING_ERROR),
2857 : errmsg("cannot override ORDER BY clause of window \"%s\"",
2858 : windef->refname),
2859 : parser_errposition(pstate, windef->location)));
2860 24 : if (orderClause)
2861 : {
2862 0 : wc->orderClause = orderClause;
2863 0 : wc->copiedOrder = false;
2864 : }
2865 : else
2866 : {
2867 24 : wc->orderClause = copyObject(refwc->orderClause);
2868 24 : wc->copiedOrder = true;
2869 : }
2870 : }
2871 : else
2872 : {
2873 2640 : wc->orderClause = orderClause;
2874 2640 : wc->copiedOrder = false;
2875 : }
2876 2664 : if (refwc && refwc->frameOptions != FRAMEOPTION_DEFAULTS)
2877 : {
2878 : /*
2879 : * Use this message if this is a WINDOW clause, or if it's an OVER
2880 : * clause that includes ORDER BY or framing clauses. (We already
2881 : * rejected PARTITION BY above, so no need to check that.)
2882 : */
2883 0 : if (windef->name ||
2884 0 : orderClause || windef->frameOptions != FRAMEOPTION_DEFAULTS)
2885 0 : ereport(ERROR,
2886 : (errcode(ERRCODE_WINDOWING_ERROR),
2887 : errmsg("cannot copy window \"%s\" because it has a frame clause",
2888 : windef->refname),
2889 : parser_errposition(pstate, windef->location)));
2890 : /* Else this clause is just OVER (foo), so say this: */
2891 0 : ereport(ERROR,
2892 : (errcode(ERRCODE_WINDOWING_ERROR),
2893 : errmsg("cannot copy window \"%s\" because it has a frame clause",
2894 : windef->refname),
2895 : errhint("Omit the parentheses in this OVER clause."),
2896 : parser_errposition(pstate, windef->location)));
2897 : }
2898 2664 : wc->frameOptions = windef->frameOptions;
2899 :
2900 : /*
2901 : * RANGE offset PRECEDING/FOLLOWING requires exactly one ORDER BY
2902 : * column; check that and get its sort opfamily info.
2903 : */
2904 2664 : if ((wc->frameOptions & FRAMEOPTION_RANGE) &&
2905 1890 : (wc->frameOptions & (FRAMEOPTION_START_OFFSET |
2906 : FRAMEOPTION_END_OFFSET)))
2907 : {
2908 : SortGroupClause *sortcl;
2909 : Node *sortkey;
2910 : int16 rangestrategy;
2911 :
2912 636 : if (list_length(wc->orderClause) != 1)
2913 18 : ereport(ERROR,
2914 : (errcode(ERRCODE_WINDOWING_ERROR),
2915 : errmsg("RANGE with offset PRECEDING/FOLLOWING requires exactly one ORDER BY column"),
2916 : parser_errposition(pstate, windef->location)));
2917 618 : sortcl = linitial_node(SortGroupClause, wc->orderClause);
2918 618 : sortkey = get_sortgroupclause_expr(sortcl, *targetlist);
2919 : /* Find the sort operator in pg_amop */
2920 618 : if (!get_ordering_op_properties(sortcl->sortop,
2921 : &rangeopfamily,
2922 : &rangeopcintype,
2923 : &rangestrategy))
2924 0 : elog(ERROR, "operator %u is not a valid ordering operator",
2925 : sortcl->sortop);
2926 : /* Record properties of sort ordering */
2927 618 : wc->inRangeColl = exprCollation(sortkey);
2928 618 : wc->inRangeAsc = (rangestrategy == BTLessStrategyNumber);
2929 618 : wc->inRangeNullsFirst = sortcl->nulls_first;
2930 : }
2931 :
2932 : /* Per spec, GROUPS mode requires an ORDER BY clause */
2933 2646 : if (wc->frameOptions & FRAMEOPTION_GROUPS)
2934 : {
2935 174 : if (wc->orderClause == NIL)
2936 6 : ereport(ERROR,
2937 : (errcode(ERRCODE_WINDOWING_ERROR),
2938 : errmsg("GROUPS mode requires an ORDER BY clause"),
2939 : parser_errposition(pstate, windef->location)));
2940 : }
2941 :
2942 : /* Process frame offset expressions */
2943 2640 : wc->startOffset = transformFrameOffset(pstate, wc->frameOptions,
2944 : rangeopfamily, rangeopcintype,
2945 : &wc->startInRangeFunc,
2946 : windef->startOffset);
2947 2616 : wc->endOffset = transformFrameOffset(pstate, wc->frameOptions,
2948 : rangeopfamily, rangeopcintype,
2949 : &wc->endInRangeFunc,
2950 : windef->endOffset);
2951 2610 : wc->runCondition = NIL;
2952 2610 : wc->winref = winref;
2953 :
2954 2610 : result = lappend(result, wc);
2955 : }
2956 :
2957 433940 : return result;
2958 : }
2959 :
2960 : /*
2961 : * transformDistinctClause -
2962 : * transform a DISTINCT clause
2963 : *
2964 : * Since we may need to add items to the query's targetlist, that list
2965 : * is passed by reference.
2966 : *
2967 : * As with GROUP BY, we absorb the sorting semantics of ORDER BY as much as
2968 : * possible into the distinctClause. This avoids a possible need to re-sort,
2969 : * and allows the user to choose the equality semantics used by DISTINCT,
2970 : * should she be working with a datatype that has more than one equality
2971 : * operator.
2972 : *
2973 : * is_agg is true if we are transforming an aggregate(DISTINCT ...)
2974 : * function call. This does not affect any behavior, only the phrasing
2975 : * of error messages.
2976 : */
2977 : List *
2978 3346 : transformDistinctClause(ParseState *pstate,
2979 : List **targetlist, List *sortClause, bool is_agg)
2980 : {
2981 3346 : List *result = NIL;
2982 : ListCell *slitem;
2983 : ListCell *tlitem;
2984 :
2985 : /*
2986 : * The distinctClause should consist of all ORDER BY items followed by all
2987 : * other non-resjunk targetlist items. There must not be any resjunk
2988 : * ORDER BY items --- that would imply that we are sorting by a value that
2989 : * isn't necessarily unique within a DISTINCT group, so the results
2990 : * wouldn't be well-defined. This construction ensures we follow the rule
2991 : * that sortClause and distinctClause match; in fact the sortClause will
2992 : * always be a prefix of distinctClause.
2993 : *
2994 : * Note a corner case: the same TLE could be in the ORDER BY list multiple
2995 : * times with different sortops. We have to include it in the
2996 : * distinctClause the same way to preserve the prefix property. The net
2997 : * effect will be that the TLE value will be made unique according to both
2998 : * sortops.
2999 : */
3000 3944 : foreach(slitem, sortClause)
3001 : {
3002 634 : SortGroupClause *scl = (SortGroupClause *) lfirst(slitem);
3003 634 : TargetEntry *tle = get_sortgroupclause_tle(scl, *targetlist);
3004 :
3005 634 : if (tle->resjunk)
3006 36 : ereport(ERROR,
3007 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3008 : is_agg ?
3009 : errmsg("in an aggregate with DISTINCT, ORDER BY expressions must appear in argument list") :
3010 : errmsg("for SELECT DISTINCT, ORDER BY expressions must appear in select list"),
3011 : parser_errposition(pstate,
3012 : exprLocation((Node *) tle->expr))));
3013 598 : result = lappend(result, copyObject(scl));
3014 : }
3015 :
3016 : /*
3017 : * Now add any remaining non-resjunk tlist items, using default sort/group
3018 : * semantics for their data types.
3019 : */
3020 12754 : foreach(tlitem, *targetlist)
3021 : {
3022 9444 : TargetEntry *tle = (TargetEntry *) lfirst(tlitem);
3023 :
3024 9444 : if (tle->resjunk)
3025 4 : continue; /* ignore junk */
3026 9440 : result = addTargetToGroupList(pstate, tle,
3027 : result, *targetlist,
3028 9440 : exprLocation((Node *) tle->expr));
3029 : }
3030 :
3031 : /*
3032 : * Complain if we found nothing to make DISTINCT. Returning an empty list
3033 : * would cause the parsed Query to look like it didn't have DISTINCT, with
3034 : * results that would probably surprise the user. Note: this case is
3035 : * presently impossible for aggregates because of grammar restrictions,
3036 : * but we check anyway.
3037 : */
3038 3310 : if (result == NIL)
3039 0 : ereport(ERROR,
3040 : (errcode(ERRCODE_SYNTAX_ERROR),
3041 : is_agg ?
3042 : errmsg("an aggregate with DISTINCT must have at least one argument") :
3043 : errmsg("SELECT DISTINCT must have at least one column")));
3044 :
3045 3310 : return result;
3046 : }
3047 :
3048 : /*
3049 : * transformDistinctOnClause -
3050 : * transform a DISTINCT ON clause
3051 : *
3052 : * Since we may need to add items to the query's targetlist, that list
3053 : * is passed by reference.
3054 : *
3055 : * As with GROUP BY, we absorb the sorting semantics of ORDER BY as much as
3056 : * possible into the distinctClause. This avoids a possible need to re-sort,
3057 : * and allows the user to choose the equality semantics used by DISTINCT,
3058 : * should she be working with a datatype that has more than one equality
3059 : * operator.
3060 : */
3061 : List *
3062 178 : transformDistinctOnClause(ParseState *pstate, List *distinctlist,
3063 : List **targetlist, List *sortClause)
3064 : {
3065 178 : List *result = NIL;
3066 178 : List *sortgrouprefs = NIL;
3067 : bool skipped_sortitem;
3068 : ListCell *lc;
3069 : ListCell *lc2;
3070 :
3071 : /*
3072 : * Add all the DISTINCT ON expressions to the tlist (if not already
3073 : * present, they are added as resjunk items). Assign sortgroupref numbers
3074 : * to them, and make a list of these numbers. (NB: we rely below on the
3075 : * sortgrouprefs list being one-for-one with the original distinctlist.
3076 : * Also notice that we could have duplicate DISTINCT ON expressions and
3077 : * hence duplicate entries in sortgrouprefs.)
3078 : */
3079 416 : foreach(lc, distinctlist)
3080 : {
3081 244 : Node *dexpr = (Node *) lfirst(lc);
3082 : int sortgroupref;
3083 : TargetEntry *tle;
3084 :
3085 244 : tle = findTargetlistEntrySQL92(pstate, dexpr, targetlist,
3086 : EXPR_KIND_DISTINCT_ON);
3087 238 : sortgroupref = assignSortGroupRef(tle, *targetlist);
3088 238 : sortgrouprefs = lappend_int(sortgrouprefs, sortgroupref);
3089 : }
3090 :
3091 : /*
3092 : * If the user writes both DISTINCT ON and ORDER BY, adopt the sorting
3093 : * semantics from ORDER BY items that match DISTINCT ON items, and also
3094 : * adopt their column sort order. We insist that the distinctClause and
3095 : * sortClause match, so throw error if we find the need to add any more
3096 : * distinctClause items after we've skipped an ORDER BY item that wasn't
3097 : * in DISTINCT ON.
3098 : */
3099 172 : skipped_sortitem = false;
3100 404 : foreach(lc, sortClause)
3101 : {
3102 238 : SortGroupClause *scl = (SortGroupClause *) lfirst(lc);
3103 :
3104 238 : if (list_member_int(sortgrouprefs, scl->tleSortGroupRef))
3105 : {
3106 178 : if (skipped_sortitem)
3107 6 : ereport(ERROR,
3108 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3109 : errmsg("SELECT DISTINCT ON expressions must match initial ORDER BY expressions"),
3110 : parser_errposition(pstate,
3111 : get_matching_location(scl->tleSortGroupRef,
3112 : sortgrouprefs,
3113 : distinctlist))));
3114 : else
3115 172 : result = lappend(result, copyObject(scl));
3116 : }
3117 : else
3118 60 : skipped_sortitem = true;
3119 : }
3120 :
3121 : /*
3122 : * Now add any remaining DISTINCT ON items, using default sort/group
3123 : * semantics for their data types. (Note: this is pretty questionable; if
3124 : * the ORDER BY list doesn't include all the DISTINCT ON items and more
3125 : * besides, you certainly aren't using DISTINCT ON in the intended way,
3126 : * and you probably aren't going to get consistent results. It might be
3127 : * better to throw an error or warning here. But historically we've
3128 : * allowed it, so keep doing so.)
3129 : */
3130 392 : forboth(lc, distinctlist, lc2, sortgrouprefs)
3131 : {
3132 226 : Node *dexpr = (Node *) lfirst(lc);
3133 226 : int sortgroupref = lfirst_int(lc2);
3134 226 : TargetEntry *tle = get_sortgroupref_tle(sortgroupref, *targetlist);
3135 :
3136 226 : if (targetIsInSortList(tle, InvalidOid, result))
3137 166 : continue; /* already in list (with some semantics) */
3138 60 : if (skipped_sortitem)
3139 0 : ereport(ERROR,
3140 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3141 : errmsg("SELECT DISTINCT ON expressions must match initial ORDER BY expressions"),
3142 : parser_errposition(pstate, exprLocation(dexpr))));
3143 60 : result = addTargetToGroupList(pstate, tle,
3144 : result, *targetlist,
3145 : exprLocation(dexpr));
3146 : }
3147 :
3148 : /*
3149 : * An empty result list is impossible here because of grammar
3150 : * restrictions.
3151 : */
3152 : Assert(result != NIL);
3153 :
3154 166 : return result;
3155 : }
3156 :
3157 : /*
3158 : * get_matching_location
3159 : * Get the exprLocation of the exprs member corresponding to the
3160 : * (first) member of sortgrouprefs that equals sortgroupref.
3161 : *
3162 : * This is used so that we can point at a troublesome DISTINCT ON entry.
3163 : * (Note that we need to use the original untransformed DISTINCT ON list
3164 : * item, as whatever TLE it corresponds to will very possibly have a
3165 : * parse location pointing to some matching entry in the SELECT list
3166 : * or ORDER BY list.)
3167 : */
3168 : static int
3169 6 : get_matching_location(int sortgroupref, List *sortgrouprefs, List *exprs)
3170 : {
3171 : ListCell *lcs;
3172 : ListCell *lce;
3173 :
3174 12 : forboth(lcs, sortgrouprefs, lce, exprs)
3175 : {
3176 12 : if (lfirst_int(lcs) == sortgroupref)
3177 6 : return exprLocation((Node *) lfirst(lce));
3178 : }
3179 : /* if no match, caller blew it */
3180 0 : elog(ERROR, "get_matching_location: no matching sortgroupref");
3181 : return -1; /* keep compiler quiet */
3182 : }
3183 :
3184 : /*
3185 : * resolve_unique_index_expr
3186 : * Infer a unique index from a list of indexElems, for ON
3187 : * CONFLICT clause
3188 : *
3189 : * Perform parse analysis of expressions and columns appearing within ON
3190 : * CONFLICT clause. During planning, the returned list of expressions is used
3191 : * to infer which unique index to use.
3192 : */
3193 : static List *
3194 1262 : resolve_unique_index_expr(ParseState *pstate, InferClause *infer,
3195 : Relation heapRel)
3196 : {
3197 1262 : List *result = NIL;
3198 : ListCell *l;
3199 :
3200 2816 : foreach(l, infer->indexElems)
3201 : {
3202 1560 : IndexElem *ielem = (IndexElem *) lfirst(l);
3203 1560 : InferenceElem *pInfer = makeNode(InferenceElem);
3204 : Node *parse;
3205 :
3206 : /*
3207 : * Raw grammar re-uses CREATE INDEX infrastructure for unique index
3208 : * inference clause, and so will accept opclasses by name and so on.
3209 : *
3210 : * Make no attempt to match ASC or DESC ordering or NULLS FIRST/NULLS
3211 : * LAST ordering, since those are not significant for inference
3212 : * purposes (any unique index matching the inference specification in
3213 : * other regards is accepted indifferently). Actively reject this as
3214 : * wrong-headed.
3215 : */
3216 1560 : if (ielem->ordering != SORTBY_DEFAULT)
3217 0 : ereport(ERROR,
3218 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3219 : errmsg("ASC/DESC is not allowed in ON CONFLICT clause"),
3220 : parser_errposition(pstate,
3221 : exprLocation((Node *) infer))));
3222 1560 : if (ielem->nulls_ordering != SORTBY_NULLS_DEFAULT)
3223 0 : ereport(ERROR,
3224 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3225 : errmsg("NULLS FIRST/LAST is not allowed in ON CONFLICT clause"),
3226 : parser_errposition(pstate,
3227 : exprLocation((Node *) infer))));
3228 :
3229 1560 : if (!ielem->expr)
3230 : {
3231 : /* Simple index attribute */
3232 : ColumnRef *n;
3233 :
3234 : /*
3235 : * Grammar won't have built raw expression for us in event of
3236 : * plain column reference. Create one directly, and perform
3237 : * expression transformation. Planner expects this, and performs
3238 : * its own normalization for the purposes of matching against
3239 : * pg_index.
3240 : */
3241 1398 : n = makeNode(ColumnRef);
3242 1398 : n->fields = list_make1(makeString(ielem->name));
3243 : /* Location is approximately that of inference specification */
3244 1398 : n->location = infer->location;
3245 1398 : parse = (Node *) n;
3246 : }
3247 : else
3248 : {
3249 : /* Do parse transformation of the raw expression */
3250 162 : parse = (Node *) ielem->expr;
3251 : }
3252 :
3253 : /*
3254 : * transformExpr() will reject subqueries, aggregates, window
3255 : * functions, and SRFs, based on being passed
3256 : * EXPR_KIND_INDEX_EXPRESSION. So we needn't worry about those
3257 : * further ... not that they would match any available index
3258 : * expression anyway.
3259 : */
3260 1560 : pInfer->expr = transformExpr(pstate, parse, EXPR_KIND_INDEX_EXPRESSION);
3261 :
3262 : /* Perform lookup of collation and operator class as required */
3263 1554 : if (!ielem->collation)
3264 1512 : pInfer->infercollid = InvalidOid;
3265 : else
3266 42 : pInfer->infercollid = LookupCollation(pstate, ielem->collation,
3267 42 : exprLocation(pInfer->expr));
3268 :
3269 1554 : if (!ielem->opclass)
3270 1512 : pInfer->inferopclass = InvalidOid;
3271 : else
3272 42 : pInfer->inferopclass = get_opclass_oid(BTREE_AM_OID,
3273 : ielem->opclass, false);
3274 :
3275 1554 : result = lappend(result, pInfer);
3276 : }
3277 :
3278 1256 : return result;
3279 : }
3280 :
3281 : /*
3282 : * transformOnConflictArbiter -
3283 : * transform arbiter expressions in an ON CONFLICT clause.
3284 : *
3285 : * Transformed expressions used to infer one unique index relation to serve as
3286 : * an ON CONFLICT arbiter. Partial unique indexes may be inferred using WHERE
3287 : * clause from inference specification clause.
3288 : */
3289 : void
3290 1472 : transformOnConflictArbiter(ParseState *pstate,
3291 : OnConflictClause *onConflictClause,
3292 : List **arbiterExpr, Node **arbiterWhere,
3293 : Oid *constraint)
3294 : {
3295 1472 : InferClause *infer = onConflictClause->infer;
3296 :
3297 1472 : *arbiterExpr = NIL;
3298 1472 : *arbiterWhere = NULL;
3299 1472 : *constraint = InvalidOid;
3300 :
3301 1472 : if (onConflictClause->action == ONCONFLICT_UPDATE && !infer)
3302 6 : ereport(ERROR,
3303 : (errcode(ERRCODE_SYNTAX_ERROR),
3304 : errmsg("ON CONFLICT DO UPDATE requires inference specification or constraint name"),
3305 : errhint("For example, ON CONFLICT (column_name)."),
3306 : parser_errposition(pstate,
3307 : exprLocation((Node *) onConflictClause))));
3308 :
3309 : /*
3310 : * To simplify certain aspects of its design, speculative insertion into
3311 : * system catalogs is disallowed
3312 : */
3313 1466 : if (IsCatalogRelation(pstate->p_target_relation))
3314 0 : ereport(ERROR,
3315 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3316 : errmsg("ON CONFLICT is not supported with system catalog tables"),
3317 : parser_errposition(pstate,
3318 : exprLocation((Node *) onConflictClause))));
3319 :
3320 : /* Same applies to table used by logical decoding as catalog table */
3321 1466 : if (RelationIsUsedAsCatalogTable(pstate->p_target_relation))
3322 0 : ereport(ERROR,
3323 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3324 : errmsg("ON CONFLICT is not supported on table \"%s\" used as a catalog table",
3325 : RelationGetRelationName(pstate->p_target_relation)),
3326 : parser_errposition(pstate,
3327 : exprLocation((Node *) onConflictClause))));
3328 :
3329 : /* ON CONFLICT DO NOTHING does not require an inference clause */
3330 1466 : if (infer)
3331 : {
3332 1310 : if (infer->indexElems)
3333 1262 : *arbiterExpr = resolve_unique_index_expr(pstate, infer,
3334 : pstate->p_target_relation);
3335 :
3336 : /*
3337 : * Handling inference WHERE clause (for partial unique index
3338 : * inference)
3339 : */
3340 1304 : if (infer->whereClause)
3341 42 : *arbiterWhere = transformExpr(pstate, infer->whereClause,
3342 : EXPR_KIND_INDEX_PREDICATE);
3343 :
3344 : /*
3345 : * If the arbiter is specified by constraint name, get the constraint
3346 : * OID and mark the constrained columns as requiring SELECT privilege,
3347 : * in the same way as would have happened if the arbiter had been
3348 : * specified by explicit reference to the constraint's index columns.
3349 : */
3350 1304 : if (infer->conname)
3351 : {
3352 48 : Oid relid = RelationGetRelid(pstate->p_target_relation);
3353 48 : RTEPermissionInfo *perminfo = pstate->p_target_nsitem->p_perminfo;
3354 : Bitmapset *conattnos;
3355 :
3356 48 : conattnos = get_relation_constraint_attnos(relid, infer->conname,
3357 : false, constraint);
3358 :
3359 : /* Make sure the rel as a whole is marked for SELECT access */
3360 48 : perminfo->requiredPerms |= ACL_SELECT;
3361 : /* Mark the constrained columns as requiring SELECT access */
3362 48 : perminfo->selectedCols = bms_add_members(perminfo->selectedCols,
3363 : conattnos);
3364 : }
3365 : }
3366 :
3367 : /*
3368 : * It's convenient to form a list of expressions based on the
3369 : * representation used by CREATE INDEX, since the same restrictions are
3370 : * appropriate (e.g. on subqueries). However, from here on, a dedicated
3371 : * primnode representation is used for inference elements, and so
3372 : * assign_query_collations() can be trusted to do the right thing with the
3373 : * post parse analysis query tree inference clause representation.
3374 : */
3375 1460 : }
3376 :
3377 : /*
3378 : * addTargetToSortList
3379 : * If the given targetlist entry isn't already in the SortGroupClause
3380 : * list, add it to the end of the list, using the given sort ordering
3381 : * info.
3382 : *
3383 : * Returns the updated SortGroupClause list.
3384 : */
3385 : List *
3386 84626 : addTargetToSortList(ParseState *pstate, TargetEntry *tle,
3387 : List *sortlist, List *targetlist, SortBy *sortby)
3388 : {
3389 84626 : Oid restype = exprType((Node *) tle->expr);
3390 : Oid sortop;
3391 : Oid eqop;
3392 : bool hashable;
3393 : bool reverse;
3394 : int location;
3395 : ParseCallbackState pcbstate;
3396 :
3397 : /* if tlist item is an UNKNOWN literal, change it to TEXT */
3398 84626 : if (restype == UNKNOWNOID)
3399 : {
3400 12 : tle->expr = (Expr *) coerce_type(pstate, (Node *) tle->expr,
3401 : restype, TEXTOID, -1,
3402 : COERCION_IMPLICIT,
3403 : COERCE_IMPLICIT_CAST,
3404 : -1);
3405 12 : restype = TEXTOID;
3406 : }
3407 :
3408 : /*
3409 : * Rather than clutter the API of get_sort_group_operators and the other
3410 : * functions we're about to use, make use of error context callback to
3411 : * mark any error reports with a parse position. We point to the operator
3412 : * location if present, else to the expression being sorted. (NB: use the
3413 : * original untransformed expression here; the TLE entry might well point
3414 : * at a duplicate expression in the regular SELECT list.)
3415 : */
3416 84626 : location = sortby->location;
3417 84626 : if (location < 0)
3418 84406 : location = exprLocation(sortby->node);
3419 84626 : setup_parser_errposition_callback(&pcbstate, pstate, location);
3420 :
3421 : /* determine the sortop, eqop, and directionality */
3422 84626 : switch (sortby->sortby_dir)
3423 : {
3424 81802 : case SORTBY_DEFAULT:
3425 : case SORTBY_ASC:
3426 81802 : get_sort_group_operators(restype,
3427 : true, true, false,
3428 : &sortop, &eqop, NULL,
3429 : &hashable);
3430 81796 : reverse = false;
3431 81796 : break;
3432 2604 : case SORTBY_DESC:
3433 2604 : get_sort_group_operators(restype,
3434 : false, true, true,
3435 : NULL, &eqop, &sortop,
3436 : &hashable);
3437 2604 : reverse = true;
3438 2604 : break;
3439 220 : case SORTBY_USING:
3440 : Assert(sortby->useOp != NIL);
3441 220 : sortop = compatible_oper_opid(sortby->useOp,
3442 : restype,
3443 : restype,
3444 : false);
3445 :
3446 : /*
3447 : * Verify it's a valid ordering operator, fetch the corresponding
3448 : * equality operator, and determine whether to consider it like
3449 : * ASC or DESC.
3450 : */
3451 220 : eqop = get_equality_op_for_ordering_op(sortop, &reverse);
3452 220 : if (!OidIsValid(eqop))
3453 0 : ereport(ERROR,
3454 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
3455 : errmsg("operator %s is not a valid ordering operator",
3456 : strVal(llast(sortby->useOp))),
3457 : errhint("Ordering operators must be \"<\" or \">\" members of btree operator families.")));
3458 :
3459 : /*
3460 : * Also see if the equality operator is hashable.
3461 : */
3462 220 : hashable = op_hashjoinable(eqop, restype);
3463 220 : break;
3464 0 : default:
3465 0 : elog(ERROR, "unrecognized sortby_dir: %d", sortby->sortby_dir);
3466 : sortop = InvalidOid; /* keep compiler quiet */
3467 : eqop = InvalidOid;
3468 : hashable = false;
3469 : reverse = false;
3470 : break;
3471 : }
3472 :
3473 84620 : cancel_parser_errposition_callback(&pcbstate);
3474 :
3475 : /* avoid making duplicate sortlist entries */
3476 84620 : if (!targetIsInSortList(tle, sortop, sortlist))
3477 : {
3478 84620 : SortGroupClause *sortcl = makeNode(SortGroupClause);
3479 :
3480 84620 : sortcl->tleSortGroupRef = assignSortGroupRef(tle, targetlist);
3481 :
3482 84620 : sortcl->eqop = eqop;
3483 84620 : sortcl->sortop = sortop;
3484 84620 : sortcl->hashable = hashable;
3485 :
3486 84620 : switch (sortby->sortby_nulls)
3487 : {
3488 82678 : case SORTBY_NULLS_DEFAULT:
3489 : /* NULLS FIRST is default for DESC; other way for ASC */
3490 82678 : sortcl->nulls_first = reverse;
3491 82678 : break;
3492 274 : case SORTBY_NULLS_FIRST:
3493 274 : sortcl->nulls_first = true;
3494 274 : break;
3495 1668 : case SORTBY_NULLS_LAST:
3496 1668 : sortcl->nulls_first = false;
3497 1668 : break;
3498 0 : default:
3499 0 : elog(ERROR, "unrecognized sortby_nulls: %d",
3500 : sortby->sortby_nulls);
3501 : break;
3502 : }
3503 :
3504 84620 : sortlist = lappend(sortlist, sortcl);
3505 : }
3506 :
3507 84620 : return sortlist;
3508 : }
3509 :
3510 : /*
3511 : * addTargetToGroupList
3512 : * If the given targetlist entry isn't already in the SortGroupClause
3513 : * list, add it to the end of the list, using default sort/group
3514 : * semantics.
3515 : *
3516 : * This is very similar to addTargetToSortList, except that we allow the
3517 : * case where only a grouping (equality) operator can be found, and that
3518 : * the TLE is considered "already in the list" if it appears there with any
3519 : * sorting semantics.
3520 : *
3521 : * location is the parse location to be fingered in event of trouble. Note
3522 : * that we can't rely on exprLocation(tle->expr), because that might point
3523 : * to a SELECT item that matches the GROUP BY item; it'd be pretty confusing
3524 : * to report such a location.
3525 : *
3526 : * Returns the updated SortGroupClause list.
3527 : */
3528 : static List *
3529 15588 : addTargetToGroupList(ParseState *pstate, TargetEntry *tle,
3530 : List *grouplist, List *targetlist, int location)
3531 : {
3532 15588 : Oid restype = exprType((Node *) tle->expr);
3533 :
3534 : /* if tlist item is an UNKNOWN literal, change it to TEXT */
3535 15588 : if (restype == UNKNOWNOID)
3536 : {
3537 16 : tle->expr = (Expr *) coerce_type(pstate, (Node *) tle->expr,
3538 : restype, TEXTOID, -1,
3539 : COERCION_IMPLICIT,
3540 : COERCE_IMPLICIT_CAST,
3541 : -1);
3542 16 : restype = TEXTOID;
3543 : }
3544 :
3545 : /* avoid making duplicate grouplist entries */
3546 15588 : if (!targetIsInSortList(tle, InvalidOid, grouplist))
3547 : {
3548 15014 : SortGroupClause *grpcl = makeNode(SortGroupClause);
3549 : Oid sortop;
3550 : Oid eqop;
3551 : bool hashable;
3552 : ParseCallbackState pcbstate;
3553 :
3554 15014 : setup_parser_errposition_callback(&pcbstate, pstate, location);
3555 :
3556 : /* determine the eqop and optional sortop */
3557 15014 : get_sort_group_operators(restype,
3558 : false, true, false,
3559 : &sortop, &eqop, NULL,
3560 : &hashable);
3561 :
3562 15014 : cancel_parser_errposition_callback(&pcbstate);
3563 :
3564 15014 : grpcl->tleSortGroupRef = assignSortGroupRef(tle, targetlist);
3565 15014 : grpcl->eqop = eqop;
3566 15014 : grpcl->sortop = sortop;
3567 15014 : grpcl->nulls_first = false; /* OK with or without sortop */
3568 15014 : grpcl->hashable = hashable;
3569 :
3570 15014 : grouplist = lappend(grouplist, grpcl);
3571 : }
3572 :
3573 15588 : return grouplist;
3574 : }
3575 :
3576 : /*
3577 : * assignSortGroupRef
3578 : * Assign the targetentry an unused ressortgroupref, if it doesn't
3579 : * already have one. Return the assigned or pre-existing refnumber.
3580 : *
3581 : * 'tlist' is the targetlist containing (or to contain) the given targetentry.
3582 : */
3583 : Index
3584 130882 : assignSortGroupRef(TargetEntry *tle, List *tlist)
3585 : {
3586 : Index maxRef;
3587 : ListCell *l;
3588 :
3589 130882 : if (tle->ressortgroupref) /* already has one? */
3590 746 : return tle->ressortgroupref;
3591 :
3592 : /* easiest way to pick an unused refnumber: max used + 1 */
3593 130136 : maxRef = 0;
3594 734980 : foreach(l, tlist)
3595 : {
3596 604844 : Index ref = ((TargetEntry *) lfirst(l))->ressortgroupref;
3597 :
3598 604844 : if (ref > maxRef)
3599 88672 : maxRef = ref;
3600 : }
3601 130136 : tle->ressortgroupref = maxRef + 1;
3602 130136 : return tle->ressortgroupref;
3603 : }
3604 :
3605 : /*
3606 : * targetIsInSortList
3607 : * Is the given target item already in the sortlist?
3608 : * If sortop is not InvalidOid, also test for a match to the sortop.
3609 : *
3610 : * It is not an oversight that this function ignores the nulls_first flag.
3611 : * We check sortop when determining if an ORDER BY item is redundant with
3612 : * earlier ORDER BY items, because it's conceivable that "ORDER BY
3613 : * foo USING <, foo USING <<<" is not redundant, if <<< distinguishes
3614 : * values that < considers equal. We need not check nulls_first
3615 : * however, because a lower-order column with the same sortop but
3616 : * opposite nulls direction is redundant. Also, we can consider
3617 : * ORDER BY foo ASC, foo DESC redundant, so check for a commutator match.
3618 : *
3619 : * Works for both ordering and grouping lists (sortop would normally be
3620 : * InvalidOid when considering grouping). Note that the main reason we need
3621 : * this routine (and not just a quick test for nonzeroness of ressortgroupref)
3622 : * is that a TLE might be in only one of the lists.
3623 : */
3624 : bool
3625 104582 : targetIsInSortList(TargetEntry *tle, Oid sortop, List *sortList)
3626 : {
3627 104582 : Index ref = tle->ressortgroupref;
3628 : ListCell *l;
3629 :
3630 : /* no need to scan list if tle has no marker */
3631 104582 : if (ref == 0)
3632 100232 : return false;
3633 :
3634 5510 : foreach(l, sortList)
3635 : {
3636 2486 : SortGroupClause *scl = (SortGroupClause *) lfirst(l);
3637 :
3638 2486 : if (scl->tleSortGroupRef == ref &&
3639 0 : (sortop == InvalidOid ||
3640 0 : sortop == scl->sortop ||
3641 0 : sortop == get_commutator(scl->sortop)))
3642 1326 : return true;
3643 : }
3644 3024 : return false;
3645 : }
3646 :
3647 : /*
3648 : * findWindowClause
3649 : * Find the named WindowClause in the list, or return NULL if not there
3650 : */
3651 : static WindowClause *
3652 564 : findWindowClause(List *wclist, const char *name)
3653 : {
3654 : ListCell *l;
3655 :
3656 570 : foreach(l, wclist)
3657 : {
3658 36 : WindowClause *wc = (WindowClause *) lfirst(l);
3659 :
3660 36 : if (wc->name && strcmp(wc->name, name) == 0)
3661 30 : return wc;
3662 : }
3663 :
3664 534 : return NULL;
3665 : }
3666 :
3667 : /*
3668 : * transformFrameOffset
3669 : * Process a window frame offset expression
3670 : *
3671 : * In RANGE mode, rangeopfamily is the sort opfamily for the input ORDER BY
3672 : * column, and rangeopcintype is the input data type the sort operator is
3673 : * registered with. We expect the in_range function to be registered with
3674 : * that same type. (In binary-compatible cases, it might be different from
3675 : * the input column's actual type, so we can't use that for the lookups.)
3676 : * We'll return the OID of the in_range function to *inRangeFunc.
3677 : */
3678 : static Node *
3679 5256 : transformFrameOffset(ParseState *pstate, int frameOptions,
3680 : Oid rangeopfamily, Oid rangeopcintype, Oid *inRangeFunc,
3681 : Node *clause)
3682 : {
3683 5256 : const char *constructName = NULL;
3684 : Node *node;
3685 :
3686 5256 : *inRangeFunc = InvalidOid; /* default result */
3687 :
3688 : /* Quick exit if no offset expression */
3689 5256 : if (clause == NULL)
3690 3426 : return NULL;
3691 :
3692 1830 : if (frameOptions & FRAMEOPTION_ROWS)
3693 : {
3694 : /* Transform the raw expression tree */
3695 396 : node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_ROWS);
3696 :
3697 : /*
3698 : * Like LIMIT clause, simply coerce to int8
3699 : */
3700 396 : constructName = "ROWS";
3701 396 : node = coerce_to_specific_type(pstate, node, INT8OID, constructName);
3702 : }
3703 1434 : else if (frameOptions & FRAMEOPTION_RANGE)
3704 : {
3705 : /*
3706 : * We must look up the in_range support function that's to be used,
3707 : * possibly choosing one of several, and coerce the "offset" value to
3708 : * the appropriate input type.
3709 : */
3710 : Oid nodeType;
3711 : Oid preferredType;
3712 1152 : int nfuncs = 0;
3713 1152 : int nmatches = 0;
3714 1152 : Oid selectedType = InvalidOid;
3715 1152 : Oid selectedFunc = InvalidOid;
3716 : CatCList *proclist;
3717 : int i;
3718 :
3719 : /* Transform the raw expression tree */
3720 1152 : node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_RANGE);
3721 1152 : nodeType = exprType(node);
3722 :
3723 : /*
3724 : * If there are multiple candidates, we'll prefer the one that exactly
3725 : * matches nodeType; or if nodeType is as yet unknown, prefer the one
3726 : * that exactly matches the sort column type. (The second rule is
3727 : * like what we do for "known_type operator unknown".)
3728 : */
3729 1152 : preferredType = (nodeType != UNKNOWNOID) ? nodeType : rangeopcintype;
3730 :
3731 : /* Find the in_range support functions applicable to this case */
3732 1152 : proclist = SearchSysCacheList2(AMPROCNUM,
3733 : ObjectIdGetDatum(rangeopfamily),
3734 : ObjectIdGetDatum(rangeopcintype));
3735 7344 : for (i = 0; i < proclist->n_members; i++)
3736 : {
3737 6192 : HeapTuple proctup = &proclist->members[i]->tuple;
3738 6192 : Form_pg_amproc procform = (Form_pg_amproc) GETSTRUCT(proctup);
3739 :
3740 : /* The search will find all support proc types; ignore others */
3741 6192 : if (procform->amprocnum != BTINRANGE_PROC)
3742 4434 : continue;
3743 1758 : nfuncs++;
3744 :
3745 : /* Ignore function if given value can't be coerced to that type */
3746 1758 : if (!can_coerce_type(1, &nodeType, &procform->amprocrighttype,
3747 : COERCION_IMPLICIT))
3748 330 : continue;
3749 1428 : nmatches++;
3750 :
3751 : /* Remember preferred match, or any match if didn't find that */
3752 1428 : if (selectedType != preferredType)
3753 : {
3754 1368 : selectedType = procform->amprocrighttype;
3755 1368 : selectedFunc = procform->amproc;
3756 : }
3757 : }
3758 1152 : ReleaseCatCacheList(proclist);
3759 :
3760 : /*
3761 : * Throw error if needed. It seems worth taking the trouble to
3762 : * distinguish "no support at all" from "you didn't match any
3763 : * available offset type".
3764 : */
3765 1152 : if (nfuncs == 0)
3766 6 : ereport(ERROR,
3767 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3768 : errmsg("RANGE with offset PRECEDING/FOLLOWING is not supported for column type %s",
3769 : format_type_be(rangeopcintype)),
3770 : parser_errposition(pstate, exprLocation(node))));
3771 1146 : if (nmatches == 0)
3772 18 : ereport(ERROR,
3773 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3774 : errmsg("RANGE with offset PRECEDING/FOLLOWING is not supported for column type %s and offset type %s",
3775 : format_type_be(rangeopcintype),
3776 : format_type_be(nodeType)),
3777 : errhint("Cast the offset value to an appropriate type."),
3778 : parser_errposition(pstate, exprLocation(node))));
3779 1128 : if (nmatches != 1 && selectedType != preferredType)
3780 0 : ereport(ERROR,
3781 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3782 : errmsg("RANGE with offset PRECEDING/FOLLOWING has multiple interpretations for column type %s and offset type %s",
3783 : format_type_be(rangeopcintype),
3784 : format_type_be(nodeType)),
3785 : errhint("Cast the offset value to the exact intended type."),
3786 : parser_errposition(pstate, exprLocation(node))));
3787 :
3788 : /* OK, coerce the offset to the right type */
3789 1128 : constructName = "RANGE";
3790 1128 : node = coerce_to_specific_type(pstate, node,
3791 : selectedType, constructName);
3792 1128 : *inRangeFunc = selectedFunc;
3793 : }
3794 282 : else if (frameOptions & FRAMEOPTION_GROUPS)
3795 : {
3796 : /* Transform the raw expression tree */
3797 282 : node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_GROUPS);
3798 :
3799 : /*
3800 : * Like LIMIT clause, simply coerce to int8
3801 : */
3802 282 : constructName = "GROUPS";
3803 282 : node = coerce_to_specific_type(pstate, node, INT8OID, constructName);
3804 : }
3805 : else
3806 : {
3807 : Assert(false);
3808 0 : node = NULL;
3809 : }
3810 :
3811 : /* Disallow variables in frame offsets */
3812 1806 : checkExprIsVarFree(pstate, node, constructName);
3813 :
3814 1800 : return node;
3815 : }
|