Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * parse_agg.c
4 : * handle aggregates and window functions in parser
5 : *
6 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/parser/parse_agg.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 : #include "postgres.h"
16 :
17 : #include "access/htup_details.h"
18 : #include "catalog/pg_aggregate.h"
19 : #include "catalog/pg_constraint.h"
20 : #include "catalog/pg_type.h"
21 : #include "common/int.h"
22 : #include "nodes/makefuncs.h"
23 : #include "nodes/nodeFuncs.h"
24 : #include "optimizer/optimizer.h"
25 : #include "parser/parse_agg.h"
26 : #include "parser/parse_clause.h"
27 : #include "parser/parse_coerce.h"
28 : #include "parser/parse_expr.h"
29 : #include "parser/parse_relation.h"
30 : #include "parser/parsetree.h"
31 : #include "rewrite/rewriteManip.h"
32 : #include "utils/builtins.h"
33 : #include "utils/lsyscache.h"
34 : #include "utils/syscache.h"
35 :
36 : typedef struct
37 : {
38 : ParseState *pstate;
39 : int min_varlevel;
40 : int min_agglevel;
41 : int min_ctelevel;
42 : RangeTblEntry *min_cte;
43 : int sublevels_up;
44 : } check_agg_arguments_context;
45 :
46 : typedef struct
47 : {
48 : ParseState *pstate;
49 : Query *qry;
50 : bool hasJoinRTEs;
51 : List *groupClauses; /* list of TargetEntry */
52 : List *groupClauseCommonVars; /* list of Vars */
53 : List *groupClauseSubLevels; /* list of lists of TargetEntry */
54 : List *gset_common; /* integer list of sortgrouprefs */
55 : bool have_non_var_grouping;
56 : List **func_grouped_rels;
57 : int sublevels_up;
58 : bool in_agg_direct_args;
59 : } substitute_grouped_columns_context;
60 :
61 : static int check_agg_arguments(ParseState *pstate,
62 : List *directargs,
63 : List *args,
64 : Expr *filter,
65 : int agglocation);
66 : static bool check_agg_arguments_walker(Node *node,
67 : check_agg_arguments_context *context);
68 : static Node *substitute_grouped_columns(Node *node, ParseState *pstate, Query *qry,
69 : List *groupClauses, List *groupClauseCommonVars,
70 : List *gset_common,
71 : bool have_non_var_grouping,
72 : List **func_grouped_rels);
73 : static Node *substitute_grouped_columns_mutator(Node *node,
74 : substitute_grouped_columns_context *context);
75 : static void finalize_grouping_exprs(Node *node, ParseState *pstate, Query *qry,
76 : List *groupClauses, bool hasJoinRTEs,
77 : bool have_non_var_grouping);
78 : static bool finalize_grouping_exprs_walker(Node *node,
79 : substitute_grouped_columns_context *context);
80 : static Var *buildGroupedVar(int attnum, Index ressortgroupref,
81 : substitute_grouped_columns_context *context);
82 : static void check_agglevels_and_constraints(ParseState *pstate, Node *expr);
83 : static List *expand_groupingset_node(GroupingSet *gs);
84 : static Node *make_agg_arg(Oid argtype, Oid argcollation);
85 :
86 :
87 : /*
88 : * transformAggregateCall -
89 : * Finish initial transformation of an aggregate call
90 : *
91 : * parse_func.c has recognized the function as an aggregate, and has set up
92 : * all the fields of the Aggref except aggargtypes, aggdirectargs, args,
93 : * aggorder, aggdistinct and agglevelsup. The passed-in args list has been
94 : * through standard expression transformation and type coercion to match the
95 : * agg's declared arg types, while the passed-in aggorder list hasn't been
96 : * transformed at all.
97 : *
98 : * Here we separate the args list into direct and aggregated args, storing the
99 : * former in agg->aggdirectargs and the latter in agg->args. The regular
100 : * args, but not the direct args, are converted into a targetlist by inserting
101 : * TargetEntry nodes. We then transform the aggorder and agg_distinct
102 : * specifications to produce lists of SortGroupClause nodes for agg->aggorder
103 : * and agg->aggdistinct. (For a regular aggregate, this might result in
104 : * adding resjunk expressions to the targetlist; but for ordered-set
105 : * aggregates the aggorder list will always be one-to-one with the aggregated
106 : * args.)
107 : *
108 : * We must also determine which query level the aggregate actually belongs to,
109 : * set agglevelsup accordingly, and mark p_hasAggs true in the corresponding
110 : * pstate level.
111 : */
112 : void
113 30390 : transformAggregateCall(ParseState *pstate, Aggref *agg,
114 : List *args, List *aggorder, bool agg_distinct)
115 : {
116 30390 : List *argtypes = NIL;
117 30390 : List *tlist = NIL;
118 30390 : List *torder = NIL;
119 30390 : List *tdistinct = NIL;
120 30390 : AttrNumber attno = 1;
121 : int save_next_resno;
122 : ListCell *lc;
123 :
124 30390 : if (AGGKIND_IS_ORDERED_SET(agg->aggkind))
125 : {
126 : /*
127 : * For an ordered-set agg, the args list includes direct args and
128 : * aggregated args; we must split them apart.
129 : */
130 208 : int numDirectArgs = list_length(args) - list_length(aggorder);
131 : List *aargs;
132 : ListCell *lc2;
133 :
134 : Assert(numDirectArgs >= 0);
135 :
136 208 : aargs = list_copy_tail(args, numDirectArgs);
137 208 : agg->aggdirectargs = list_truncate(args, numDirectArgs);
138 :
139 : /*
140 : * Build a tlist from the aggregated args, and make a sortlist entry
141 : * for each one. Note that the expressions in the SortBy nodes are
142 : * ignored (they are the raw versions of the transformed args); we are
143 : * just looking at the sort information in the SortBy nodes.
144 : */
145 451 : forboth(lc, aargs, lc2, aggorder)
146 : {
147 243 : Expr *arg = (Expr *) lfirst(lc);
148 243 : SortBy *sortby = (SortBy *) lfirst(lc2);
149 : TargetEntry *tle;
150 :
151 : /* We don't bother to assign column names to the entries */
152 243 : tle = makeTargetEntry(arg, attno++, NULL, false);
153 243 : tlist = lappend(tlist, tle);
154 :
155 243 : torder = addTargetToSortList(pstate, tle,
156 : torder, tlist, sortby);
157 : }
158 :
159 : /* Never any DISTINCT in an ordered-set agg */
160 : Assert(!agg_distinct);
161 : }
162 : else
163 : {
164 : /* Regular aggregate, so it has no direct args */
165 30182 : agg->aggdirectargs = NIL;
166 :
167 : /*
168 : * Transform the plain list of Exprs into a targetlist.
169 : */
170 52582 : foreach(lc, args)
171 : {
172 22400 : Expr *arg = (Expr *) lfirst(lc);
173 : TargetEntry *tle;
174 :
175 : /* We don't bother to assign column names to the entries */
176 22400 : tle = makeTargetEntry(arg, attno++, NULL, false);
177 22400 : tlist = lappend(tlist, tle);
178 : }
179 :
180 : /*
181 : * If we have an ORDER BY, transform it. This will add columns to the
182 : * tlist if they appear in ORDER BY but weren't already in the arg
183 : * list. They will be marked resjunk = true so we can tell them apart
184 : * from regular aggregate arguments later.
185 : *
186 : * We need to mess with p_next_resno since it will be used to number
187 : * any new targetlist entries.
188 : */
189 30182 : save_next_resno = pstate->p_next_resno;
190 30182 : pstate->p_next_resno = attno;
191 :
192 30182 : torder = transformSortClause(pstate,
193 : aggorder,
194 : &tlist,
195 : EXPR_KIND_ORDER_BY,
196 : true /* force SQL99 rules */ );
197 :
198 : /*
199 : * If we have DISTINCT, transform that to produce a distinctList.
200 : */
201 30182 : if (agg_distinct)
202 : {
203 364 : tdistinct = transformDistinctClause(pstate, &tlist, torder, true);
204 :
205 : /*
206 : * Remove this check if executor support for hashed distinct for
207 : * aggregates is ever added.
208 : */
209 800 : foreach(lc, tdistinct)
210 : {
211 460 : SortGroupClause *sortcl = (SortGroupClause *) lfirst(lc);
212 :
213 460 : if (!OidIsValid(sortcl->sortop))
214 : {
215 0 : Node *expr = get_sortgroupclause_expr(sortcl, tlist);
216 :
217 0 : ereport(ERROR,
218 : (errcode(ERRCODE_UNDEFINED_FUNCTION),
219 : errmsg("could not identify an ordering operator for type %s",
220 : format_type_be(exprType(expr))),
221 : errdetail("Aggregates with DISTINCT must be able to sort their inputs."),
222 : parser_errposition(pstate, exprLocation(expr))));
223 : }
224 : }
225 : }
226 :
227 30158 : pstate->p_next_resno = save_next_resno;
228 : }
229 :
230 : /* Update the Aggref with the transformation results */
231 30366 : agg->args = tlist;
232 30366 : agg->aggorder = torder;
233 30366 : agg->aggdistinct = tdistinct;
234 :
235 : /*
236 : * Now build the aggargtypes list with the type OIDs of the direct and
237 : * aggregated args, ignoring any resjunk entries that might have been
238 : * added by ORDER BY/DISTINCT processing. We can't do this earlier
239 : * because said processing can modify some args' data types, in particular
240 : * by resolving previously-unresolved "unknown" literals.
241 : */
242 30605 : foreach(lc, agg->aggdirectargs)
243 : {
244 239 : Expr *arg = (Expr *) lfirst(lc);
245 :
246 239 : argtypes = lappend_oid(argtypes, exprType((Node *) arg));
247 : }
248 53992 : foreach(lc, tlist)
249 : {
250 23626 : TargetEntry *tle = (TargetEntry *) lfirst(lc);
251 :
252 23626 : if (tle->resjunk)
253 1047 : continue; /* ignore junk */
254 22579 : argtypes = lappend_oid(argtypes, exprType((Node *) tle->expr));
255 : }
256 30366 : agg->aggargtypes = argtypes;
257 :
258 30366 : check_agglevels_and_constraints(pstate, (Node *) agg);
259 30250 : }
260 :
261 : /*
262 : * transformGroupingFunc
263 : * Transform a GROUPING expression
264 : *
265 : * GROUPING() behaves very like an aggregate. Processing of levels and nesting
266 : * is done as for aggregates. We set p_hasAggs for these expressions too.
267 : */
268 : Node *
269 252 : transformGroupingFunc(ParseState *pstate, GroupingFunc *p)
270 : {
271 : ListCell *lc;
272 252 : List *args = p->args;
273 252 : List *result_list = NIL;
274 252 : GroupingFunc *result = makeNode(GroupingFunc);
275 :
276 252 : if (list_length(args) > 31)
277 0 : ereport(ERROR,
278 : (errcode(ERRCODE_TOO_MANY_ARGUMENTS),
279 : errmsg("GROUPING must have fewer than 32 arguments"),
280 : parser_errposition(pstate, p->location)));
281 :
282 642 : foreach(lc, args)
283 : {
284 : Node *current_result;
285 :
286 390 : current_result = transformExpr(pstate, (Node *) lfirst(lc), pstate->p_expr_kind);
287 :
288 : /* acceptability of expressions is checked later */
289 :
290 390 : result_list = lappend(result_list, current_result);
291 : }
292 :
293 252 : result->args = result_list;
294 252 : result->location = p->location;
295 :
296 252 : check_agglevels_and_constraints(pstate, (Node *) result);
297 :
298 252 : return (Node *) result;
299 : }
300 :
301 : /*
302 : * Aggregate functions and grouping operations (which are combined in the spec
303 : * as <set function specification>) are very similar with regard to level and
304 : * nesting restrictions (though we allow a lot more things than the spec does).
305 : * Centralise those restrictions here.
306 : */
307 : static void
308 30618 : check_agglevels_and_constraints(ParseState *pstate, Node *expr)
309 : {
310 30618 : List *directargs = NIL;
311 30618 : List *args = NIL;
312 30618 : Expr *filter = NULL;
313 : int min_varlevel;
314 30618 : int location = -1;
315 : Index *p_levelsup;
316 : const char *err;
317 : bool errkind;
318 30618 : bool isAgg = IsA(expr, Aggref);
319 :
320 30618 : if (isAgg)
321 : {
322 30366 : Aggref *agg = (Aggref *) expr;
323 :
324 30366 : directargs = agg->aggdirectargs;
325 30366 : args = agg->args;
326 30366 : filter = agg->aggfilter;
327 30366 : location = agg->location;
328 30366 : p_levelsup = &agg->agglevelsup;
329 : }
330 : else
331 : {
332 252 : GroupingFunc *grp = (GroupingFunc *) expr;
333 :
334 252 : args = grp->args;
335 252 : location = grp->location;
336 252 : p_levelsup = &grp->agglevelsup;
337 : }
338 :
339 : /*
340 : * Check the arguments to compute the aggregate's level and detect
341 : * improper nesting.
342 : */
343 30618 : min_varlevel = check_agg_arguments(pstate,
344 : directargs,
345 : args,
346 : filter,
347 : location);
348 :
349 30582 : *p_levelsup = min_varlevel;
350 :
351 : /* Mark the correct pstate level as having aggregates */
352 30715 : while (min_varlevel-- > 0)
353 133 : pstate = pstate->parentParseState;
354 30582 : pstate->p_hasAggs = true;
355 :
356 : /*
357 : * Check to see if the aggregate function is in an invalid place within
358 : * its aggregation query.
359 : *
360 : * For brevity we support two schemes for reporting an error here: set
361 : * "err" to a custom message, or set "errkind" true if the error context
362 : * is sufficiently identified by what ParseExprKindName will return, *and*
363 : * what it will return is just a SQL keyword. (Otherwise, use a custom
364 : * message to avoid creating translation problems.)
365 : */
366 30582 : err = NULL;
367 30582 : errkind = false;
368 30582 : switch (pstate->p_expr_kind)
369 : {
370 0 : case EXPR_KIND_NONE:
371 : Assert(false); /* can't happen */
372 0 : break;
373 0 : case EXPR_KIND_OTHER:
374 :
375 : /*
376 : * Accept aggregate/grouping here; caller must throw error if
377 : * wanted
378 : */
379 0 : break;
380 0 : case EXPR_KIND_JOIN_ON:
381 : case EXPR_KIND_JOIN_USING:
382 0 : if (isAgg)
383 0 : err = _("aggregate functions are not allowed in JOIN conditions");
384 : else
385 0 : err = _("grouping operations are not allowed in JOIN conditions");
386 :
387 0 : break;
388 16 : case EXPR_KIND_FROM_SUBSELECT:
389 :
390 : /*
391 : * Aggregate/grouping scope rules make it worth being explicit
392 : * here
393 : */
394 16 : if (isAgg)
395 16 : err = _("aggregate functions are not allowed in FROM clause of their own query level");
396 : else
397 0 : err = _("grouping operations are not allowed in FROM clause of their own query level");
398 :
399 16 : break;
400 0 : case EXPR_KIND_FROM_FUNCTION:
401 0 : if (isAgg)
402 0 : err = _("aggregate functions are not allowed in functions in FROM");
403 : else
404 0 : err = _("grouping operations are not allowed in functions in FROM");
405 :
406 0 : break;
407 8 : case EXPR_KIND_WHERE:
408 8 : errkind = true;
409 8 : break;
410 4 : case EXPR_KIND_POLICY:
411 4 : if (isAgg)
412 4 : err = _("aggregate functions are not allowed in policy expressions");
413 : else
414 0 : err = _("grouping operations are not allowed in policy expressions");
415 :
416 4 : break;
417 436 : case EXPR_KIND_HAVING:
418 : /* okay */
419 436 : break;
420 8 : case EXPR_KIND_FILTER:
421 8 : errkind = true;
422 8 : break;
423 0 : case EXPR_KIND_WINDOW_PARTITION:
424 : /* okay */
425 0 : break;
426 8 : case EXPR_KIND_WINDOW_ORDER:
427 : /* okay */
428 8 : break;
429 0 : case EXPR_KIND_WINDOW_FRAME_RANGE:
430 0 : if (isAgg)
431 0 : err = _("aggregate functions are not allowed in window RANGE");
432 : else
433 0 : err = _("grouping operations are not allowed in window RANGE");
434 :
435 0 : break;
436 0 : case EXPR_KIND_WINDOW_FRAME_ROWS:
437 0 : if (isAgg)
438 0 : err = _("aggregate functions are not allowed in window ROWS");
439 : else
440 0 : err = _("grouping operations are not allowed in window ROWS");
441 :
442 0 : break;
443 0 : case EXPR_KIND_WINDOW_FRAME_GROUPS:
444 0 : if (isAgg)
445 0 : err = _("aggregate functions are not allowed in window GROUPS");
446 : else
447 0 : err = _("grouping operations are not allowed in window GROUPS");
448 :
449 0 : break;
450 30010 : case EXPR_KIND_SELECT_TARGET:
451 : /* okay */
452 30010 : break;
453 0 : case EXPR_KIND_INSERT_TARGET:
454 : case EXPR_KIND_UPDATE_SOURCE:
455 : case EXPR_KIND_UPDATE_TARGET:
456 0 : errkind = true;
457 0 : break;
458 0 : case EXPR_KIND_MERGE_WHEN:
459 0 : if (isAgg)
460 0 : err = _("aggregate functions are not allowed in MERGE WHEN conditions");
461 : else
462 0 : err = _("grouping operations are not allowed in MERGE WHEN conditions");
463 :
464 0 : break;
465 0 : case EXPR_KIND_GROUP_BY:
466 0 : errkind = true;
467 0 : break;
468 48 : case EXPR_KIND_ORDER_BY:
469 : /* okay */
470 48 : break;
471 0 : case EXPR_KIND_DISTINCT_ON:
472 : /* okay */
473 0 : break;
474 0 : case EXPR_KIND_LIMIT:
475 : case EXPR_KIND_OFFSET:
476 0 : errkind = true;
477 0 : break;
478 0 : case EXPR_KIND_RETURNING:
479 : case EXPR_KIND_MERGE_RETURNING:
480 0 : errkind = true;
481 0 : break;
482 0 : case EXPR_KIND_VALUES:
483 : case EXPR_KIND_VALUES_SINGLE:
484 0 : errkind = true;
485 0 : break;
486 0 : case EXPR_KIND_CHECK_CONSTRAINT:
487 : case EXPR_KIND_DOMAIN_CHECK:
488 0 : if (isAgg)
489 0 : err = _("aggregate functions are not allowed in check constraints");
490 : else
491 0 : err = _("grouping operations are not allowed in check constraints");
492 :
493 0 : break;
494 4 : case EXPR_KIND_COLUMN_DEFAULT:
495 : case EXPR_KIND_FUNCTION_DEFAULT:
496 :
497 4 : if (isAgg)
498 4 : err = _("aggregate functions are not allowed in DEFAULT expressions");
499 : else
500 0 : err = _("grouping operations are not allowed in DEFAULT expressions");
501 :
502 4 : break;
503 16 : case EXPR_KIND_INDEX_EXPRESSION:
504 16 : if (isAgg)
505 16 : err = _("aggregate functions are not allowed in index expressions");
506 : else
507 0 : err = _("grouping operations are not allowed in index expressions");
508 :
509 16 : break;
510 0 : case EXPR_KIND_INDEX_PREDICATE:
511 0 : if (isAgg)
512 0 : err = _("aggregate functions are not allowed in index predicates");
513 : else
514 0 : err = _("grouping operations are not allowed in index predicates");
515 :
516 0 : break;
517 0 : case EXPR_KIND_STATS_EXPRESSION:
518 0 : if (isAgg)
519 0 : err = _("aggregate functions are not allowed in statistics expressions");
520 : else
521 0 : err = _("grouping operations are not allowed in statistics expressions");
522 :
523 0 : break;
524 0 : case EXPR_KIND_ALTER_COL_TRANSFORM:
525 0 : if (isAgg)
526 0 : err = _("aggregate functions are not allowed in transform expressions");
527 : else
528 0 : err = _("grouping operations are not allowed in transform expressions");
529 :
530 0 : break;
531 0 : case EXPR_KIND_EXECUTE_PARAMETER:
532 0 : if (isAgg)
533 0 : err = _("aggregate functions are not allowed in EXECUTE parameters");
534 : else
535 0 : err = _("grouping operations are not allowed in EXECUTE parameters");
536 :
537 0 : break;
538 0 : case EXPR_KIND_TRIGGER_WHEN:
539 0 : if (isAgg)
540 0 : err = _("aggregate functions are not allowed in trigger WHEN conditions");
541 : else
542 0 : err = _("grouping operations are not allowed in trigger WHEN conditions");
543 :
544 0 : break;
545 8 : case EXPR_KIND_PARTITION_BOUND:
546 8 : if (isAgg)
547 8 : err = _("aggregate functions are not allowed in partition bound");
548 : else
549 0 : err = _("grouping operations are not allowed in partition bound");
550 :
551 8 : break;
552 4 : case EXPR_KIND_PARTITION_EXPRESSION:
553 4 : if (isAgg)
554 4 : err = _("aggregate functions are not allowed in partition key expressions");
555 : else
556 0 : err = _("grouping operations are not allowed in partition key expressions");
557 :
558 4 : break;
559 8 : case EXPR_KIND_GENERATED_COLUMN:
560 :
561 8 : if (isAgg)
562 8 : err = _("aggregate functions are not allowed in column generation expressions");
563 : else
564 0 : err = _("grouping operations are not allowed in column generation expressions");
565 :
566 8 : break;
567 :
568 0 : case EXPR_KIND_CALL_ARGUMENT:
569 0 : if (isAgg)
570 0 : err = _("aggregate functions are not allowed in CALL arguments");
571 : else
572 0 : err = _("grouping operations are not allowed in CALL arguments");
573 :
574 0 : break;
575 :
576 4 : case EXPR_KIND_COPY_WHERE:
577 4 : if (isAgg)
578 4 : err = _("aggregate functions are not allowed in COPY FROM WHERE conditions");
579 : else
580 0 : err = _("grouping operations are not allowed in COPY FROM WHERE conditions");
581 :
582 4 : break;
583 :
584 0 : case EXPR_KIND_CYCLE_MARK:
585 0 : errkind = true;
586 0 : break;
587 0 : case EXPR_KIND_FOR_PORTION:
588 0 : if (isAgg)
589 0 : err = _("aggregate functions are not allowed in FOR PORTION OF expressions");
590 : else
591 0 : err = _("grouping operations are not allowed in FOR PORTION OF expressions");
592 :
593 0 : break;
594 :
595 0 : case EXPR_KIND_PROPGRAPH_PROPERTY:
596 0 : if (isAgg)
597 0 : err = _("aggregate functions are not allowed in property definition expressions");
598 : else
599 0 : err = _("grouping operations are not allowed in property definition expressions");
600 :
601 0 : break;
602 :
603 : /*
604 : * There is intentionally no default: case here, so that the
605 : * compiler will warn if we add a new ParseExprKind without
606 : * extending this switch. If we do see an unrecognized value at
607 : * runtime, the behavior will be the same as for EXPR_KIND_OTHER,
608 : * which is sane anyway.
609 : */
610 : }
611 :
612 30582 : if (err)
613 64 : ereport(ERROR,
614 : (errcode(ERRCODE_GROUPING_ERROR),
615 : errmsg_internal("%s", err),
616 : parser_errposition(pstate, location)));
617 :
618 30518 : if (errkind)
619 : {
620 16 : if (isAgg)
621 : /* translator: %s is name of a SQL construct, eg GROUP BY */
622 16 : err = _("aggregate functions are not allowed in %s");
623 : else
624 : /* translator: %s is name of a SQL construct, eg GROUP BY */
625 0 : err = _("grouping operations are not allowed in %s");
626 :
627 16 : ereport(ERROR,
628 : (errcode(ERRCODE_GROUPING_ERROR),
629 : errmsg_internal(err,
630 : ParseExprKindName(pstate->p_expr_kind)),
631 : parser_errposition(pstate, location)));
632 : }
633 30502 : }
634 :
635 : /*
636 : * check_agg_arguments
637 : * Scan the arguments of an aggregate function to determine the
638 : * aggregate's semantic level (zero is the current select's level,
639 : * one is its parent, etc).
640 : *
641 : * The aggregate's level is the same as the level of the lowest-level variable
642 : * or aggregate in its aggregated arguments (including any ORDER BY columns)
643 : * or filter expression; or if it contains no variables at all, we presume it
644 : * to be local.
645 : *
646 : * Vars/Aggs in direct arguments are *not* counted towards determining the
647 : * agg's level, as those arguments aren't evaluated per-row but only
648 : * per-group, and so in some sense aren't really agg arguments. However,
649 : * this can mean that we decide an agg is upper-level even when its direct
650 : * args contain lower-level Vars/Aggs, and that case has to be disallowed.
651 : * (This is a little strange, but the SQL standard seems pretty definite that
652 : * direct args are not to be considered when setting the agg's level.)
653 : *
654 : * We also take this opportunity to detect any aggregates or window functions
655 : * nested within the arguments. We can throw error immediately if we find
656 : * a window function. Aggregates are a bit trickier because it's only an
657 : * error if the inner aggregate is of the same semantic level as the outer,
658 : * which we can't know until we finish scanning the arguments.
659 : */
660 : static int
661 30618 : check_agg_arguments(ParseState *pstate,
662 : List *directargs,
663 : List *args,
664 : Expr *filter,
665 : int agglocation)
666 : {
667 : int agglevel;
668 : check_agg_arguments_context context;
669 :
670 30618 : context.pstate = pstate;
671 30618 : context.min_varlevel = -1; /* signifies nothing found yet */
672 30618 : context.min_agglevel = -1;
673 30618 : context.min_ctelevel = -1;
674 30618 : context.min_cte = NULL;
675 30618 : context.sublevels_up = 0;
676 :
677 30618 : (void) check_agg_arguments_walker((Node *) args, &context);
678 30614 : (void) check_agg_arguments_walker((Node *) filter, &context);
679 :
680 : /*
681 : * If we found no vars nor aggs at all, it's a level-zero aggregate;
682 : * otherwise, its level is the minimum of vars or aggs.
683 : */
684 30614 : if (context.min_varlevel < 0)
685 : {
686 11646 : if (context.min_agglevel < 0)
687 11646 : agglevel = 0;
688 : else
689 0 : agglevel = context.min_agglevel;
690 : }
691 18968 : else if (context.min_agglevel < 0)
692 18944 : agglevel = context.min_varlevel;
693 : else
694 24 : agglevel = Min(context.min_varlevel, context.min_agglevel);
695 :
696 : /*
697 : * If there's a nested aggregate of the same semantic level, complain.
698 : */
699 30614 : if (agglevel == context.min_agglevel)
700 : {
701 : int aggloc;
702 :
703 20 : aggloc = locate_agg_of_level((Node *) args, agglevel);
704 20 : if (aggloc < 0)
705 8 : aggloc = locate_agg_of_level((Node *) filter, agglevel);
706 20 : ereport(ERROR,
707 : (errcode(ERRCODE_GROUPING_ERROR),
708 : errmsg("aggregate function calls cannot be nested"),
709 : parser_errposition(pstate, aggloc)));
710 : }
711 :
712 : /*
713 : * If there's a non-local CTE that's below the aggregate's semantic level,
714 : * complain. It's not quite clear what we should do to fix up such a case
715 : * (treating the CTE reference like a Var seems wrong), and it's also
716 : * unclear whether there is a real-world use for such cases.
717 : */
718 30594 : if (context.min_ctelevel >= 0 && context.min_ctelevel < agglevel)
719 4 : ereport(ERROR,
720 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
721 : errmsg("outer-level aggregate cannot use a nested CTE"),
722 : errdetail("CTE \"%s\" is below the aggregate's semantic level.",
723 : context.min_cte->eref->aliasname),
724 : parser_errposition(pstate, agglocation)));
725 :
726 : /*
727 : * Now check for vars/aggs in the direct arguments, and throw error if
728 : * needed. Note that we allow a Var of the agg's semantic level, but not
729 : * an Agg of that level. In principle such Aggs could probably be
730 : * supported, but it would create an ordering dependency among the
731 : * aggregates at execution time. Since the case appears neither to be
732 : * required by spec nor particularly useful, we just treat it as a
733 : * nested-aggregate situation.
734 : */
735 30590 : if (directargs)
736 : {
737 204 : context.min_varlevel = -1;
738 204 : context.min_agglevel = -1;
739 204 : context.min_ctelevel = -1;
740 204 : (void) check_agg_arguments_walker((Node *) directargs, &context);
741 204 : if (context.min_varlevel >= 0 && context.min_varlevel < agglevel)
742 4 : ereport(ERROR,
743 : (errcode(ERRCODE_GROUPING_ERROR),
744 : errmsg("outer-level aggregate cannot contain a lower-level variable in its direct arguments"),
745 : parser_errposition(pstate,
746 : locate_var_of_level((Node *) directargs,
747 : context.min_varlevel))));
748 200 : if (context.min_agglevel >= 0 && context.min_agglevel <= agglevel)
749 4 : ereport(ERROR,
750 : (errcode(ERRCODE_GROUPING_ERROR),
751 : errmsg("aggregate function calls cannot be nested"),
752 : parser_errposition(pstate,
753 : locate_agg_of_level((Node *) directargs,
754 : context.min_agglevel))));
755 196 : if (context.min_ctelevel >= 0 && context.min_ctelevel < agglevel)
756 0 : ereport(ERROR,
757 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
758 : errmsg("outer-level aggregate cannot use a nested CTE"),
759 : errdetail("CTE \"%s\" is below the aggregate's semantic level.",
760 : context.min_cte->eref->aliasname),
761 : parser_errposition(pstate, agglocation)));
762 : }
763 30582 : return agglevel;
764 : }
765 :
766 : static bool
767 122296 : check_agg_arguments_walker(Node *node,
768 : check_agg_arguments_context *context)
769 : {
770 122296 : if (node == NULL)
771 42768 : return false;
772 79528 : if (IsA(node, Var))
773 : {
774 21532 : int varlevelsup = ((Var *) node)->varlevelsup;
775 :
776 : /* convert levelsup to frame of reference of original query */
777 21532 : varlevelsup -= context->sublevels_up;
778 : /* ignore local vars of subqueries */
779 21532 : if (varlevelsup >= 0)
780 : {
781 21438 : if (context->min_varlevel < 0 ||
782 2433 : context->min_varlevel > varlevelsup)
783 19073 : context->min_varlevel = varlevelsup;
784 : }
785 21532 : return false;
786 : }
787 57996 : if (IsA(node, Aggref))
788 : {
789 44 : int agglevelsup = ((Aggref *) node)->agglevelsup;
790 :
791 : /* convert levelsup to frame of reference of original query */
792 44 : agglevelsup -= context->sublevels_up;
793 : /* ignore local aggs of subqueries */
794 44 : if (agglevelsup >= 0)
795 : {
796 28 : if (context->min_agglevel < 0 ||
797 0 : context->min_agglevel > agglevelsup)
798 28 : context->min_agglevel = agglevelsup;
799 : }
800 : /* Continue and descend into subtree */
801 : }
802 57996 : if (IsA(node, GroupingFunc))
803 : {
804 0 : int agglevelsup = ((GroupingFunc *) node)->agglevelsup;
805 :
806 : /* convert levelsup to frame of reference of original query */
807 0 : agglevelsup -= context->sublevels_up;
808 : /* ignore local aggs of subqueries */
809 0 : if (agglevelsup >= 0)
810 : {
811 0 : if (context->min_agglevel < 0 ||
812 0 : context->min_agglevel > agglevelsup)
813 0 : context->min_agglevel = agglevelsup;
814 : }
815 : /* Continue and descend into subtree */
816 : }
817 :
818 : /*
819 : * SRFs and window functions can be rejected immediately, unless we are
820 : * within a sub-select within the aggregate's arguments; in that case
821 : * they're OK.
822 : */
823 57996 : if (context->sublevels_up == 0)
824 : {
825 57369 : if ((IsA(node, FuncExpr) && ((FuncExpr *) node)->funcretset) ||
826 57365 : (IsA(node, OpExpr) && ((OpExpr *) node)->opretset))
827 4 : ereport(ERROR,
828 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
829 : errmsg("aggregate function calls cannot contain set-returning function calls"),
830 : errhint("You might be able to move the set-returning function into a LATERAL FROM item."),
831 : parser_errposition(context->pstate, exprLocation(node))));
832 57365 : if (IsA(node, WindowFunc))
833 0 : ereport(ERROR,
834 : (errcode(ERRCODE_GROUPING_ERROR),
835 : errmsg("aggregate function calls cannot contain window function calls"),
836 : parser_errposition(context->pstate,
837 : ((WindowFunc *) node)->location)));
838 : }
839 :
840 57992 : if (IsA(node, RangeTblEntry))
841 : {
842 85 : RangeTblEntry *rte = (RangeTblEntry *) node;
843 :
844 85 : if (rte->rtekind == RTE_CTE)
845 : {
846 12 : int ctelevelsup = rte->ctelevelsup;
847 :
848 : /* convert levelsup to frame of reference of original query */
849 12 : ctelevelsup -= context->sublevels_up;
850 : /* ignore local CTEs of subqueries */
851 12 : if (ctelevelsup >= 0)
852 : {
853 12 : if (context->min_ctelevel < 0 ||
854 0 : context->min_ctelevel > ctelevelsup)
855 : {
856 12 : context->min_ctelevel = ctelevelsup;
857 12 : context->min_cte = rte;
858 : }
859 : }
860 : }
861 85 : return false; /* allow range_table_walker to continue */
862 : }
863 57907 : if (IsA(node, Query))
864 : {
865 : /* Recurse into subselects */
866 : bool result;
867 :
868 91 : context->sublevels_up++;
869 91 : result = query_tree_walker((Query *) node,
870 : check_agg_arguments_walker,
871 : context,
872 : QTW_EXAMINE_RTES_BEFORE);
873 91 : context->sublevels_up--;
874 91 : return result;
875 : }
876 :
877 57816 : return expression_tree_walker(node,
878 : check_agg_arguments_walker,
879 : context);
880 : }
881 :
882 : /*
883 : * transformWindowFuncCall -
884 : * Finish initial transformation of a window function call
885 : *
886 : * parse_func.c has recognized the function as a window function, and has set
887 : * up all the fields of the WindowFunc except winref. Here we must (1) add
888 : * the WindowDef to the pstate (if not a duplicate of one already present) and
889 : * set winref to link to it; and (2) mark p_hasWindowFuncs true in the pstate.
890 : * Unlike aggregates, only the most closely nested pstate level need be
891 : * considered --- there are no "outer window functions" per SQL spec.
892 : */
893 : void
894 2628 : transformWindowFuncCall(ParseState *pstate, WindowFunc *wfunc,
895 : WindowDef *windef)
896 : {
897 : const char *err;
898 : bool errkind;
899 :
900 : /*
901 : * A window function call can't contain another one (but aggs are OK). XXX
902 : * is this required by spec, or just an unimplemented feature?
903 : *
904 : * Note: we don't need to check the filter expression here, because the
905 : * context checks done below and in transformAggregateCall would have
906 : * already rejected any window funcs or aggs within the filter.
907 : */
908 3336 : if (pstate->p_hasWindowFuncs &&
909 708 : contain_windowfuncs((Node *) wfunc->args))
910 0 : ereport(ERROR,
911 : (errcode(ERRCODE_WINDOWING_ERROR),
912 : errmsg("window function calls cannot be nested"),
913 : parser_errposition(pstate,
914 : locate_windowfunc((Node *) wfunc->args))));
915 :
916 : /*
917 : * Check to see if the window function is in an invalid place within the
918 : * query.
919 : *
920 : * For brevity we support two schemes for reporting an error here: set
921 : * "err" to a custom message, or set "errkind" true if the error context
922 : * is sufficiently identified by what ParseExprKindName will return, *and*
923 : * what it will return is just a SQL keyword. (Otherwise, use a custom
924 : * message to avoid creating translation problems.)
925 : */
926 2628 : err = NULL;
927 2628 : errkind = false;
928 2628 : switch (pstate->p_expr_kind)
929 : {
930 0 : case EXPR_KIND_NONE:
931 : Assert(false); /* can't happen */
932 0 : break;
933 0 : case EXPR_KIND_OTHER:
934 : /* Accept window func here; caller must throw error if wanted */
935 0 : break;
936 4 : case EXPR_KIND_JOIN_ON:
937 : case EXPR_KIND_JOIN_USING:
938 4 : err = _("window functions are not allowed in JOIN conditions");
939 4 : break;
940 0 : case EXPR_KIND_FROM_SUBSELECT:
941 : /* can't get here, but just in case, throw an error */
942 0 : errkind = true;
943 0 : break;
944 0 : case EXPR_KIND_FROM_FUNCTION:
945 0 : err = _("window functions are not allowed in functions in FROM");
946 0 : break;
947 8 : case EXPR_KIND_WHERE:
948 8 : errkind = true;
949 8 : break;
950 0 : case EXPR_KIND_POLICY:
951 0 : err = _("window functions are not allowed in policy expressions");
952 0 : break;
953 0 : case EXPR_KIND_HAVING:
954 0 : errkind = true;
955 0 : break;
956 0 : case EXPR_KIND_FILTER:
957 0 : errkind = true;
958 0 : break;
959 4 : case EXPR_KIND_WINDOW_PARTITION:
960 : case EXPR_KIND_WINDOW_ORDER:
961 : case EXPR_KIND_WINDOW_FRAME_RANGE:
962 : case EXPR_KIND_WINDOW_FRAME_ROWS:
963 : case EXPR_KIND_WINDOW_FRAME_GROUPS:
964 4 : err = _("window functions are not allowed in window definitions");
965 4 : break;
966 2587 : case EXPR_KIND_SELECT_TARGET:
967 : /* okay */
968 2587 : break;
969 0 : case EXPR_KIND_INSERT_TARGET:
970 : case EXPR_KIND_UPDATE_SOURCE:
971 : case EXPR_KIND_UPDATE_TARGET:
972 0 : errkind = true;
973 0 : break;
974 0 : case EXPR_KIND_MERGE_WHEN:
975 0 : err = _("window functions are not allowed in MERGE WHEN conditions");
976 0 : break;
977 0 : case EXPR_KIND_GROUP_BY:
978 0 : errkind = true;
979 0 : break;
980 5 : case EXPR_KIND_ORDER_BY:
981 : /* okay */
982 5 : break;
983 0 : case EXPR_KIND_DISTINCT_ON:
984 : /* okay */
985 0 : break;
986 0 : case EXPR_KIND_LIMIT:
987 : case EXPR_KIND_OFFSET:
988 0 : errkind = true;
989 0 : break;
990 4 : case EXPR_KIND_RETURNING:
991 : case EXPR_KIND_MERGE_RETURNING:
992 4 : errkind = true;
993 4 : break;
994 0 : case EXPR_KIND_VALUES:
995 : case EXPR_KIND_VALUES_SINGLE:
996 0 : errkind = true;
997 0 : break;
998 0 : case EXPR_KIND_CHECK_CONSTRAINT:
999 : case EXPR_KIND_DOMAIN_CHECK:
1000 0 : err = _("window functions are not allowed in check constraints");
1001 0 : break;
1002 0 : case EXPR_KIND_COLUMN_DEFAULT:
1003 : case EXPR_KIND_FUNCTION_DEFAULT:
1004 0 : err = _("window functions are not allowed in DEFAULT expressions");
1005 0 : break;
1006 0 : case EXPR_KIND_INDEX_EXPRESSION:
1007 0 : err = _("window functions are not allowed in index expressions");
1008 0 : break;
1009 0 : case EXPR_KIND_STATS_EXPRESSION:
1010 0 : err = _("window functions are not allowed in statistics expressions");
1011 0 : break;
1012 0 : case EXPR_KIND_INDEX_PREDICATE:
1013 0 : err = _("window functions are not allowed in index predicates");
1014 0 : break;
1015 0 : case EXPR_KIND_ALTER_COL_TRANSFORM:
1016 0 : err = _("window functions are not allowed in transform expressions");
1017 0 : break;
1018 0 : case EXPR_KIND_EXECUTE_PARAMETER:
1019 0 : err = _("window functions are not allowed in EXECUTE parameters");
1020 0 : break;
1021 0 : case EXPR_KIND_TRIGGER_WHEN:
1022 0 : err = _("window functions are not allowed in trigger WHEN conditions");
1023 0 : break;
1024 0 : case EXPR_KIND_PARTITION_BOUND:
1025 0 : err = _("window functions are not allowed in partition bound");
1026 0 : break;
1027 4 : case EXPR_KIND_PARTITION_EXPRESSION:
1028 4 : err = _("window functions are not allowed in partition key expressions");
1029 4 : break;
1030 0 : case EXPR_KIND_CALL_ARGUMENT:
1031 0 : err = _("window functions are not allowed in CALL arguments");
1032 0 : break;
1033 4 : case EXPR_KIND_COPY_WHERE:
1034 4 : err = _("window functions are not allowed in COPY FROM WHERE conditions");
1035 4 : break;
1036 8 : case EXPR_KIND_GENERATED_COLUMN:
1037 8 : err = _("window functions are not allowed in column generation expressions");
1038 8 : break;
1039 0 : case EXPR_KIND_CYCLE_MARK:
1040 0 : errkind = true;
1041 0 : break;
1042 0 : case EXPR_KIND_PROPGRAPH_PROPERTY:
1043 0 : err = _("window functions are not allowed in property definition expressions");
1044 0 : break;
1045 0 : case EXPR_KIND_FOR_PORTION:
1046 0 : err = _("window functions are not allowed in FOR PORTION OF expressions");
1047 0 : break;
1048 :
1049 : /*
1050 : * There is intentionally no default: case here, so that the
1051 : * compiler will warn if we add a new ParseExprKind without
1052 : * extending this switch. If we do see an unrecognized value at
1053 : * runtime, the behavior will be the same as for EXPR_KIND_OTHER,
1054 : * which is sane anyway.
1055 : */
1056 : }
1057 2628 : if (err)
1058 24 : ereport(ERROR,
1059 : (errcode(ERRCODE_WINDOWING_ERROR),
1060 : errmsg_internal("%s", err),
1061 : parser_errposition(pstate, wfunc->location)));
1062 2604 : if (errkind)
1063 12 : ereport(ERROR,
1064 : (errcode(ERRCODE_WINDOWING_ERROR),
1065 : /* translator: %s is name of a SQL construct, eg GROUP BY */
1066 : errmsg("window functions are not allowed in %s",
1067 : ParseExprKindName(pstate->p_expr_kind)),
1068 : parser_errposition(pstate, wfunc->location)));
1069 :
1070 : /*
1071 : * If the OVER clause just specifies a window name, find that WINDOW
1072 : * clause (which had better be present). Otherwise, try to match all the
1073 : * properties of the OVER clause, and make a new entry in the p_windowdefs
1074 : * list if no luck.
1075 : */
1076 2592 : if (windef->name)
1077 : {
1078 790 : Index winref = 0;
1079 : ListCell *lc;
1080 :
1081 : Assert(windef->refname == NULL &&
1082 : windef->partitionClause == NIL &&
1083 : windef->orderClause == NIL &&
1084 : windef->frameOptions == FRAMEOPTION_DEFAULTS);
1085 :
1086 818 : foreach(lc, pstate->p_windowdefs)
1087 : {
1088 818 : WindowDef *refwin = (WindowDef *) lfirst(lc);
1089 :
1090 818 : winref++;
1091 818 : if (refwin->name && strcmp(refwin->name, windef->name) == 0)
1092 : {
1093 790 : wfunc->winref = winref;
1094 790 : break;
1095 : }
1096 : }
1097 790 : if (lc == NULL) /* didn't find it? */
1098 0 : ereport(ERROR,
1099 : (errcode(ERRCODE_UNDEFINED_OBJECT),
1100 : errmsg("window \"%s\" does not exist", windef->name),
1101 : parser_errposition(pstate, windef->location)));
1102 : }
1103 : else
1104 : {
1105 1802 : Index winref = 0;
1106 : ListCell *lc;
1107 :
1108 2042 : foreach(lc, pstate->p_windowdefs)
1109 : {
1110 408 : WindowDef *refwin = (WindowDef *) lfirst(lc);
1111 :
1112 408 : winref++;
1113 416 : if (refwin->refname && windef->refname &&
1114 8 : strcmp(refwin->refname, windef->refname) == 0)
1115 : /* matched on refname */ ;
1116 400 : else if (!refwin->refname && !windef->refname)
1117 : /* matched, no refname */ ;
1118 : else
1119 32 : continue;
1120 :
1121 : /*
1122 : * Also see similar de-duplication code in optimize_window_clauses
1123 : */
1124 692 : if (equal(refwin->partitionClause, windef->partitionClause) &&
1125 316 : equal(refwin->orderClause, windef->orderClause) &&
1126 444 : refwin->frameOptions == windef->frameOptions &&
1127 336 : equal(refwin->startOffset, windef->startOffset) &&
1128 168 : equal(refwin->endOffset, windef->endOffset))
1129 : {
1130 : /* found a duplicate window specification */
1131 168 : wfunc->winref = winref;
1132 168 : break;
1133 : }
1134 : }
1135 1802 : if (lc == NULL) /* didn't find it? */
1136 : {
1137 1634 : pstate->p_windowdefs = lappend(pstate->p_windowdefs, windef);
1138 1634 : wfunc->winref = list_length(pstate->p_windowdefs);
1139 : }
1140 : }
1141 :
1142 2592 : pstate->p_hasWindowFuncs = true;
1143 2592 : }
1144 :
1145 : /*
1146 : * parseCheckAggregates
1147 : * Check for aggregates where they shouldn't be and improper grouping, and
1148 : * replace grouped variables in the targetlist and HAVING clause with Vars
1149 : * that reference the RTE_GROUP RTE.
1150 : * This function should be called after the target list and qualifications
1151 : * are finalized.
1152 : *
1153 : * Misplaced aggregates are now mostly detected in transformAggregateCall,
1154 : * but it seems more robust to check for aggregates in recursive queries
1155 : * only after everything is finalized. In any case it's hard to detect
1156 : * improper grouping on-the-fly, so we have to make another pass over the
1157 : * query for that.
1158 : */
1159 : void
1160 27270 : parseCheckAggregates(ParseState *pstate, Query *qry)
1161 : {
1162 27270 : List *gset_common = NIL;
1163 27270 : List *groupClauses = NIL;
1164 27270 : List *groupClauseCommonVars = NIL;
1165 : bool have_non_var_grouping;
1166 27270 : List *func_grouped_rels = NIL;
1167 : ListCell *l;
1168 : bool hasJoinRTEs;
1169 : bool hasSelfRefRTEs;
1170 : Node *clause;
1171 :
1172 : /* This should only be called if we found aggregates or grouping */
1173 : Assert(pstate->p_hasAggs || qry->groupClause || qry->havingQual || qry->groupingSets);
1174 :
1175 : /*
1176 : * If we have grouping sets, expand them and find the intersection of all
1177 : * sets.
1178 : */
1179 27270 : if (qry->groupingSets)
1180 : {
1181 : /*
1182 : * The limit of 4096 is arbitrary and exists simply to avoid resource
1183 : * issues from pathological constructs.
1184 : */
1185 724 : List *gsets = expand_grouping_sets(qry->groupingSets, qry->groupDistinct, 4096);
1186 :
1187 724 : if (!gsets)
1188 0 : ereport(ERROR,
1189 : (errcode(ERRCODE_STATEMENT_TOO_COMPLEX),
1190 : errmsg("too many grouping sets present (maximum 4096)"),
1191 : parser_errposition(pstate,
1192 : qry->groupClause
1193 : ? exprLocation((Node *) qry->groupClause)
1194 : : exprLocation((Node *) qry->groupingSets))));
1195 :
1196 : /*
1197 : * The intersection will often be empty, so help things along by
1198 : * seeding the intersect with the smallest set.
1199 : */
1200 724 : gset_common = linitial(gsets);
1201 :
1202 724 : if (gset_common)
1203 : {
1204 454 : for_each_from(l, gsets, 1)
1205 : {
1206 314 : gset_common = list_intersection_int(gset_common, lfirst(l));
1207 314 : if (!gset_common)
1208 158 : break;
1209 : }
1210 : }
1211 :
1212 : /*
1213 : * If there was only one grouping set in the expansion, AND if the
1214 : * groupClause is non-empty (meaning that the grouping set is not
1215 : * empty either), then we can ditch the grouping set and pretend we
1216 : * just had a normal GROUP BY.
1217 : */
1218 724 : if (list_length(gsets) == 1 && qry->groupClause)
1219 16 : qry->groupingSets = NIL;
1220 : }
1221 :
1222 : /*
1223 : * Scan the range table to see if there are JOIN or self-reference CTE
1224 : * entries. We'll need this info below.
1225 : */
1226 27270 : hasJoinRTEs = hasSelfRefRTEs = false;
1227 59936 : foreach(l, pstate->p_rtable)
1228 : {
1229 32666 : RangeTblEntry *rte = (RangeTblEntry *) lfirst(l);
1230 :
1231 32666 : if (rte->rtekind == RTE_JOIN)
1232 1447 : hasJoinRTEs = true;
1233 31219 : else if (rte->rtekind == RTE_CTE && rte->self_reference)
1234 8 : hasSelfRefRTEs = true;
1235 : }
1236 :
1237 : /*
1238 : * Build a list of the acceptable GROUP BY expressions to save in the
1239 : * RTE_GROUP RTE, and for use by substitute_grouped_columns().
1240 : *
1241 : * We get the TLE, not just the expr, because GROUPING wants to know the
1242 : * sortgroupref.
1243 : */
1244 32697 : foreach(l, qry->groupClause)
1245 : {
1246 5427 : SortGroupClause *grpcl = (SortGroupClause *) lfirst(l);
1247 : TargetEntry *expr;
1248 :
1249 5427 : expr = get_sortgroupclause_tle(grpcl, qry->targetList);
1250 5427 : if (expr == NULL)
1251 0 : continue; /* probably cannot happen */
1252 :
1253 5427 : groupClauses = lappend(groupClauses, expr);
1254 : }
1255 :
1256 : /*
1257 : * If there are any acceptable GROUP BY expressions, build an RTE and
1258 : * nsitem for the result of the grouping step. (It's important to do this
1259 : * before flattening join alias vars in groupClauses, because the RTE
1260 : * should preserve any alias vars that were in the input.)
1261 : */
1262 27270 : if (groupClauses)
1263 : {
1264 3229 : pstate->p_grouping_nsitem =
1265 3229 : addRangeTableEntryForGroup(pstate, groupClauses);
1266 :
1267 : /* Set qry->rtable again in case it was previously NIL */
1268 3229 : qry->rtable = pstate->p_rtable;
1269 : /* Mark the Query as having RTE_GROUP RTE */
1270 3229 : qry->hasGroupRTE = true;
1271 : }
1272 :
1273 : /*
1274 : * If there are join alias vars involved, we have to flatten them to the
1275 : * underlying vars, so that aliased and unaliased vars will be correctly
1276 : * taken as equal. We can skip the expense of doing this if no rangetable
1277 : * entries are RTE_JOIN kind.
1278 : */
1279 27270 : if (hasJoinRTEs)
1280 : groupClauses = (List *)
1281 1217 : flatten_join_alias_for_parser(qry, (Node *) groupClauses, 0);
1282 :
1283 : /*
1284 : * Detect whether any of the grouping expressions aren't simple Vars; if
1285 : * they're all Vars then we don't have to work so hard in the recursive
1286 : * scans. (Note we have to flatten aliases before this.)
1287 : *
1288 : * Track Vars that are included in all grouping sets separately in
1289 : * groupClauseCommonVars, since these are the only ones we can use to
1290 : * check for functional dependencies.
1291 : */
1292 27270 : have_non_var_grouping = false;
1293 32697 : foreach(l, groupClauses)
1294 : {
1295 5427 : TargetEntry *tle = lfirst(l);
1296 :
1297 5427 : if (!IsA(tle->expr, Var))
1298 : {
1299 990 : have_non_var_grouping = true;
1300 : }
1301 5641 : else if (!qry->groupingSets ||
1302 1204 : list_member_int(gset_common, tle->ressortgroupref))
1303 : {
1304 3357 : groupClauseCommonVars = lappend(groupClauseCommonVars, tle->expr);
1305 : }
1306 : }
1307 :
1308 : /*
1309 : * Replace grouped variables in the targetlist and HAVING clause with Vars
1310 : * that reference the RTE_GROUP RTE. Emit an error message if we find any
1311 : * ungrouped variables.
1312 : *
1313 : * Note: because we check resjunk tlist elements as well as regular ones,
1314 : * this will also find ungrouped variables that came from ORDER BY and
1315 : * WINDOW clauses. For that matter, it's also going to examine the
1316 : * grouping expressions themselves --- but they'll all pass the test ...
1317 : *
1318 : * We also finalize GROUPING expressions, but for that we need to traverse
1319 : * the original (unflattened) clause in order to modify nodes.
1320 : */
1321 27270 : clause = (Node *) qry->targetList;
1322 27270 : finalize_grouping_exprs(clause, pstate, qry,
1323 : groupClauses, hasJoinRTEs,
1324 : have_non_var_grouping);
1325 27266 : if (hasJoinRTEs)
1326 1217 : clause = flatten_join_alias_for_parser(qry, clause, 0);
1327 27206 : qry->targetList = (List *)
1328 27266 : substitute_grouped_columns(clause, pstate, qry,
1329 : groupClauses, groupClauseCommonVars,
1330 : gset_common,
1331 : have_non_var_grouping,
1332 : &func_grouped_rels);
1333 :
1334 27206 : clause = (Node *) qry->havingQual;
1335 27206 : finalize_grouping_exprs(clause, pstate, qry,
1336 : groupClauses, hasJoinRTEs,
1337 : have_non_var_grouping);
1338 27206 : if (hasJoinRTEs)
1339 1201 : clause = flatten_join_alias_for_parser(qry, clause, 0);
1340 27202 : qry->havingQual =
1341 27206 : substitute_grouped_columns(clause, pstate, qry,
1342 : groupClauses, groupClauseCommonVars,
1343 : gset_common,
1344 : have_non_var_grouping,
1345 : &func_grouped_rels);
1346 :
1347 : /*
1348 : * Per spec, aggregates can't appear in a recursive term.
1349 : */
1350 27202 : if (pstate->p_hasAggs && hasSelfRefRTEs)
1351 8 : ereport(ERROR,
1352 : (errcode(ERRCODE_INVALID_RECURSION),
1353 : errmsg("aggregate functions are not allowed in a recursive query's recursive term"),
1354 : parser_errposition(pstate,
1355 : locate_agg_of_level((Node *) qry, 0))));
1356 27194 : }
1357 :
1358 : /*
1359 : * substitute_grouped_columns -
1360 : * Scan the given expression tree for grouped variables (variables that
1361 : * are listed in the groupClauses list) and replace them with Vars that
1362 : * reference the RTE_GROUP RTE. Emit a suitable error message if any
1363 : * ungrouped variables (variables that are not listed in the groupClauses
1364 : * list and are not within the arguments of aggregate functions) are
1365 : * found.
1366 : *
1367 : * NOTE: we assume that the given clause has been transformed suitably for
1368 : * parser output. This means we can use expression_tree_mutator.
1369 : */
1370 : static Node *
1371 54472 : substitute_grouped_columns(Node *node, ParseState *pstate, Query *qry,
1372 : List *groupClauses, List *groupClauseCommonVars,
1373 : List *gset_common,
1374 : bool have_non_var_grouping,
1375 : List **func_grouped_rels)
1376 : {
1377 : substitute_grouped_columns_context context;
1378 :
1379 54472 : context.pstate = pstate;
1380 54472 : context.qry = qry;
1381 54472 : context.hasJoinRTEs = false; /* assume caller flattened join Vars */
1382 54472 : context.groupClauses = groupClauses;
1383 54472 : context.groupClauseCommonVars = groupClauseCommonVars;
1384 54472 : context.groupClauseSubLevels = NIL;
1385 54472 : context.gset_common = gset_common;
1386 54472 : context.have_non_var_grouping = have_non_var_grouping;
1387 54472 : context.func_grouped_rels = func_grouped_rels;
1388 54472 : context.sublevels_up = 0;
1389 54472 : context.in_agg_direct_args = false;
1390 54472 : return substitute_grouped_columns_mutator(node, &context);
1391 : }
1392 :
1393 : static Node *
1394 184202 : substitute_grouped_columns_mutator(Node *node,
1395 : substitute_grouped_columns_context *context)
1396 : {
1397 : ListCell *gl;
1398 :
1399 184202 : if (node == NULL)
1400 61372 : return NULL;
1401 :
1402 122830 : if (IsA(node, Aggref))
1403 : {
1404 30658 : Aggref *agg = (Aggref *) node;
1405 :
1406 30658 : if ((int) agg->agglevelsup == context->sublevels_up)
1407 : {
1408 : /*
1409 : * If we find an aggregate call of the original level, do not
1410 : * recurse into its normal arguments, ORDER BY arguments, or
1411 : * filter; grouped vars there do not need to be replaced and
1412 : * ungrouped vars there are not an error. But we should check
1413 : * direct arguments as though they weren't in an aggregate. We
1414 : * set a special flag in the context to help produce a useful
1415 : * error message for ungrouped vars in direct arguments.
1416 : */
1417 30654 : agg = copyObject(agg);
1418 :
1419 : Assert(!context->in_agg_direct_args);
1420 30654 : context->in_agg_direct_args = true;
1421 30650 : agg->aggdirectargs = (List *)
1422 30654 : substitute_grouped_columns_mutator((Node *) agg->aggdirectargs,
1423 : context);
1424 30650 : context->in_agg_direct_args = false;
1425 30650 : return (Node *) agg;
1426 : }
1427 :
1428 : /*
1429 : * We can skip recursing into aggregates of higher levels altogether,
1430 : * since they could not possibly contain Vars of concern to us (see
1431 : * transformAggregateCall). We do need to look at aggregates of lower
1432 : * levels, however.
1433 : */
1434 4 : if ((int) agg->agglevelsup > context->sublevels_up)
1435 0 : return node;
1436 : }
1437 :
1438 92176 : if (IsA(node, GroupingFunc))
1439 : {
1440 274 : GroupingFunc *grp = (GroupingFunc *) node;
1441 :
1442 : /* handled GroupingFunc separately, no need to recheck at this level */
1443 :
1444 274 : if ((int) grp->agglevelsup >= context->sublevels_up)
1445 253 : return node;
1446 : }
1447 :
1448 : /*
1449 : * If we have any GROUP BY items that are not simple Vars, check to see if
1450 : * subexpression as a whole matches any GROUP BY item. We need to do this
1451 : * at every recursion level so that we recognize GROUPed-BY expressions
1452 : * before reaching variables within them. (Since this approach is pretty
1453 : * expensive, we don't do it this way if the items are all simple Vars.)
1454 : */
1455 91923 : if (context->have_non_var_grouping)
1456 : {
1457 : List *groupClauses;
1458 5901 : int attnum = 0;
1459 :
1460 : /* Within a subquery, we need a mutated version of the groupClauses */
1461 5901 : if (context->sublevels_up == 0)
1462 5409 : groupClauses = context->groupClauses;
1463 : else
1464 492 : groupClauses = list_nth(context->groupClauseSubLevels,
1465 492 : context->sublevels_up - 1);
1466 :
1467 15974 : foreach(gl, groupClauses)
1468 : {
1469 11611 : TargetEntry *tle = (TargetEntry *) lfirst(gl);
1470 :
1471 11611 : attnum++;
1472 11611 : if (equal(node, tle->expr))
1473 : {
1474 : /* acceptable, replace it with a GROUP Var */
1475 1538 : return (Node *) buildGroupedVar(attnum,
1476 : tle->ressortgroupref,
1477 : context);
1478 : }
1479 : }
1480 : }
1481 :
1482 : /*
1483 : * Constants are always acceptable. We have to do this after we checked
1484 : * the subexpression as a whole for a match, because it is possible that
1485 : * we have GROUP BY items that are constants, and the constants would
1486 : * become not so constant after the grouping step.
1487 : */
1488 90385 : if (IsA(node, Const) ||
1489 87733 : IsA(node, Param))
1490 2810 : return node;
1491 :
1492 : /*
1493 : * If we have an ungrouped Var of the original query level, we have a
1494 : * failure. Vars below the original query level are not a problem, and
1495 : * neither are Vars from above it. (If such Vars are ungrouped as far as
1496 : * their own query level is concerned, that's someone else's problem...)
1497 : */
1498 87575 : if (IsA(node, Var))
1499 : {
1500 6409 : Var *var = (Var *) node;
1501 : RangeTblEntry *rte;
1502 : char *attname;
1503 :
1504 6409 : if (var->varlevelsup != context->sublevels_up)
1505 290 : return node; /* it's not local to my query, ignore */
1506 :
1507 : /*
1508 : * Check for a match, if we didn't do it above.
1509 : */
1510 6119 : if (!context->have_non_var_grouping)
1511 : {
1512 6115 : int attnum = 0;
1513 :
1514 8730 : foreach(gl, context->groupClauses)
1515 : {
1516 8498 : TargetEntry *tle = (TargetEntry *) lfirst(gl);
1517 8498 : Var *gvar = (Var *) tle->expr;
1518 :
1519 8498 : attnum++;
1520 8498 : if (IsA(gvar, Var) &&
1521 8498 : gvar->varno == var->varno &&
1522 8076 : gvar->varattno == var->varattno &&
1523 5883 : gvar->varlevelsup == 0)
1524 : {
1525 : /* acceptable, replace it with a GROUP Var */
1526 5883 : return (Node *) buildGroupedVar(attnum,
1527 : tle->ressortgroupref,
1528 : context);
1529 : }
1530 : }
1531 : }
1532 :
1533 : /*
1534 : * Check whether the Var is known functionally dependent on the GROUP
1535 : * BY columns. If so, we can allow the Var to be used, because the
1536 : * grouping is really a no-op for this table. However, this deduction
1537 : * depends on one or more constraints of the table, so we have to add
1538 : * those constraints to the query's constraintDeps list, because it's
1539 : * not semantically valid anymore if the constraint(s) get dropped.
1540 : * (Therefore, this check must be the last-ditch effort before raising
1541 : * error: we don't want to add dependencies unnecessarily.)
1542 : *
1543 : * Because this is a pretty expensive check, and will have the same
1544 : * outcome for all columns of a table, we remember which RTEs we've
1545 : * already proven functional dependency for in the func_grouped_rels
1546 : * list. This test also prevents us from adding duplicate entries to
1547 : * the constraintDeps list.
1548 : */
1549 236 : if (list_member_int(*context->func_grouped_rels, var->varno))
1550 92 : return node; /* previously proven acceptable */
1551 :
1552 : Assert(var->varno > 0 &&
1553 : (int) var->varno <= list_length(context->pstate->p_rtable));
1554 144 : rte = rt_fetch(var->varno, context->pstate->p_rtable);
1555 144 : if (rte->rtekind == RTE_RELATION)
1556 : {
1557 140 : if (check_functional_grouping(rte->relid,
1558 140 : var->varno,
1559 : 0,
1560 : context->groupClauseCommonVars,
1561 140 : &context->qry->constraintDeps))
1562 : {
1563 160 : *context->func_grouped_rels =
1564 80 : lappend_int(*context->func_grouped_rels, var->varno);
1565 80 : return node; /* acceptable */
1566 : }
1567 : }
1568 :
1569 : /* Found an ungrouped local variable; generate error message */
1570 64 : attname = get_rte_attribute_name(rte, var->varattno);
1571 64 : if (context->sublevels_up == 0)
1572 64 : ereport(ERROR,
1573 : (errcode(ERRCODE_GROUPING_ERROR),
1574 : errmsg("column \"%s.%s\" must appear in the GROUP BY clause or be used in an aggregate function",
1575 : rte->eref->aliasname, attname),
1576 : context->in_agg_direct_args ?
1577 : errdetail("Direct arguments of an ordered-set aggregate must use only grouped columns.") : 0,
1578 : parser_errposition(context->pstate, var->location)));
1579 : else
1580 0 : ereport(ERROR,
1581 : (errcode(ERRCODE_GROUPING_ERROR),
1582 : errmsg("subquery uses ungrouped column \"%s.%s\" from outer query",
1583 : rte->eref->aliasname, attname),
1584 : parser_errposition(context->pstate, var->location)));
1585 : }
1586 :
1587 81166 : if (IsA(node, Query))
1588 : {
1589 : /* Recurse into subselects */
1590 : Query *newnode;
1591 :
1592 247 : context->sublevels_up++;
1593 :
1594 : /*
1595 : * If we have non-Var grouping expressions, we'll need a copy of the
1596 : * groupClauses list that's mutated to match this sublevels_up depth.
1597 : * Build one if we've not yet visited a subquery at this depth.
1598 : */
1599 307 : if (context->have_non_var_grouping &&
1600 60 : context->sublevels_up > list_length(context->groupClauseSubLevels))
1601 : {
1602 36 : List *subGroupClauses = copyObject(context->groupClauses);
1603 :
1604 36 : IncrementVarSublevelsUp((Node *) subGroupClauses,
1605 : context->sublevels_up, 0);
1606 36 : context->groupClauseSubLevels =
1607 36 : lappend(context->groupClauseSubLevels, subGroupClauses);
1608 : Assert(context->sublevels_up == list_length(context->groupClauseSubLevels));
1609 : }
1610 :
1611 247 : newnode = query_tree_mutator((Query *) node,
1612 : substitute_grouped_columns_mutator,
1613 : context,
1614 : 0);
1615 247 : context->sublevels_up--;
1616 247 : return (Node *) newnode;
1617 : }
1618 80919 : return expression_tree_mutator(node, substitute_grouped_columns_mutator,
1619 : context);
1620 : }
1621 :
1622 : /*
1623 : * finalize_grouping_exprs -
1624 : * Scan the given expression tree for GROUPING() and related calls,
1625 : * and validate and process their arguments.
1626 : *
1627 : * This is split out from substitute_grouped_columns above because it needs
1628 : * to modify the nodes (which it does in-place, not via a mutator) while
1629 : * substitute_grouped_columns may see only a copy of the original thanks to
1630 : * flattening of join alias vars. So here, we flatten each individual
1631 : * GROUPING argument as we see it before comparing it.
1632 : */
1633 : static void
1634 54476 : finalize_grouping_exprs(Node *node, ParseState *pstate, Query *qry,
1635 : List *groupClauses, bool hasJoinRTEs,
1636 : bool have_non_var_grouping)
1637 : {
1638 : substitute_grouped_columns_context context;
1639 :
1640 54476 : context.pstate = pstate;
1641 54476 : context.qry = qry;
1642 54476 : context.hasJoinRTEs = hasJoinRTEs;
1643 54476 : context.groupClauses = groupClauses;
1644 54476 : context.groupClauseCommonVars = NIL;
1645 54476 : context.groupClauseSubLevels = NIL;
1646 54476 : context.gset_common = NIL;
1647 54476 : context.have_non_var_grouping = have_non_var_grouping;
1648 54476 : context.func_grouped_rels = NULL;
1649 54476 : context.sublevels_up = 0;
1650 54476 : context.in_agg_direct_args = false;
1651 54476 : finalize_grouping_exprs_walker(node, &context);
1652 54472 : }
1653 :
1654 : static bool
1655 181733 : finalize_grouping_exprs_walker(Node *node,
1656 : substitute_grouped_columns_context *context)
1657 : {
1658 : ListCell *gl;
1659 :
1660 181733 : if (node == NULL)
1661 62227 : return false;
1662 119506 : if (IsA(node, Const) ||
1663 115988 : IsA(node, Param))
1664 3684 : return false; /* constants are always acceptable */
1665 :
1666 115822 : if (IsA(node, Aggref))
1667 : {
1668 30662 : Aggref *agg = (Aggref *) node;
1669 :
1670 30662 : if ((int) agg->agglevelsup == context->sublevels_up)
1671 : {
1672 : /*
1673 : * If we find an aggregate call of the original level, do not
1674 : * recurse into its normal arguments, ORDER BY arguments, or
1675 : * filter; GROUPING exprs of this level are not allowed there. But
1676 : * check direct arguments as though they weren't in an aggregate.
1677 : */
1678 : bool result;
1679 :
1680 : Assert(!context->in_agg_direct_args);
1681 30658 : context->in_agg_direct_args = true;
1682 30658 : result = finalize_grouping_exprs_walker((Node *) agg->aggdirectargs,
1683 : context);
1684 30658 : context->in_agg_direct_args = false;
1685 30658 : return result;
1686 : }
1687 :
1688 : /*
1689 : * We can skip recursing into aggregates of higher levels altogether,
1690 : * since they could not possibly contain exprs of concern to us (see
1691 : * transformAggregateCall). We do need to look at aggregates of lower
1692 : * levels, however.
1693 : */
1694 4 : if ((int) agg->agglevelsup > context->sublevels_up)
1695 0 : return false;
1696 : }
1697 :
1698 85164 : if (IsA(node, GroupingFunc))
1699 : {
1700 278 : GroupingFunc *grp = (GroupingFunc *) node;
1701 :
1702 : /*
1703 : * We only need to check GroupingFunc nodes at the exact level to
1704 : * which they belong, since they cannot mix levels in arguments.
1705 : */
1706 :
1707 278 : if ((int) grp->agglevelsup == context->sublevels_up)
1708 : {
1709 : ListCell *lc;
1710 252 : List *ref_list = NIL;
1711 :
1712 634 : foreach(lc, grp->args)
1713 : {
1714 386 : Node *expr = lfirst(lc);
1715 386 : Index ref = 0;
1716 :
1717 386 : if (context->hasJoinRTEs)
1718 48 : expr = flatten_join_alias_for_parser(context->qry,
1719 : expr,
1720 : context->sublevels_up);
1721 :
1722 : /*
1723 : * Each expression must match a grouping entry at the current
1724 : * query level. Unlike the general expression case, we don't
1725 : * allow functional dependencies or outer references.
1726 : */
1727 :
1728 386 : if (IsA(expr, Var))
1729 : {
1730 330 : Var *var = (Var *) expr;
1731 :
1732 330 : if (var->varlevelsup == context->sublevels_up)
1733 : {
1734 485 : foreach(gl, context->groupClauses)
1735 : {
1736 481 : TargetEntry *tle = lfirst(gl);
1737 481 : Var *gvar = (Var *) tle->expr;
1738 :
1739 481 : if (IsA(gvar, Var) &&
1740 481 : gvar->varno == var->varno &&
1741 473 : gvar->varattno == var->varattno &&
1742 326 : gvar->varlevelsup == 0)
1743 : {
1744 326 : ref = tle->ressortgroupref;
1745 326 : break;
1746 : }
1747 : }
1748 : }
1749 : }
1750 56 : else if (context->have_non_var_grouping)
1751 : {
1752 : List *groupClauses;
1753 :
1754 : /*
1755 : * Within a subquery, we need a mutated version of the
1756 : * groupClauses
1757 : */
1758 56 : if (context->sublevels_up == 0)
1759 48 : groupClauses = context->groupClauses;
1760 : else
1761 8 : groupClauses = list_nth(context->groupClauseSubLevels,
1762 8 : context->sublevels_up - 1);
1763 :
1764 104 : foreach(gl, groupClauses)
1765 : {
1766 104 : TargetEntry *tle = lfirst(gl);
1767 :
1768 104 : if (equal(expr, tle->expr))
1769 : {
1770 56 : ref = tle->ressortgroupref;
1771 56 : break;
1772 : }
1773 : }
1774 : }
1775 :
1776 386 : if (ref == 0)
1777 4 : ereport(ERROR,
1778 : (errcode(ERRCODE_GROUPING_ERROR),
1779 : errmsg("arguments to GROUPING must be grouping expressions of the associated query level"),
1780 : parser_errposition(context->pstate,
1781 : exprLocation(expr))));
1782 :
1783 382 : ref_list = lappend_int(ref_list, ref);
1784 : }
1785 :
1786 248 : grp->refs = ref_list;
1787 : }
1788 :
1789 274 : if ((int) grp->agglevelsup > context->sublevels_up)
1790 5 : return false;
1791 : }
1792 :
1793 85155 : if (IsA(node, Query))
1794 : {
1795 : /* Recurse into subselects */
1796 : bool result;
1797 :
1798 313 : context->sublevels_up++;
1799 :
1800 : /*
1801 : * If we have non-Var grouping expressions, we'll need a copy of the
1802 : * groupClauses list that's mutated to match this sublevels_up depth.
1803 : * Build one if we've not yet visited a subquery at this depth.
1804 : */
1805 439 : if (context->have_non_var_grouping &&
1806 126 : context->sublevels_up > list_length(context->groupClauseSubLevels))
1807 : {
1808 54 : List *subGroupClauses = copyObject(context->groupClauses);
1809 :
1810 54 : IncrementVarSublevelsUp((Node *) subGroupClauses,
1811 : context->sublevels_up, 0);
1812 54 : context->groupClauseSubLevels =
1813 54 : lappend(context->groupClauseSubLevels, subGroupClauses);
1814 : Assert(context->sublevels_up == list_length(context->groupClauseSubLevels));
1815 : }
1816 :
1817 313 : result = query_tree_walker((Query *) node,
1818 : finalize_grouping_exprs_walker,
1819 : context,
1820 : 0);
1821 313 : context->sublevels_up--;
1822 313 : return result;
1823 : }
1824 84842 : return expression_tree_walker(node, finalize_grouping_exprs_walker,
1825 : context);
1826 : }
1827 :
1828 : /*
1829 : * buildGroupedVar -
1830 : * build a Var node that references the RTE_GROUP RTE
1831 : */
1832 : static Var *
1833 7421 : buildGroupedVar(int attnum, Index ressortgroupref,
1834 : substitute_grouped_columns_context *context)
1835 : {
1836 : Var *var;
1837 7421 : ParseNamespaceItem *grouping_nsitem = context->pstate->p_grouping_nsitem;
1838 7421 : ParseNamespaceColumn *nscol = grouping_nsitem->p_nscolumns + attnum - 1;
1839 :
1840 : Assert(nscol->p_varno == grouping_nsitem->p_rtindex);
1841 : Assert(nscol->p_varattno == attnum);
1842 7421 : var = makeVar(nscol->p_varno,
1843 7421 : nscol->p_varattno,
1844 : nscol->p_vartype,
1845 : nscol->p_vartypmod,
1846 : nscol->p_varcollid,
1847 7421 : context->sublevels_up);
1848 : /* makeVar doesn't offer parameters for these, so set by hand: */
1849 7421 : var->varnosyn = nscol->p_varnosyn;
1850 7421 : var->varattnosyn = nscol->p_varattnosyn;
1851 :
1852 7421 : if (context->qry->groupingSets &&
1853 1524 : !list_member_int(context->gset_common, ressortgroupref))
1854 1372 : var->varnullingrels =
1855 1372 : bms_add_member(var->varnullingrels, grouping_nsitem->p_rtindex);
1856 :
1857 7421 : return var;
1858 : }
1859 :
1860 :
1861 : /*
1862 : * Given a GroupingSet node, expand it and return a list of lists.
1863 : *
1864 : * For EMPTY nodes, return a list of one empty list.
1865 : *
1866 : * For SIMPLE nodes, return a list of one list, which is the node content.
1867 : *
1868 : * For CUBE and ROLLUP nodes, return a list of the expansions.
1869 : *
1870 : * For SET nodes, recursively expand contained CUBE and ROLLUP.
1871 : */
1872 : static List *
1873 3927 : expand_groupingset_node(GroupingSet *gs)
1874 : {
1875 3927 : List *result = NIL;
1876 :
1877 3927 : switch (gs->kind)
1878 : {
1879 520 : case GROUPING_SET_EMPTY:
1880 520 : result = list_make1(NIL);
1881 520 : break;
1882 :
1883 1753 : case GROUPING_SET_SIMPLE:
1884 1753 : result = list_make1(gs->content);
1885 1753 : break;
1886 :
1887 469 : case GROUPING_SET_ROLLUP:
1888 : {
1889 469 : List *rollup_val = gs->content;
1890 : ListCell *lc;
1891 469 : int curgroup_size = list_length(gs->content);
1892 :
1893 1196 : while (curgroup_size > 0)
1894 : {
1895 727 : List *current_result = NIL;
1896 727 : int i = curgroup_size;
1897 :
1898 985 : foreach(lc, rollup_val)
1899 : {
1900 985 : GroupingSet *gs_current = (GroupingSet *) lfirst(lc);
1901 :
1902 : Assert(gs_current->kind == GROUPING_SET_SIMPLE);
1903 :
1904 985 : current_result = list_concat(current_result,
1905 985 : gs_current->content);
1906 :
1907 : /* If we are done with making the current group, break */
1908 985 : if (--i == 0)
1909 727 : break;
1910 : }
1911 :
1912 727 : result = lappend(result, current_result);
1913 727 : --curgroup_size;
1914 : }
1915 :
1916 469 : result = lappend(result, NIL);
1917 : }
1918 469 : break;
1919 :
1920 274 : case GROUPING_SET_CUBE:
1921 : {
1922 274 : List *cube_list = gs->content;
1923 274 : int number_bits = list_length(cube_list);
1924 : uint32 num_sets;
1925 : uint32 i;
1926 :
1927 : /* parser should cap this much lower */
1928 : Assert(number_bits < 31);
1929 :
1930 274 : num_sets = (1U << number_bits);
1931 :
1932 1524 : for (i = 0; i < num_sets; i++)
1933 : {
1934 1250 : List *current_result = NIL;
1935 : ListCell *lc;
1936 1250 : uint32 mask = 1U;
1937 :
1938 4156 : foreach(lc, cube_list)
1939 : {
1940 2906 : GroupingSet *gs_current = (GroupingSet *) lfirst(lc);
1941 :
1942 : Assert(gs_current->kind == GROUPING_SET_SIMPLE);
1943 :
1944 2906 : if (mask & i)
1945 1453 : current_result = list_concat(current_result,
1946 1453 : gs_current->content);
1947 :
1948 2906 : mask <<= 1;
1949 : }
1950 :
1951 1250 : result = lappend(result, current_result);
1952 : }
1953 : }
1954 274 : break;
1955 :
1956 911 : case GROUPING_SET_SETS:
1957 : {
1958 : ListCell *lc;
1959 :
1960 3104 : foreach(lc, gs->content)
1961 : {
1962 2193 : List *current_result = expand_groupingset_node(lfirst(lc));
1963 :
1964 2193 : result = list_concat(result, current_result);
1965 : }
1966 : }
1967 911 : break;
1968 : }
1969 :
1970 3927 : return result;
1971 : }
1972 :
1973 : /* list_sort comparator to sort sub-lists by length */
1974 : static int
1975 4566 : cmp_list_len_asc(const ListCell *a, const ListCell *b)
1976 : {
1977 4566 : int la = list_length((const List *) lfirst(a));
1978 4566 : int lb = list_length((const List *) lfirst(b));
1979 :
1980 4566 : return pg_cmp_s32(la, lb);
1981 : }
1982 :
1983 : /* list_sort comparator to sort sub-lists by length and contents */
1984 : static int
1985 224 : cmp_list_len_contents_asc(const ListCell *a, const ListCell *b)
1986 : {
1987 224 : int res = cmp_list_len_asc(a, b);
1988 :
1989 224 : if (res == 0)
1990 : {
1991 70 : List *la = (List *) lfirst(a);
1992 70 : List *lb = (List *) lfirst(b);
1993 : ListCell *lca;
1994 : ListCell *lcb;
1995 :
1996 158 : forboth(lca, la, lcb, lb)
1997 : {
1998 110 : int va = lfirst_int(lca);
1999 110 : int vb = lfirst_int(lcb);
2000 :
2001 110 : if (va > vb)
2002 22 : return 1;
2003 99 : if (va < vb)
2004 11 : return -1;
2005 : }
2006 : }
2007 :
2008 202 : return res;
2009 : }
2010 :
2011 : /*
2012 : * Expand a groupingSets clause to a flat list of grouping sets.
2013 : * The returned list is sorted by length, shortest sets first.
2014 : *
2015 : * This is mainly for the planner, but we use it here too to do
2016 : * some consistency checks.
2017 : */
2018 : List *
2019 1614 : expand_grouping_sets(List *groupingSets, bool groupDistinct, int limit)
2020 : {
2021 1614 : List *expanded_groups = NIL;
2022 1614 : List *result = NIL;
2023 1614 : double numsets = 1;
2024 : ListCell *lc;
2025 :
2026 1614 : if (groupingSets == NIL)
2027 0 : return NIL;
2028 :
2029 3348 : foreach(lc, groupingSets)
2030 : {
2031 1734 : List *current_result = NIL;
2032 1734 : GroupingSet *gs = lfirst(lc);
2033 :
2034 1734 : current_result = expand_groupingset_node(gs);
2035 :
2036 : Assert(current_result != NIL);
2037 :
2038 1734 : numsets *= list_length(current_result);
2039 :
2040 1734 : if (limit >= 0 && numsets > limit)
2041 0 : return NIL;
2042 :
2043 1734 : expanded_groups = lappend(expanded_groups, current_result);
2044 : }
2045 :
2046 : /*
2047 : * Do cartesian product between sublists of expanded_groups. While at it,
2048 : * remove any duplicate elements from individual grouping sets (we must
2049 : * NOT change the number of sets though)
2050 : */
2051 :
2052 6097 : foreach(lc, (List *) linitial(expanded_groups))
2053 : {
2054 4483 : result = lappend(result, list_union_int(NIL, (List *) lfirst(lc)));
2055 : }
2056 :
2057 1734 : for_each_from(lc, expanded_groups, 1)
2058 : {
2059 120 : List *p = lfirst(lc);
2060 120 : List *new_result = NIL;
2061 : ListCell *lc2;
2062 :
2063 365 : foreach(lc2, result)
2064 : {
2065 245 : List *q = lfirst(lc2);
2066 : ListCell *lc3;
2067 :
2068 784 : foreach(lc3, p)
2069 : {
2070 539 : new_result = lappend(new_result,
2071 539 : list_union_int(q, (List *) lfirst(lc3)));
2072 : }
2073 : }
2074 120 : result = new_result;
2075 : }
2076 :
2077 : /* Now sort the lists by length and deduplicate if necessary */
2078 1614 : if (!groupDistinct || list_length(result) < 2)
2079 1599 : list_sort(result, cmp_list_len_asc);
2080 : else
2081 : {
2082 : ListCell *cell;
2083 : List *prev;
2084 :
2085 : /* Sort each groupset individually */
2086 122 : foreach(cell, result)
2087 107 : list_sort(lfirst(cell), list_int_cmp);
2088 :
2089 : /* Now sort the list of groupsets by length and contents */
2090 15 : list_sort(result, cmp_list_len_contents_asc);
2091 :
2092 : /* Finally, remove duplicates */
2093 15 : prev = linitial(result);
2094 107 : for_each_from(cell, result, 1)
2095 : {
2096 92 : if (equal(lfirst(cell), prev))
2097 48 : result = foreach_delete_current(result, cell);
2098 : else
2099 44 : prev = lfirst(cell);
2100 : }
2101 : }
2102 :
2103 1614 : return result;
2104 : }
2105 :
2106 : /*
2107 : * get_aggregate_argtypes
2108 : * Identify the specific datatypes passed to an aggregate call.
2109 : *
2110 : * Given an Aggref, extract the actual datatypes of the input arguments.
2111 : * The input datatypes are reported in a way that matches up with the
2112 : * aggregate's declaration, ie, any ORDER BY columns attached to a plain
2113 : * aggregate are ignored, but we report both direct and aggregated args of
2114 : * an ordered-set aggregate.
2115 : *
2116 : * Datatypes are returned into inputTypes[], which must reference an array
2117 : * of length FUNC_MAX_ARGS.
2118 : *
2119 : * The function result is the number of actual arguments.
2120 : */
2121 : int
2122 78425 : get_aggregate_argtypes(Aggref *aggref, Oid *inputTypes)
2123 : {
2124 78425 : int numArguments = 0;
2125 : ListCell *lc;
2126 :
2127 : Assert(list_length(aggref->aggargtypes) <= FUNC_MAX_ARGS);
2128 :
2129 137829 : foreach(lc, aggref->aggargtypes)
2130 : {
2131 59404 : inputTypes[numArguments++] = lfirst_oid(lc);
2132 : }
2133 :
2134 78425 : return numArguments;
2135 : }
2136 :
2137 : /*
2138 : * resolve_aggregate_transtype
2139 : * Identify the transition state value's datatype for an aggregate call.
2140 : *
2141 : * This function resolves a polymorphic aggregate's state datatype.
2142 : * It must be passed the aggtranstype from the aggregate's catalog entry,
2143 : * as well as the actual argument types extracted by get_aggregate_argtypes.
2144 : * (We could fetch pg_aggregate.aggtranstype internally, but all existing
2145 : * callers already have the value at hand, so we make them pass it.)
2146 : */
2147 : Oid
2148 38839 : resolve_aggregate_transtype(Oid aggfuncid,
2149 : Oid aggtranstype,
2150 : Oid *inputTypes,
2151 : int numArguments)
2152 : {
2153 : /* resolve actual type of transition state, if polymorphic */
2154 38839 : if (IsPolymorphicType(aggtranstype))
2155 : {
2156 : /* have to fetch the agg's declared input types... */
2157 : Oid *declaredArgTypes;
2158 : int agg_nargs;
2159 :
2160 449 : (void) get_func_signature(aggfuncid, &declaredArgTypes, &agg_nargs);
2161 :
2162 : /*
2163 : * VARIADIC ANY aggs could have more actual than declared args, but
2164 : * such extra args can't affect polymorphic type resolution.
2165 : */
2166 : Assert(agg_nargs <= numArguments);
2167 :
2168 449 : aggtranstype = enforce_generic_type_consistency(inputTypes,
2169 : declaredArgTypes,
2170 : agg_nargs,
2171 : aggtranstype,
2172 : false);
2173 449 : pfree(declaredArgTypes);
2174 : }
2175 38839 : return aggtranstype;
2176 : }
2177 :
2178 : /*
2179 : * agg_args_support_sendreceive
2180 : * Returns true if all non-byval types of aggref's args have send and
2181 : * receive functions.
2182 : */
2183 : bool
2184 10588 : agg_args_support_sendreceive(Aggref *aggref)
2185 : {
2186 : ListCell *lc;
2187 :
2188 21092 : foreach(lc, aggref->args)
2189 : {
2190 : HeapTuple typeTuple;
2191 : Form_pg_type pt;
2192 10588 : TargetEntry *tle = (TargetEntry *) lfirst(lc);
2193 10588 : Oid type = exprType((Node *) tle->expr);
2194 :
2195 : /*
2196 : * RECORD is a special case: it has typsend/typreceive functions, but
2197 : * record_recv only works if passed the correct typmod to identify the
2198 : * specific anonymous record type. array_agg_deserialize cannot do
2199 : * that, so we have to disclaim support for the case.
2200 : */
2201 10588 : if (type == RECORDOID)
2202 84 : return false;
2203 :
2204 10553 : typeTuple = SearchSysCache1(TYPEOID, ObjectIdGetDatum(type));
2205 10553 : if (!HeapTupleIsValid(typeTuple))
2206 0 : elog(ERROR, "cache lookup failed for type %u", type);
2207 :
2208 10553 : pt = (Form_pg_type) GETSTRUCT(typeTuple);
2209 :
2210 10553 : if (!pt->typbyval &&
2211 9092 : (!OidIsValid(pt->typsend) || !OidIsValid(pt->typreceive)))
2212 : {
2213 49 : ReleaseSysCache(typeTuple);
2214 49 : return false;
2215 : }
2216 10504 : ReleaseSysCache(typeTuple);
2217 : }
2218 10504 : return true;
2219 : }
2220 :
2221 : /*
2222 : * Create an expression tree for the transition function of an aggregate.
2223 : * This is needed so that polymorphic functions can be used within an
2224 : * aggregate --- without the expression tree, such functions would not know
2225 : * the datatypes they are supposed to use. (The trees will never actually
2226 : * be executed, however, so we can skimp a bit on correctness.)
2227 : *
2228 : * agg_input_types and agg_state_type identifies the input types of the
2229 : * aggregate. These should be resolved to actual types (ie, none should
2230 : * ever be ANYELEMENT etc).
2231 : * agg_input_collation is the aggregate function's input collation.
2232 : *
2233 : * For an ordered-set aggregate, remember that agg_input_types describes
2234 : * the direct arguments followed by the aggregated arguments.
2235 : *
2236 : * transfn_oid and invtransfn_oid identify the funcs to be called; the
2237 : * latter may be InvalidOid, however if invtransfn_oid is set then
2238 : * transfn_oid must also be set.
2239 : *
2240 : * transfn_oid may also be passed as the aggcombinefn when the *transfnexpr is
2241 : * to be used for a combine aggregate phase. We expect invtransfn_oid to be
2242 : * InvalidOid in this case since there is no such thing as an inverse
2243 : * combinefn.
2244 : *
2245 : * Pointers to the constructed trees are returned into *transfnexpr,
2246 : * *invtransfnexpr. If there is no invtransfn, the respective pointer is set
2247 : * to NULL. Since use of the invtransfn is optional, NULL may be passed for
2248 : * invtransfnexpr.
2249 : */
2250 : void
2251 38966 : build_aggregate_transfn_expr(Oid *agg_input_types,
2252 : int agg_num_inputs,
2253 : int agg_num_direct_inputs,
2254 : bool agg_variadic,
2255 : Oid agg_state_type,
2256 : Oid agg_input_collation,
2257 : Oid transfn_oid,
2258 : Oid invtransfn_oid,
2259 : Expr **transfnexpr,
2260 : Expr **invtransfnexpr)
2261 : {
2262 : List *args;
2263 : FuncExpr *fexpr;
2264 : int i;
2265 :
2266 : /*
2267 : * Build arg list to use in the transfn FuncExpr node.
2268 : */
2269 38966 : args = list_make1(make_agg_arg(agg_state_type, agg_input_collation));
2270 :
2271 69822 : for (i = agg_num_direct_inputs; i < agg_num_inputs; i++)
2272 : {
2273 30856 : args = lappend(args,
2274 30856 : make_agg_arg(agg_input_types[i], agg_input_collation));
2275 : }
2276 :
2277 38966 : fexpr = makeFuncExpr(transfn_oid,
2278 : agg_state_type,
2279 : args,
2280 : InvalidOid,
2281 : agg_input_collation,
2282 : COERCE_EXPLICIT_CALL);
2283 38966 : fexpr->funcvariadic = agg_variadic;
2284 38966 : *transfnexpr = (Expr *) fexpr;
2285 :
2286 : /*
2287 : * Build invtransfn expression if requested, with same args as transfn
2288 : */
2289 38966 : if (invtransfnexpr != NULL)
2290 : {
2291 1087 : if (OidIsValid(invtransfn_oid))
2292 : {
2293 532 : fexpr = makeFuncExpr(invtransfn_oid,
2294 : agg_state_type,
2295 : args,
2296 : InvalidOid,
2297 : agg_input_collation,
2298 : COERCE_EXPLICIT_CALL);
2299 532 : fexpr->funcvariadic = agg_variadic;
2300 532 : *invtransfnexpr = (Expr *) fexpr;
2301 : }
2302 : else
2303 555 : *invtransfnexpr = NULL;
2304 : }
2305 38966 : }
2306 :
2307 : /*
2308 : * Like build_aggregate_transfn_expr, but creates an expression tree for the
2309 : * serialization function of an aggregate.
2310 : */
2311 : void
2312 224 : build_aggregate_serialfn_expr(Oid serialfn_oid,
2313 : Expr **serialfnexpr)
2314 : {
2315 : List *args;
2316 : FuncExpr *fexpr;
2317 :
2318 : /* serialfn always takes INTERNAL and returns BYTEA */
2319 224 : args = list_make1(make_agg_arg(INTERNALOID, InvalidOid));
2320 :
2321 224 : fexpr = makeFuncExpr(serialfn_oid,
2322 : BYTEAOID,
2323 : args,
2324 : InvalidOid,
2325 : InvalidOid,
2326 : COERCE_EXPLICIT_CALL);
2327 224 : *serialfnexpr = (Expr *) fexpr;
2328 224 : }
2329 :
2330 : /*
2331 : * Like build_aggregate_transfn_expr, but creates an expression tree for the
2332 : * deserialization function of an aggregate.
2333 : */
2334 : void
2335 80 : build_aggregate_deserialfn_expr(Oid deserialfn_oid,
2336 : Expr **deserialfnexpr)
2337 : {
2338 : List *args;
2339 : FuncExpr *fexpr;
2340 :
2341 : /* deserialfn always takes BYTEA, INTERNAL and returns INTERNAL */
2342 80 : args = list_make2(make_agg_arg(BYTEAOID, InvalidOid),
2343 : make_agg_arg(INTERNALOID, InvalidOid));
2344 :
2345 80 : fexpr = makeFuncExpr(deserialfn_oid,
2346 : INTERNALOID,
2347 : args,
2348 : InvalidOid,
2349 : InvalidOid,
2350 : COERCE_EXPLICIT_CALL);
2351 80 : *deserialfnexpr = (Expr *) fexpr;
2352 80 : }
2353 :
2354 : /*
2355 : * Like build_aggregate_transfn_expr, but creates an expression tree for the
2356 : * final function of an aggregate, rather than the transition function.
2357 : */
2358 : void
2359 17205 : build_aggregate_finalfn_expr(Oid *agg_input_types,
2360 : int num_finalfn_inputs,
2361 : Oid agg_state_type,
2362 : Oid agg_result_type,
2363 : Oid agg_input_collation,
2364 : Oid finalfn_oid,
2365 : Expr **finalfnexpr)
2366 : {
2367 : List *args;
2368 : int i;
2369 :
2370 : /*
2371 : * Build expr tree for final function
2372 : */
2373 17205 : args = list_make1(make_agg_arg(agg_state_type, agg_input_collation));
2374 :
2375 : /* finalfn may take additional args, which match agg's input types */
2376 28554 : for (i = 0; i < num_finalfn_inputs - 1; i++)
2377 : {
2378 11349 : args = lappend(args,
2379 11349 : make_agg_arg(agg_input_types[i], agg_input_collation));
2380 : }
2381 :
2382 17205 : *finalfnexpr = (Expr *) makeFuncExpr(finalfn_oid,
2383 : agg_result_type,
2384 : args,
2385 : InvalidOid,
2386 : agg_input_collation,
2387 : COERCE_EXPLICIT_CALL);
2388 : /* finalfn is currently never treated as variadic */
2389 17205 : }
2390 :
2391 : /*
2392 : * Convenience function to build dummy argument expressions for aggregates.
2393 : *
2394 : * We really only care that an aggregate support function can discover its
2395 : * actual argument types at runtime using get_fn_expr_argtype(), so it's okay
2396 : * to use Param nodes that don't correspond to any real Param.
2397 : */
2398 : static Node *
2399 98760 : make_agg_arg(Oid argtype, Oid argcollation)
2400 : {
2401 98760 : Param *argp = makeNode(Param);
2402 :
2403 98760 : argp->paramkind = PARAM_EXEC;
2404 98760 : argp->paramid = -1;
2405 98760 : argp->paramtype = argtype;
2406 98760 : argp->paramtypmod = -1;
2407 98760 : argp->paramcollid = argcollation;
2408 98760 : argp->location = -1;
2409 98760 : return (Node *) argp;
2410 : }
|
