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