Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * parse_clause.c
4 : * handle clauses in parser
5 : *
6 : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/parser/parse_clause.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 :
16 : #include "postgres.h"
17 :
18 : #include "access/htup_details.h"
19 : #include "access/nbtree.h"
20 : #include "access/table.h"
21 : #include "access/tsmapi.h"
22 : #include "catalog/catalog.h"
23 : #include "catalog/pg_am.h"
24 : #include "catalog/pg_amproc.h"
25 : #include "catalog/pg_constraint.h"
26 : #include "catalog/pg_type.h"
27 : #include "commands/defrem.h"
28 : #include "miscadmin.h"
29 : #include "nodes/makefuncs.h"
30 : #include "nodes/nodeFuncs.h"
31 : #include "optimizer/optimizer.h"
32 : #include "parser/analyze.h"
33 : #include "parser/parse_clause.h"
34 : #include "parser/parse_coerce.h"
35 : #include "parser/parse_collate.h"
36 : #include "parser/parse_expr.h"
37 : #include "parser/parse_func.h"
38 : #include "parser/parse_oper.h"
39 : #include "parser/parse_relation.h"
40 : #include "parser/parse_target.h"
41 : #include "parser/parse_type.h"
42 : #include "parser/parser.h"
43 : #include "rewrite/rewriteManip.h"
44 : #include "utils/builtins.h"
45 : #include "utils/catcache.h"
46 : #include "utils/lsyscache.h"
47 : #include "utils/rel.h"
48 : #include "utils/syscache.h"
49 :
50 :
51 : static int extractRemainingColumns(ParseState *pstate,
52 : ParseNamespaceColumn *src_nscolumns,
53 : List *src_colnames,
54 : List **src_colnos,
55 : List **res_colnames, List **res_colvars,
56 : ParseNamespaceColumn *res_nscolumns);
57 : static Node *transformJoinUsingClause(ParseState *pstate,
58 : List *leftVars, List *rightVars);
59 : static Node *transformJoinOnClause(ParseState *pstate, JoinExpr *j,
60 : List *namespace);
61 : static ParseNamespaceItem *transformTableEntry(ParseState *pstate, RangeVar *r);
62 : static ParseNamespaceItem *transformRangeSubselect(ParseState *pstate,
63 : RangeSubselect *r);
64 : static ParseNamespaceItem *transformRangeFunction(ParseState *pstate,
65 : RangeFunction *r);
66 : static ParseNamespaceItem *transformRangeTableFunc(ParseState *pstate,
67 : RangeTableFunc *rtf);
68 : static TableSampleClause *transformRangeTableSample(ParseState *pstate,
69 : RangeTableSample *rts);
70 : static ParseNamespaceItem *getNSItemForSpecialRelationTypes(ParseState *pstate,
71 : RangeVar *rv);
72 : static Node *transformFromClauseItem(ParseState *pstate, Node *n,
73 : ParseNamespaceItem **top_nsitem,
74 : List **namespace);
75 : static Var *buildVarFromNSColumn(ParseState *pstate,
76 : ParseNamespaceColumn *nscol);
77 : static Node *buildMergedJoinVar(ParseState *pstate, JoinType jointype,
78 : Var *l_colvar, Var *r_colvar);
79 : static void markRelsAsNulledBy(ParseState *pstate, Node *n, int jindex);
80 : static void setNamespaceColumnVisibility(List *namespace, bool cols_visible);
81 : static void setNamespaceLateralState(List *namespace,
82 : bool lateral_only, bool lateral_ok);
83 : static void checkExprIsVarFree(ParseState *pstate, Node *n,
84 : const char *constructName);
85 : static TargetEntry *findTargetlistEntrySQL92(ParseState *pstate, Node *node,
86 : List **tlist, ParseExprKind exprKind);
87 : static TargetEntry *findTargetlistEntrySQL99(ParseState *pstate, Node *node,
88 : List **tlist, ParseExprKind exprKind);
89 : static int get_matching_location(int sortgroupref,
90 : List *sortgrouprefs, List *exprs);
91 : static List *resolve_unique_index_expr(ParseState *pstate, InferClause *infer,
92 : Relation heapRel);
93 : static List *addTargetToGroupList(ParseState *pstate, TargetEntry *tle,
94 : List *grouplist, List *targetlist, int location);
95 : static WindowClause *findWindowClause(List *wclist, const char *name);
96 : static Node *transformFrameOffset(ParseState *pstate, int frameOptions,
97 : Oid rangeopfamily, Oid rangeopcintype, Oid *inRangeFunc,
98 : Node *clause);
99 :
100 :
101 : /*
102 : * transformFromClause -
103 : * Process the FROM clause and add items to the query's range table,
104 : * joinlist, and namespace.
105 : *
106 : * Note: we assume that the pstate's p_rtable, p_joinlist, and p_namespace
107 : * lists were initialized to NIL when the pstate was created.
108 : * We will add onto any entries already present --- this is needed for rule
109 : * processing, as well as for UPDATE and DELETE.
110 : */
111 : void
112 486492 : transformFromClause(ParseState *pstate, List *frmList)
113 : {
114 : ListCell *fl;
115 :
116 : /*
117 : * The grammar will have produced a list of RangeVars, RangeSubselects,
118 : * RangeFunctions, and/or JoinExprs. Transform each one (possibly adding
119 : * entries to the rtable), check for duplicate refnames, and then add it
120 : * to the joinlist and namespace.
121 : *
122 : * Note we must process the items left-to-right for proper handling of
123 : * LATERAL references.
124 : */
125 829268 : foreach(fl, frmList)
126 : {
127 343452 : Node *n = lfirst(fl);
128 : ParseNamespaceItem *nsitem;
129 : List *namespace;
130 :
131 343452 : n = transformFromClauseItem(pstate, n,
132 : &nsitem,
133 : &namespace);
134 :
135 342782 : checkNameSpaceConflicts(pstate, pstate->p_namespace, namespace);
136 :
137 : /* Mark the new namespace items as visible only to LATERAL */
138 342776 : setNamespaceLateralState(namespace, true, true);
139 :
140 342776 : pstate->p_joinlist = lappend(pstate->p_joinlist, n);
141 342776 : pstate->p_namespace = list_concat(pstate->p_namespace, namespace);
142 : }
143 :
144 : /*
145 : * We're done parsing the FROM list, so make all namespace items
146 : * unconditionally visible. Note that this will also reset lateral_only
147 : * for any namespace items that were already present when we were called;
148 : * but those should have been that way already.
149 : */
150 485816 : setNamespaceLateralState(pstate->p_namespace, false, true);
151 485816 : }
152 :
153 : /*
154 : * setTargetTable
155 : * Add the target relation of INSERT/UPDATE/DELETE/MERGE to the range table,
156 : * and make the special links to it in the ParseState.
157 : *
158 : * We also open the target relation and acquire a write lock on it.
159 : * This must be done before processing the FROM list, in case the target
160 : * is also mentioned as a source relation --- we want to be sure to grab
161 : * the write lock before any read lock.
162 : *
163 : * If alsoSource is true, add the target to the query's joinlist and
164 : * namespace. For INSERT, we don't want the target to be joined to;
165 : * it's a destination of tuples, not a source. MERGE is actually
166 : * both, but we'll add it separately to joinlist and namespace, so
167 : * doing nothing (like INSERT) is correct here. For UPDATE/DELETE,
168 : * we do need to scan or join the target. (NOTE: we do not bother
169 : * to check for namespace conflict; we assume that the namespace was
170 : * initially empty in these cases.)
171 : *
172 : * Finally, we mark the relation as requiring the permissions specified
173 : * by requiredPerms.
174 : *
175 : * Returns the rangetable index of the target relation.
176 : */
177 : int
178 91866 : setTargetTable(ParseState *pstate, RangeVar *relation,
179 : bool inh, bool alsoSource, AclMode requiredPerms)
180 : {
181 : ParseNamespaceItem *nsitem;
182 :
183 : /*
184 : * ENRs hide tables of the same name, so we need to check for them first.
185 : * In contrast, CTEs don't hide tables (for this purpose).
186 : */
187 173152 : if (relation->schemaname == NULL &&
188 81286 : scanNameSpaceForENR(pstate, relation->relname))
189 6 : ereport(ERROR,
190 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
191 : errmsg("relation \"%s\" cannot be the target of a modifying statement",
192 : relation->relname)));
193 :
194 : /* Close old target; this could only happen for multi-action rules */
195 91860 : if (pstate->p_target_relation != NULL)
196 0 : table_close(pstate->p_target_relation, NoLock);
197 :
198 : /*
199 : * Open target rel and grab suitable lock (which we will hold till end of
200 : * transaction).
201 : *
202 : * free_parsestate() will eventually do the corresponding table_close(),
203 : * but *not* release the lock.
204 : */
205 91860 : pstate->p_target_relation = parserOpenTable(pstate, relation,
206 : RowExclusiveLock);
207 :
208 : /*
209 : * Now build an RTE and a ParseNamespaceItem.
210 : */
211 91840 : nsitem = addRangeTableEntryForRelation(pstate, pstate->p_target_relation,
212 : RowExclusiveLock,
213 : relation->alias, inh, false);
214 :
215 : /* remember the RTE/nsitem as being the query target */
216 91840 : pstate->p_target_nsitem = nsitem;
217 :
218 : /*
219 : * Override addRangeTableEntry's default ACL_SELECT permissions check, and
220 : * instead mark target table as requiring exactly the specified
221 : * permissions.
222 : *
223 : * If we find an explicit reference to the rel later during parse
224 : * analysis, we will add the ACL_SELECT bit back again; see
225 : * markVarForSelectPriv and its callers.
226 : */
227 91840 : nsitem->p_perminfo->requiredPerms = requiredPerms;
228 :
229 : /*
230 : * If UPDATE/DELETE, add table to joinlist and namespace.
231 : */
232 91840 : if (alsoSource)
233 17388 : addNSItemToQuery(pstate, nsitem, true, true, true);
234 :
235 91840 : return nsitem->p_rtindex;
236 : }
237 :
238 : /*
239 : * Extract all not-in-common columns from column lists of a source table
240 : *
241 : * src_nscolumns and src_colnames describe the source table.
242 : *
243 : * *src_colnos initially contains the column numbers of the already-merged
244 : * columns. We add to it the column number of each additional column.
245 : * Also append to *res_colnames the name of each additional column,
246 : * append to *res_colvars a Var for each additional column, and copy the
247 : * columns' nscolumns data into res_nscolumns[] (which is caller-allocated
248 : * space that had better be big enough).
249 : *
250 : * Returns the number of columns added.
251 : */
252 : static int
253 149756 : extractRemainingColumns(ParseState *pstate,
254 : ParseNamespaceColumn *src_nscolumns,
255 : List *src_colnames,
256 : List **src_colnos,
257 : List **res_colnames, List **res_colvars,
258 : ParseNamespaceColumn *res_nscolumns)
259 : {
260 149756 : int colcount = 0;
261 : Bitmapset *prevcols;
262 : int attnum;
263 : ListCell *lc;
264 :
265 : /*
266 : * While we could just test "list_member_int(*src_colnos, attnum)" to
267 : * detect already-merged columns in the loop below, that would be O(N^2)
268 : * for a wide input table. Instead build a bitmapset of just the merged
269 : * USING columns, which we won't add to within the main loop.
270 : */
271 149756 : prevcols = NULL;
272 153072 : foreach(lc, *src_colnos)
273 : {
274 3316 : prevcols = bms_add_member(prevcols, lfirst_int(lc));
275 : }
276 :
277 149756 : attnum = 0;
278 2969740 : foreach(lc, src_colnames)
279 : {
280 2819984 : char *colname = strVal(lfirst(lc));
281 :
282 2819984 : attnum++;
283 : /* Non-dropped and not already merged? */
284 2819984 : if (colname[0] != '\0' && !bms_is_member(attnum, prevcols))
285 : {
286 : /* Yes, so emit it as next output column */
287 2816188 : *src_colnos = lappend_int(*src_colnos, attnum);
288 2816188 : *res_colnames = lappend(*res_colnames, lfirst(lc));
289 2816188 : *res_colvars = lappend(*res_colvars,
290 2816188 : buildVarFromNSColumn(pstate,
291 2816188 : src_nscolumns + attnum - 1));
292 : /* Copy the input relation's nscolumn data for this column */
293 2816188 : res_nscolumns[colcount] = src_nscolumns[attnum - 1];
294 2816188 : colcount++;
295 : }
296 : }
297 149756 : return colcount;
298 : }
299 :
300 : /* transformJoinUsingClause()
301 : * Build a complete ON clause from a partially-transformed USING list.
302 : * We are given lists of nodes representing left and right match columns.
303 : * Result is a transformed qualification expression.
304 : */
305 : static Node *
306 1458 : transformJoinUsingClause(ParseState *pstate,
307 : List *leftVars, List *rightVars)
308 : {
309 : Node *result;
310 1458 : List *andargs = NIL;
311 : ListCell *lvars,
312 : *rvars;
313 :
314 : /*
315 : * We cheat a little bit here by building an untransformed operator tree
316 : * whose leaves are the already-transformed Vars. This requires collusion
317 : * from transformExpr(), which normally could be expected to complain
318 : * about already-transformed subnodes. However, this does mean that we
319 : * have to mark the columns as requiring SELECT privilege for ourselves;
320 : * transformExpr() won't do it.
321 : */
322 3116 : forboth(lvars, leftVars, rvars, rightVars)
323 : {
324 1658 : Var *lvar = (Var *) lfirst(lvars);
325 1658 : Var *rvar = (Var *) lfirst(rvars);
326 : A_Expr *e;
327 :
328 : /* Require read access to the join variables */
329 1658 : markVarForSelectPriv(pstate, lvar);
330 1658 : markVarForSelectPriv(pstate, rvar);
331 :
332 : /* Now create the lvar = rvar join condition */
333 1658 : e = makeSimpleA_Expr(AEXPR_OP, "=",
334 1658 : (Node *) copyObject(lvar), (Node *) copyObject(rvar),
335 : -1);
336 :
337 : /* Prepare to combine into an AND clause, if multiple join columns */
338 1658 : andargs = lappend(andargs, e);
339 : }
340 :
341 : /* Only need an AND if there's more than one join column */
342 1458 : if (list_length(andargs) == 1)
343 1280 : result = (Node *) linitial(andargs);
344 : else
345 178 : result = (Node *) makeBoolExpr(AND_EXPR, andargs, -1);
346 :
347 : /*
348 : * Since the references are already Vars, and are certainly from the input
349 : * relations, we don't have to go through the same pushups that
350 : * transformJoinOnClause() does. Just invoke transformExpr() to fix up
351 : * the operators, and we're done.
352 : */
353 1458 : result = transformExpr(pstate, result, EXPR_KIND_JOIN_USING);
354 :
355 1458 : result = coerce_to_boolean(pstate, result, "JOIN/USING");
356 :
357 1458 : return result;
358 : }
359 :
360 : /* transformJoinOnClause()
361 : * Transform the qual conditions for JOIN/ON.
362 : * Result is a transformed qualification expression.
363 : */
364 : static Node *
365 73140 : transformJoinOnClause(ParseState *pstate, JoinExpr *j, List *namespace)
366 : {
367 : Node *result;
368 : List *save_namespace;
369 :
370 : /*
371 : * The namespace that the join expression should see is just the two
372 : * subtrees of the JOIN plus any outer references from upper pstate
373 : * levels. Temporarily set this pstate's namespace accordingly. (We need
374 : * not check for refname conflicts, because transformFromClauseItem()
375 : * already did.) All namespace items are marked visible regardless of
376 : * LATERAL state.
377 : */
378 73140 : setNamespaceLateralState(namespace, false, true);
379 :
380 73140 : save_namespace = pstate->p_namespace;
381 73140 : pstate->p_namespace = namespace;
382 :
383 73140 : result = transformWhereClause(pstate, j->quals,
384 : EXPR_KIND_JOIN_ON, "JOIN/ON");
385 :
386 73122 : pstate->p_namespace = save_namespace;
387 :
388 73122 : return result;
389 : }
390 :
391 : /*
392 : * transformTableEntry --- transform a RangeVar (simple relation reference)
393 : */
394 : static ParseNamespaceItem *
395 354326 : transformTableEntry(ParseState *pstate, RangeVar *r)
396 : {
397 : /* addRangeTableEntry does all the work */
398 354326 : return addRangeTableEntry(pstate, r, r->alias, r->inh, true);
399 : }
400 :
401 : /*
402 : * transformRangeSubselect --- transform a sub-SELECT appearing in FROM
403 : */
404 : static ParseNamespaceItem *
405 14512 : transformRangeSubselect(ParseState *pstate, RangeSubselect *r)
406 : {
407 : Query *query;
408 :
409 : /*
410 : * Set p_expr_kind to show this parse level is recursing to a subselect.
411 : * We can't be nested within any expression, so don't need save-restore
412 : * logic here.
413 : */
414 : Assert(pstate->p_expr_kind == EXPR_KIND_NONE);
415 14512 : pstate->p_expr_kind = EXPR_KIND_FROM_SUBSELECT;
416 :
417 : /*
418 : * If the subselect is LATERAL, make lateral_only names of this level
419 : * visible to it. (LATERAL can't nest within a single pstate level, so we
420 : * don't need save/restore logic here.)
421 : */
422 : Assert(!pstate->p_lateral_active);
423 14512 : pstate->p_lateral_active = r->lateral;
424 :
425 : /*
426 : * Analyze and transform the subquery. Note that if the subquery doesn't
427 : * have an alias, it can't be explicitly selected for locking, but locking
428 : * might still be required (if there is an all-tables locking clause).
429 : */
430 14512 : query = parse_sub_analyze(r->subquery, pstate, NULL,
431 14512 : isLockedRefname(pstate,
432 14512 : r->alias == NULL ? NULL :
433 14374 : r->alias->aliasname),
434 : true);
435 :
436 : /* Restore state */
437 14404 : pstate->p_lateral_active = false;
438 14404 : pstate->p_expr_kind = EXPR_KIND_NONE;
439 :
440 : /*
441 : * Check that we got a SELECT. Anything else should be impossible given
442 : * restrictions of the grammar, but check anyway.
443 : */
444 14404 : if (!IsA(query, Query) ||
445 14404 : query->commandType != CMD_SELECT)
446 0 : elog(ERROR, "unexpected non-SELECT command in subquery in FROM");
447 :
448 : /*
449 : * OK, build an RTE and nsitem for the subquery.
450 : */
451 28802 : return addRangeTableEntryForSubquery(pstate,
452 : query,
453 : r->alias,
454 14404 : r->lateral,
455 : true);
456 : }
457 :
458 :
459 : /*
460 : * transformRangeFunction --- transform a function call appearing in FROM
461 : */
462 : static ParseNamespaceItem *
463 42698 : transformRangeFunction(ParseState *pstate, RangeFunction *r)
464 : {
465 42698 : List *funcexprs = NIL;
466 42698 : List *funcnames = NIL;
467 42698 : List *coldeflists = NIL;
468 : bool is_lateral;
469 : ListCell *lc;
470 :
471 : /*
472 : * We make lateral_only names of this level visible, whether or not the
473 : * RangeFunction is explicitly marked LATERAL. This is needed for SQL
474 : * spec compliance in the case of UNNEST(), and seems useful on
475 : * convenience grounds for all functions in FROM.
476 : *
477 : * (LATERAL can't nest within a single pstate level, so we don't need
478 : * save/restore logic here.)
479 : */
480 : Assert(!pstate->p_lateral_active);
481 42698 : pstate->p_lateral_active = true;
482 :
483 : /*
484 : * Transform the raw expressions.
485 : *
486 : * While transforming, also save function names for possible use as alias
487 : * and column names. We use the same transformation rules as for a SELECT
488 : * output expression. For a FuncCall node, the result will be the
489 : * function name, but it is possible for the grammar to hand back other
490 : * node types.
491 : *
492 : * We have to get this info now, because FigureColname only works on raw
493 : * parsetrees. Actually deciding what to do with the names is left up to
494 : * addRangeTableEntryForFunction.
495 : *
496 : * Likewise, collect column definition lists if there were any. But
497 : * complain if we find one here and the RangeFunction has one too.
498 : */
499 85414 : foreach(lc, r->functions)
500 : {
501 42884 : List *pair = (List *) lfirst(lc);
502 : Node *fexpr;
503 : List *coldeflist;
504 : Node *newfexpr;
505 : Node *last_srf;
506 :
507 : /* Disassemble the function-call/column-def-list pairs */
508 : Assert(list_length(pair) == 2);
509 42884 : fexpr = (Node *) linitial(pair);
510 42884 : coldeflist = (List *) lsecond(pair);
511 :
512 : /*
513 : * If we find a function call unnest() with more than one argument and
514 : * no special decoration, transform it into separate unnest() calls on
515 : * each argument. This is a kluge, for sure, but it's less nasty than
516 : * other ways of implementing the SQL-standard UNNEST() syntax.
517 : *
518 : * If there is any decoration (including a coldeflist), we don't
519 : * transform, which probably means a no-such-function error later. We
520 : * could alternatively throw an error right now, but that doesn't seem
521 : * tremendously helpful. If someone is using any such decoration,
522 : * then they're not using the SQL-standard syntax, and they're more
523 : * likely expecting an un-tweaked function call.
524 : *
525 : * Note: the transformation changes a non-schema-qualified unnest()
526 : * function name into schema-qualified pg_catalog.unnest(). This
527 : * choice is also a bit debatable, but it seems reasonable to force
528 : * use of built-in unnest() when we make this transformation.
529 : */
530 42884 : if (IsA(fexpr, FuncCall))
531 : {
532 42740 : FuncCall *fc = (FuncCall *) fexpr;
533 :
534 42740 : if (list_length(fc->funcname) == 1 &&
535 31640 : strcmp(strVal(linitial(fc->funcname)), "unnest") == 0 &&
536 2470 : list_length(fc->args) > 1 &&
537 66 : fc->agg_order == NIL &&
538 66 : fc->agg_filter == NULL &&
539 66 : fc->over == NULL &&
540 66 : !fc->agg_star &&
541 66 : !fc->agg_distinct &&
542 66 : !fc->func_variadic &&
543 : coldeflist == NIL)
544 : {
545 : ListCell *lc2;
546 :
547 240 : foreach(lc2, fc->args)
548 : {
549 174 : Node *arg = (Node *) lfirst(lc2);
550 : FuncCall *newfc;
551 :
552 174 : last_srf = pstate->p_last_srf;
553 :
554 174 : newfc = makeFuncCall(SystemFuncName("unnest"),
555 174 : list_make1(arg),
556 : COERCE_EXPLICIT_CALL,
557 : fc->location);
558 :
559 174 : newfexpr = transformExpr(pstate, (Node *) newfc,
560 : EXPR_KIND_FROM_FUNCTION);
561 :
562 : /* nodeFunctionscan.c requires SRFs to be at top level */
563 174 : if (pstate->p_last_srf != last_srf &&
564 174 : pstate->p_last_srf != newfexpr)
565 0 : ereport(ERROR,
566 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
567 : errmsg("set-returning functions must appear at top level of FROM"),
568 : parser_errposition(pstate,
569 : exprLocation(pstate->p_last_srf))));
570 :
571 174 : funcexprs = lappend(funcexprs, newfexpr);
572 :
573 174 : funcnames = lappend(funcnames,
574 174 : FigureColname((Node *) newfc));
575 :
576 : /* coldeflist is empty, so no error is possible */
577 :
578 174 : coldeflists = lappend(coldeflists, coldeflist);
579 : }
580 66 : continue; /* done with this function item */
581 : }
582 : }
583 :
584 : /* normal case ... */
585 42818 : last_srf = pstate->p_last_srf;
586 :
587 42818 : newfexpr = transformExpr(pstate, fexpr,
588 : EXPR_KIND_FROM_FUNCTION);
589 :
590 : /* nodeFunctionscan.c requires SRFs to be at top level */
591 42656 : if (pstate->p_last_srf != last_srf &&
592 37274 : pstate->p_last_srf != newfexpr)
593 6 : ereport(ERROR,
594 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
595 : errmsg("set-returning functions must appear at top level of FROM"),
596 : parser_errposition(pstate,
597 : exprLocation(pstate->p_last_srf))));
598 :
599 42650 : funcexprs = lappend(funcexprs, newfexpr);
600 :
601 42650 : funcnames = lappend(funcnames,
602 42650 : FigureColname(fexpr));
603 :
604 42650 : if (coldeflist && r->coldeflist)
605 0 : ereport(ERROR,
606 : (errcode(ERRCODE_SYNTAX_ERROR),
607 : errmsg("multiple column definition lists are not allowed for the same function"),
608 : parser_errposition(pstate,
609 : exprLocation((Node *) r->coldeflist))));
610 :
611 42650 : coldeflists = lappend(coldeflists, coldeflist);
612 : }
613 :
614 42530 : pstate->p_lateral_active = false;
615 :
616 : /*
617 : * We must assign collations now so that the RTE exposes correct collation
618 : * info for Vars created from it.
619 : */
620 42530 : assign_list_collations(pstate, funcexprs);
621 :
622 : /*
623 : * Install the top-level coldeflist if there was one (we already checked
624 : * that there was no conflicting per-function coldeflist).
625 : *
626 : * We only allow this when there's a single function (even after UNNEST
627 : * expansion) and no WITH ORDINALITY. The reason for the latter
628 : * restriction is that it's not real clear whether the ordinality column
629 : * should be in the coldeflist, and users are too likely to make mistakes
630 : * in one direction or the other. Putting the coldeflist inside ROWS
631 : * FROM() is much clearer in this case.
632 : */
633 42530 : if (r->coldeflist)
634 : {
635 748 : if (list_length(funcexprs) != 1)
636 : {
637 0 : if (r->is_rowsfrom)
638 0 : ereport(ERROR,
639 : (errcode(ERRCODE_SYNTAX_ERROR),
640 : errmsg("ROWS FROM() with multiple functions cannot have a column definition list"),
641 : errhint("Put a separate column definition list for each function inside ROWS FROM()."),
642 : parser_errposition(pstate,
643 : exprLocation((Node *) r->coldeflist))));
644 : else
645 0 : ereport(ERROR,
646 : (errcode(ERRCODE_SYNTAX_ERROR),
647 : errmsg("UNNEST() with multiple arguments cannot have a column definition list"),
648 : errhint("Use separate UNNEST() calls inside ROWS FROM(), and attach a column definition list to each one."),
649 : parser_errposition(pstate,
650 : exprLocation((Node *) r->coldeflist))));
651 : }
652 748 : if (r->ordinality)
653 0 : ereport(ERROR,
654 : (errcode(ERRCODE_SYNTAX_ERROR),
655 : errmsg("WITH ORDINALITY cannot be used with a column definition list"),
656 : errhint("Put the column definition list inside ROWS FROM()."),
657 : parser_errposition(pstate,
658 : exprLocation((Node *) r->coldeflist))));
659 :
660 748 : coldeflists = list_make1(r->coldeflist);
661 : }
662 :
663 : /*
664 : * Mark the RTE as LATERAL if the user said LATERAL explicitly, or if
665 : * there are any lateral cross-references in it.
666 : */
667 42530 : is_lateral = r->lateral || contain_vars_of_level((Node *) funcexprs, 0);
668 :
669 : /*
670 : * OK, build an RTE and nsitem for the function.
671 : */
672 42530 : return addRangeTableEntryForFunction(pstate,
673 : funcnames, funcexprs, coldeflists,
674 : r, is_lateral, true);
675 : }
676 :
677 : /*
678 : * transformRangeTableFunc -
679 : * Transform a raw RangeTableFunc into TableFunc.
680 : *
681 : * Transform the namespace clauses, the document-generating expression, the
682 : * row-generating expression, the column-generating expressions, and the
683 : * default value expressions.
684 : */
685 : static ParseNamespaceItem *
686 220 : transformRangeTableFunc(ParseState *pstate, RangeTableFunc *rtf)
687 : {
688 220 : TableFunc *tf = makeNode(TableFunc);
689 : const char *constructName;
690 : Oid docType;
691 : bool is_lateral;
692 : ListCell *col;
693 : char **names;
694 : int colno;
695 :
696 : /*
697 : * Currently we only support XMLTABLE here. See transformJsonTable() for
698 : * JSON_TABLE support.
699 : */
700 220 : tf->functype = TFT_XMLTABLE;
701 220 : constructName = "XMLTABLE";
702 220 : docType = XMLOID;
703 :
704 : /*
705 : * We make lateral_only names of this level visible, whether or not the
706 : * RangeTableFunc is explicitly marked LATERAL. This is needed for SQL
707 : * spec compliance and seems useful on convenience grounds for all
708 : * functions in FROM.
709 : *
710 : * (LATERAL can't nest within a single pstate level, so we don't need
711 : * save/restore logic here.)
712 : */
713 : Assert(!pstate->p_lateral_active);
714 220 : pstate->p_lateral_active = true;
715 :
716 : /* Transform and apply typecast to the row-generating expression ... */
717 : Assert(rtf->rowexpr != NULL);
718 220 : tf->rowexpr = coerce_to_specific_type(pstate,
719 : transformExpr(pstate, rtf->rowexpr, EXPR_KIND_FROM_FUNCTION),
720 : TEXTOID,
721 : constructName);
722 220 : assign_expr_collations(pstate, tf->rowexpr);
723 :
724 : /* ... and to the document itself */
725 : Assert(rtf->docexpr != NULL);
726 220 : tf->docexpr = coerce_to_specific_type(pstate,
727 : transformExpr(pstate, rtf->docexpr, EXPR_KIND_FROM_FUNCTION),
728 : docType,
729 : constructName);
730 220 : assign_expr_collations(pstate, tf->docexpr);
731 :
732 : /* undef ordinality column number */
733 220 : tf->ordinalitycol = -1;
734 :
735 : /* Process column specs */
736 220 : names = palloc(sizeof(char *) * list_length(rtf->columns));
737 :
738 220 : colno = 0;
739 970 : foreach(col, rtf->columns)
740 : {
741 750 : RangeTableFuncCol *rawc = (RangeTableFuncCol *) lfirst(col);
742 : Oid typid;
743 : int32 typmod;
744 : Node *colexpr;
745 : Node *coldefexpr;
746 : int j;
747 :
748 750 : tf->colnames = lappend(tf->colnames,
749 750 : makeString(pstrdup(rawc->colname)));
750 :
751 : /*
752 : * Determine the type and typmod for the new column. FOR ORDINALITY
753 : * columns are INTEGER per spec; the others are user-specified.
754 : */
755 750 : if (rawc->for_ordinality)
756 : {
757 62 : if (tf->ordinalitycol != -1)
758 0 : ereport(ERROR,
759 : (errcode(ERRCODE_SYNTAX_ERROR),
760 : errmsg("only one FOR ORDINALITY column is allowed"),
761 : parser_errposition(pstate, rawc->location)));
762 :
763 62 : typid = INT4OID;
764 62 : typmod = -1;
765 62 : tf->ordinalitycol = colno;
766 : }
767 : else
768 : {
769 688 : if (rawc->typeName->setof)
770 0 : ereport(ERROR,
771 : (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
772 : errmsg("column \"%s\" cannot be declared SETOF",
773 : rawc->colname),
774 : parser_errposition(pstate, rawc->location)));
775 :
776 688 : typenameTypeIdAndMod(pstate, rawc->typeName,
777 : &typid, &typmod);
778 : }
779 :
780 750 : tf->coltypes = lappend_oid(tf->coltypes, typid);
781 750 : tf->coltypmods = lappend_int(tf->coltypmods, typmod);
782 750 : tf->colcollations = lappend_oid(tf->colcollations,
783 : get_typcollation(typid));
784 :
785 : /* Transform the PATH and DEFAULT expressions */
786 750 : if (rawc->colexpr)
787 : {
788 490 : colexpr = coerce_to_specific_type(pstate,
789 : transformExpr(pstate, rawc->colexpr,
790 : EXPR_KIND_FROM_FUNCTION),
791 : TEXTOID,
792 : constructName);
793 490 : assign_expr_collations(pstate, colexpr);
794 : }
795 : else
796 260 : colexpr = NULL;
797 :
798 750 : if (rawc->coldefexpr)
799 : {
800 56 : coldefexpr = coerce_to_specific_type_typmod(pstate,
801 : transformExpr(pstate, rawc->coldefexpr,
802 : EXPR_KIND_FROM_FUNCTION),
803 : typid, typmod,
804 : constructName);
805 56 : assign_expr_collations(pstate, coldefexpr);
806 : }
807 : else
808 694 : coldefexpr = NULL;
809 :
810 750 : tf->colexprs = lappend(tf->colexprs, colexpr);
811 750 : tf->coldefexprs = lappend(tf->coldefexprs, coldefexpr);
812 :
813 750 : if (rawc->is_not_null)
814 56 : tf->notnulls = bms_add_member(tf->notnulls, colno);
815 :
816 : /* make sure column names are unique */
817 2534 : for (j = 0; j < colno; j++)
818 1784 : if (strcmp(names[j], rawc->colname) == 0)
819 0 : ereport(ERROR,
820 : (errcode(ERRCODE_SYNTAX_ERROR),
821 : errmsg("column name \"%s\" is not unique",
822 : rawc->colname),
823 : parser_errposition(pstate, rawc->location)));
824 750 : names[colno] = rawc->colname;
825 :
826 750 : colno++;
827 : }
828 220 : pfree(names);
829 :
830 : /* Namespaces, if any, also need to be transformed */
831 220 : if (rtf->namespaces != NIL)
832 : {
833 : ListCell *ns;
834 : ListCell *lc2;
835 20 : List *ns_uris = NIL;
836 20 : List *ns_names = NIL;
837 20 : bool default_ns_seen = false;
838 :
839 40 : foreach(ns, rtf->namespaces)
840 : {
841 20 : ResTarget *r = (ResTarget *) lfirst(ns);
842 : Node *ns_uri;
843 :
844 : Assert(IsA(r, ResTarget));
845 20 : ns_uri = transformExpr(pstate, r->val, EXPR_KIND_FROM_FUNCTION);
846 20 : ns_uri = coerce_to_specific_type(pstate, ns_uri,
847 : TEXTOID, constructName);
848 20 : assign_expr_collations(pstate, ns_uri);
849 20 : ns_uris = lappend(ns_uris, ns_uri);
850 :
851 : /* Verify consistency of name list: no dupes, only one DEFAULT */
852 20 : if (r->name != NULL)
853 : {
854 14 : foreach(lc2, ns_names)
855 : {
856 0 : String *ns_node = lfirst_node(String, lc2);
857 :
858 0 : if (ns_node == NULL)
859 0 : continue;
860 0 : if (strcmp(strVal(ns_node), r->name) == 0)
861 0 : ereport(ERROR,
862 : (errcode(ERRCODE_SYNTAX_ERROR),
863 : errmsg("namespace name \"%s\" is not unique",
864 : r->name),
865 : parser_errposition(pstate, r->location)));
866 : }
867 : }
868 : else
869 : {
870 6 : if (default_ns_seen)
871 0 : ereport(ERROR,
872 : (errcode(ERRCODE_SYNTAX_ERROR),
873 : errmsg("only one default namespace is allowed"),
874 : parser_errposition(pstate, r->location)));
875 6 : default_ns_seen = true;
876 : }
877 :
878 : /* We represent DEFAULT by a null pointer */
879 20 : ns_names = lappend(ns_names,
880 20 : r->name ? makeString(r->name) : NULL);
881 : }
882 :
883 20 : tf->ns_uris = ns_uris;
884 20 : tf->ns_names = ns_names;
885 : }
886 :
887 220 : tf->location = rtf->location;
888 :
889 220 : pstate->p_lateral_active = false;
890 :
891 : /*
892 : * Mark the RTE as LATERAL if the user said LATERAL explicitly, or if
893 : * there are any lateral cross-references in it.
894 : */
895 220 : is_lateral = rtf->lateral || contain_vars_of_level((Node *) tf, 0);
896 :
897 220 : return addRangeTableEntryForTableFunc(pstate,
898 : tf, rtf->alias, is_lateral, true);
899 : }
900 :
901 : /*
902 : * transformRangeTableSample --- transform a TABLESAMPLE clause
903 : *
904 : * Caller has already transformed rts->relation, we just have to validate
905 : * the remaining fields and create a TableSampleClause node.
906 : */
907 : static TableSampleClause *
908 242 : transformRangeTableSample(ParseState *pstate, RangeTableSample *rts)
909 : {
910 : TableSampleClause *tablesample;
911 : Oid handlerOid;
912 : Oid funcargtypes[1];
913 : TsmRoutine *tsm;
914 : List *fargs;
915 : ListCell *larg,
916 : *ltyp;
917 :
918 : /*
919 : * To validate the sample method name, look up the handler function, which
920 : * has the same name, one dummy INTERNAL argument, and a result type of
921 : * tsm_handler. (Note: tablesample method names are not schema-qualified
922 : * in the SQL standard; but since they are just functions to us, we allow
923 : * schema qualification to resolve any potential ambiguity.)
924 : */
925 242 : funcargtypes[0] = INTERNALOID;
926 :
927 242 : handlerOid = LookupFuncName(rts->method, 1, funcargtypes, true);
928 :
929 : /* we want error to complain about no-such-method, not no-such-function */
930 242 : if (!OidIsValid(handlerOid))
931 6 : ereport(ERROR,
932 : (errcode(ERRCODE_UNDEFINED_OBJECT),
933 : errmsg("tablesample method %s does not exist",
934 : NameListToString(rts->method)),
935 : parser_errposition(pstate, rts->location)));
936 :
937 : /* check that handler has correct return type */
938 236 : if (get_func_rettype(handlerOid) != TSM_HANDLEROID)
939 0 : ereport(ERROR,
940 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
941 : errmsg("function %s must return type %s",
942 : NameListToString(rts->method), "tsm_handler"),
943 : parser_errposition(pstate, rts->location)));
944 :
945 : /* OK, run the handler to get TsmRoutine, for argument type info */
946 236 : tsm = GetTsmRoutine(handlerOid);
947 :
948 236 : tablesample = makeNode(TableSampleClause);
949 236 : tablesample->tsmhandler = handlerOid;
950 :
951 : /* check user provided the expected number of arguments */
952 236 : if (list_length(rts->args) != list_length(tsm->parameterTypes))
953 0 : ereport(ERROR,
954 : (errcode(ERRCODE_INVALID_TABLESAMPLE_ARGUMENT),
955 : errmsg_plural("tablesample method %s requires %d argument, not %d",
956 : "tablesample method %s requires %d arguments, not %d",
957 : list_length(tsm->parameterTypes),
958 : NameListToString(rts->method),
959 : list_length(tsm->parameterTypes),
960 : list_length(rts->args)),
961 : parser_errposition(pstate, rts->location)));
962 :
963 : /*
964 : * Transform the arguments, typecasting them as needed. Note we must also
965 : * assign collations now, because assign_query_collations() doesn't
966 : * examine any substructure of RTEs.
967 : */
968 236 : fargs = NIL;
969 472 : forboth(larg, rts->args, ltyp, tsm->parameterTypes)
970 : {
971 236 : Node *arg = (Node *) lfirst(larg);
972 236 : Oid argtype = lfirst_oid(ltyp);
973 :
974 236 : arg = transformExpr(pstate, arg, EXPR_KIND_FROM_FUNCTION);
975 236 : arg = coerce_to_specific_type(pstate, arg, argtype, "TABLESAMPLE");
976 236 : assign_expr_collations(pstate, arg);
977 236 : fargs = lappend(fargs, arg);
978 : }
979 236 : tablesample->args = fargs;
980 :
981 : /* Process REPEATABLE (seed) */
982 236 : if (rts->repeatable != NULL)
983 : {
984 : Node *arg;
985 :
986 102 : if (!tsm->repeatable_across_queries)
987 4 : ereport(ERROR,
988 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
989 : errmsg("tablesample method %s does not support REPEATABLE",
990 : NameListToString(rts->method)),
991 : parser_errposition(pstate, rts->location)));
992 :
993 98 : arg = transformExpr(pstate, rts->repeatable, EXPR_KIND_FROM_FUNCTION);
994 98 : arg = coerce_to_specific_type(pstate, arg, FLOAT8OID, "REPEATABLE");
995 98 : assign_expr_collations(pstate, arg);
996 98 : tablesample->repeatable = (Expr *) arg;
997 : }
998 : else
999 134 : tablesample->repeatable = NULL;
1000 :
1001 232 : return tablesample;
1002 : }
1003 :
1004 : /*
1005 : * getNSItemForSpecialRelationTypes
1006 : *
1007 : * If given RangeVar refers to a CTE or an EphemeralNamedRelation,
1008 : * build and return an appropriate ParseNamespaceItem, otherwise return NULL
1009 : */
1010 : static ParseNamespaceItem *
1011 360430 : getNSItemForSpecialRelationTypes(ParseState *pstate, RangeVar *rv)
1012 : {
1013 : ParseNamespaceItem *nsitem;
1014 : CommonTableExpr *cte;
1015 : Index levelsup;
1016 :
1017 : /*
1018 : * if it is a qualified name, it can't be a CTE or tuplestore reference
1019 : */
1020 360430 : if (rv->schemaname)
1021 177988 : return NULL;
1022 :
1023 182442 : cte = scanNameSpaceForCTE(pstate, rv->relname, &levelsup);
1024 182442 : if (cte)
1025 5652 : nsitem = addRangeTableEntryForCTE(pstate, cte, levelsup, rv, true);
1026 176790 : else if (scanNameSpaceForENR(pstate, rv->relname))
1027 452 : nsitem = addRangeTableEntryForENR(pstate, rv, true);
1028 : else
1029 176338 : nsitem = NULL;
1030 :
1031 182430 : return nsitem;
1032 : }
1033 :
1034 : /*
1035 : * transformFromClauseItem -
1036 : * Transform a FROM-clause item, adding any required entries to the
1037 : * range table list being built in the ParseState, and return the
1038 : * transformed item ready to include in the joinlist. Also build a
1039 : * ParseNamespaceItem list describing the names exposed by this item.
1040 : * This routine can recurse to handle SQL92 JOIN expressions.
1041 : *
1042 : * The function return value is the node to add to the jointree (a
1043 : * RangeTblRef or JoinExpr). Additional output parameters are:
1044 : *
1045 : * *top_nsitem: receives the ParseNamespaceItem directly corresponding to the
1046 : * jointree item. (This is only used during internal recursion, not by
1047 : * outside callers.)
1048 : *
1049 : * *namespace: receives a List of ParseNamespaceItems for the RTEs exposed
1050 : * as table/column names by this item. (The lateral_only flags in these items
1051 : * are indeterminate and should be explicitly set by the caller before use.)
1052 : */
1053 : static Node *
1054 493570 : transformFromClauseItem(ParseState *pstate, Node *n,
1055 : ParseNamespaceItem **top_nsitem,
1056 : List **namespace)
1057 : {
1058 : /* Guard against stack overflow due to overly deep subtree */
1059 493570 : check_stack_depth();
1060 :
1061 493570 : if (IsA(n, RangeVar))
1062 : {
1063 : /* Plain relation reference, or perhaps a CTE reference */
1064 360430 : RangeVar *rv = (RangeVar *) n;
1065 : RangeTblRef *rtr;
1066 : ParseNamespaceItem *nsitem;
1067 :
1068 : /* Check if it's a CTE or tuplestore reference */
1069 360430 : nsitem = getNSItemForSpecialRelationTypes(pstate, rv);
1070 :
1071 : /* if not found above, must be a table reference */
1072 360418 : if (!nsitem)
1073 354326 : nsitem = transformTableEntry(pstate, rv);
1074 :
1075 360238 : *top_nsitem = nsitem;
1076 360238 : *namespace = list_make1(nsitem);
1077 360238 : rtr = makeNode(RangeTblRef);
1078 360238 : rtr->rtindex = nsitem->p_rtindex;
1079 360238 : return (Node *) rtr;
1080 : }
1081 133140 : else if (IsA(n, RangeSubselect))
1082 : {
1083 : /* sub-SELECT is like a plain relation */
1084 : RangeTblRef *rtr;
1085 : ParseNamespaceItem *nsitem;
1086 :
1087 14512 : nsitem = transformRangeSubselect(pstate, (RangeSubselect *) n);
1088 14398 : *top_nsitem = nsitem;
1089 14398 : *namespace = list_make1(nsitem);
1090 14398 : rtr = makeNode(RangeTblRef);
1091 14398 : rtr->rtindex = nsitem->p_rtindex;
1092 14398 : return (Node *) rtr;
1093 : }
1094 118628 : else if (IsA(n, RangeFunction))
1095 : {
1096 : /* function is like a plain relation */
1097 : RangeTblRef *rtr;
1098 : ParseNamespaceItem *nsitem;
1099 :
1100 42698 : nsitem = transformRangeFunction(pstate, (RangeFunction *) n);
1101 42476 : *top_nsitem = nsitem;
1102 42476 : *namespace = list_make1(nsitem);
1103 42476 : rtr = makeNode(RangeTblRef);
1104 42476 : rtr->rtindex = nsitem->p_rtindex;
1105 42476 : return (Node *) rtr;
1106 : }
1107 75930 : else if (IsA(n, RangeTableFunc) || IsA(n, JsonTable))
1108 : {
1109 : /* table function is like a plain relation */
1110 : RangeTblRef *rtr;
1111 : ParseNamespaceItem *nsitem;
1112 :
1113 744 : if (IsA(n, JsonTable))
1114 524 : nsitem = transformJsonTable(pstate, (JsonTable *) n);
1115 : else
1116 220 : nsitem = transformRangeTableFunc(pstate, (RangeTableFunc *) n);
1117 :
1118 654 : *top_nsitem = nsitem;
1119 654 : *namespace = list_make1(nsitem);
1120 654 : rtr = makeNode(RangeTblRef);
1121 654 : rtr->rtindex = nsitem->p_rtindex;
1122 654 : return (Node *) rtr;
1123 : }
1124 75186 : else if (IsA(n, RangeTableSample))
1125 : {
1126 : /* TABLESAMPLE clause (wrapping some other valid FROM node) */
1127 254 : RangeTableSample *rts = (RangeTableSample *) n;
1128 : Node *rel;
1129 : RangeTblEntry *rte;
1130 :
1131 : /* Recursively transform the contained relation */
1132 254 : rel = transformFromClauseItem(pstate, rts->relation,
1133 : top_nsitem, namespace);
1134 254 : rte = (*top_nsitem)->p_rte;
1135 : /* We only support this on plain relations and matviews */
1136 254 : if (rte->rtekind != RTE_RELATION ||
1137 248 : (rte->relkind != RELKIND_RELATION &&
1138 24 : rte->relkind != RELKIND_MATVIEW &&
1139 24 : rte->relkind != RELKIND_PARTITIONED_TABLE))
1140 12 : ereport(ERROR,
1141 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1142 : errmsg("TABLESAMPLE clause can only be applied to tables and materialized views"),
1143 : parser_errposition(pstate, exprLocation(rts->relation))));
1144 :
1145 : /* Transform TABLESAMPLE details and attach to the RTE */
1146 242 : rte->tablesample = transformRangeTableSample(pstate, rts);
1147 232 : return rel;
1148 : }
1149 74932 : else if (IsA(n, JoinExpr))
1150 : {
1151 : /* A newfangled join expression */
1152 74932 : JoinExpr *j = (JoinExpr *) n;
1153 : ParseNamespaceItem *nsitem;
1154 : ParseNamespaceItem *l_nsitem;
1155 : ParseNamespaceItem *r_nsitem;
1156 : List *l_namespace,
1157 : *r_namespace,
1158 : *my_namespace,
1159 : *l_colnames,
1160 : *r_colnames,
1161 : *res_colnames,
1162 : *l_colnos,
1163 : *r_colnos,
1164 : *res_colvars;
1165 : ParseNamespaceColumn *l_nscolumns,
1166 : *r_nscolumns,
1167 : *res_nscolumns;
1168 : int res_colindex;
1169 : bool lateral_ok;
1170 : int sv_namespace_length;
1171 : int k;
1172 :
1173 : /*
1174 : * Recursively process the left subtree, then the right. We must do
1175 : * it in this order for correct visibility of LATERAL references.
1176 : */
1177 74932 : j->larg = transformFromClauseItem(pstate, j->larg,
1178 : &l_nsitem,
1179 : &l_namespace);
1180 :
1181 : /*
1182 : * Make the left-side RTEs available for LATERAL access within the
1183 : * right side, by temporarily adding them to the pstate's namespace
1184 : * list. Per SQL:2008, if the join type is not INNER or LEFT then the
1185 : * left-side names must still be exposed, but it's an error to
1186 : * reference them. (Stupid design, but that's what it says.) Hence,
1187 : * we always push them into the namespace, but mark them as not
1188 : * lateral_ok if the jointype is wrong.
1189 : *
1190 : * Notice that we don't require the merged namespace list to be
1191 : * conflict-free. See the comments for scanNameSpaceForRefname().
1192 : */
1193 74932 : lateral_ok = (j->jointype == JOIN_INNER || j->jointype == JOIN_LEFT);
1194 74932 : setNamespaceLateralState(l_namespace, true, lateral_ok);
1195 :
1196 74932 : sv_namespace_length = list_length(pstate->p_namespace);
1197 74932 : pstate->p_namespace = list_concat(pstate->p_namespace, l_namespace);
1198 :
1199 : /* And now we can process the RHS */
1200 74932 : j->rarg = transformFromClauseItem(pstate, j->rarg,
1201 : &r_nsitem,
1202 : &r_namespace);
1203 :
1204 : /* Remove the left-side RTEs from the namespace list again */
1205 74896 : pstate->p_namespace = list_truncate(pstate->p_namespace,
1206 : sv_namespace_length);
1207 :
1208 : /*
1209 : * Check for conflicting refnames in left and right subtrees. Must do
1210 : * this because higher levels will assume I hand back a self-
1211 : * consistent namespace list.
1212 : */
1213 74896 : checkNameSpaceConflicts(pstate, l_namespace, r_namespace);
1214 :
1215 : /*
1216 : * Generate combined namespace info for possible use below.
1217 : */
1218 74896 : my_namespace = list_concat(l_namespace, r_namespace);
1219 :
1220 : /*
1221 : * We'll work from the nscolumns data and eref alias column names for
1222 : * each of the input nsitems. Note that these include dropped
1223 : * columns, which is helpful because we can keep track of physical
1224 : * input column numbers more easily.
1225 : */
1226 74896 : l_nscolumns = l_nsitem->p_nscolumns;
1227 74896 : l_colnames = l_nsitem->p_names->colnames;
1228 74896 : r_nscolumns = r_nsitem->p_nscolumns;
1229 74896 : r_colnames = r_nsitem->p_names->colnames;
1230 :
1231 : /*
1232 : * Natural join does not explicitly specify columns; must generate
1233 : * columns to join. Need to run through the list of columns from each
1234 : * table or join result and match up the column names. Use the first
1235 : * table, and check every column in the second table for a match.
1236 : * (We'll check that the matches were unique later on.) The result of
1237 : * this step is a list of column names just like an explicitly-written
1238 : * USING list.
1239 : */
1240 74896 : if (j->isNatural)
1241 : {
1242 258 : List *rlist = NIL;
1243 : ListCell *lx,
1244 : *rx;
1245 :
1246 : Assert(j->usingClause == NIL); /* shouldn't have USING() too */
1247 :
1248 1140 : foreach(lx, l_colnames)
1249 : {
1250 882 : char *l_colname = strVal(lfirst(lx));
1251 882 : String *m_name = NULL;
1252 :
1253 882 : if (l_colname[0] == '\0')
1254 12 : continue; /* ignore dropped columns */
1255 :
1256 2412 : foreach(rx, r_colnames)
1257 : {
1258 1848 : char *r_colname = strVal(lfirst(rx));
1259 :
1260 1848 : if (strcmp(l_colname, r_colname) == 0)
1261 : {
1262 306 : m_name = makeString(l_colname);
1263 306 : break;
1264 : }
1265 : }
1266 :
1267 : /* matched a right column? then keep as join column... */
1268 870 : if (m_name != NULL)
1269 306 : rlist = lappend(rlist, m_name);
1270 : }
1271 :
1272 258 : j->usingClause = rlist;
1273 : }
1274 :
1275 : /*
1276 : * If a USING clause alias was specified, save the USING columns as
1277 : * its column list.
1278 : */
1279 74896 : if (j->join_using_alias)
1280 84 : j->join_using_alias->colnames = j->usingClause;
1281 :
1282 : /*
1283 : * Now transform the join qualifications, if any.
1284 : */
1285 74896 : l_colnos = NIL;
1286 74896 : r_colnos = NIL;
1287 74896 : res_colnames = NIL;
1288 74896 : res_colvars = NIL;
1289 :
1290 : /* this may be larger than needed, but it's not worth being exact */
1291 : res_nscolumns = (ParseNamespaceColumn *)
1292 74896 : palloc0((list_length(l_colnames) + list_length(r_colnames)) *
1293 : sizeof(ParseNamespaceColumn));
1294 74896 : res_colindex = 0;
1295 :
1296 74896 : if (j->usingClause)
1297 : {
1298 : /*
1299 : * JOIN/USING (or NATURAL JOIN, as transformed above). Transform
1300 : * the list into an explicit ON-condition.
1301 : */
1302 1458 : List *ucols = j->usingClause;
1303 1458 : List *l_usingvars = NIL;
1304 1458 : List *r_usingvars = NIL;
1305 : ListCell *ucol;
1306 :
1307 : Assert(j->quals == NULL); /* shouldn't have ON() too */
1308 :
1309 3116 : foreach(ucol, ucols)
1310 : {
1311 1658 : char *u_colname = strVal(lfirst(ucol));
1312 : ListCell *col;
1313 : int ndx;
1314 1658 : int l_index = -1;
1315 1658 : int r_index = -1;
1316 : Var *l_colvar,
1317 : *r_colvar;
1318 :
1319 : Assert(u_colname[0] != '\0');
1320 :
1321 : /* Check for USING(foo,foo) */
1322 1886 : foreach(col, res_colnames)
1323 : {
1324 228 : char *res_colname = strVal(lfirst(col));
1325 :
1326 228 : if (strcmp(res_colname, u_colname) == 0)
1327 0 : ereport(ERROR,
1328 : (errcode(ERRCODE_DUPLICATE_COLUMN),
1329 : errmsg("column name \"%s\" appears more than once in USING clause",
1330 : u_colname)));
1331 : }
1332 :
1333 : /* Find it in left input */
1334 1658 : ndx = 0;
1335 8130 : foreach(col, l_colnames)
1336 : {
1337 6472 : char *l_colname = strVal(lfirst(col));
1338 :
1339 6472 : if (strcmp(l_colname, u_colname) == 0)
1340 : {
1341 1658 : if (l_index >= 0)
1342 0 : ereport(ERROR,
1343 : (errcode(ERRCODE_AMBIGUOUS_COLUMN),
1344 : errmsg("common column name \"%s\" appears more than once in left table",
1345 : u_colname)));
1346 1658 : l_index = ndx;
1347 : }
1348 6472 : ndx++;
1349 : }
1350 1658 : if (l_index < 0)
1351 0 : ereport(ERROR,
1352 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1353 : errmsg("column \"%s\" specified in USING clause does not exist in left table",
1354 : u_colname)));
1355 1658 : l_colnos = lappend_int(l_colnos, l_index + 1);
1356 :
1357 : /* Find it in right input */
1358 1658 : ndx = 0;
1359 8030 : foreach(col, r_colnames)
1360 : {
1361 6372 : char *r_colname = strVal(lfirst(col));
1362 :
1363 6372 : if (strcmp(r_colname, u_colname) == 0)
1364 : {
1365 1658 : if (r_index >= 0)
1366 0 : ereport(ERROR,
1367 : (errcode(ERRCODE_AMBIGUOUS_COLUMN),
1368 : errmsg("common column name \"%s\" appears more than once in right table",
1369 : u_colname)));
1370 1658 : r_index = ndx;
1371 : }
1372 6372 : ndx++;
1373 : }
1374 1658 : if (r_index < 0)
1375 0 : ereport(ERROR,
1376 : (errcode(ERRCODE_UNDEFINED_COLUMN),
1377 : errmsg("column \"%s\" specified in USING clause does not exist in right table",
1378 : u_colname)));
1379 1658 : r_colnos = lappend_int(r_colnos, r_index + 1);
1380 :
1381 : /* Build Vars to use in the generated JOIN ON clause */
1382 1658 : l_colvar = buildVarFromNSColumn(pstate, l_nscolumns + l_index);
1383 1658 : l_usingvars = lappend(l_usingvars, l_colvar);
1384 1658 : r_colvar = buildVarFromNSColumn(pstate, r_nscolumns + r_index);
1385 1658 : r_usingvars = lappend(r_usingvars, r_colvar);
1386 :
1387 : /*
1388 : * While we're here, add column names to the res_colnames
1389 : * list. It's a bit ugly to do this here while the
1390 : * corresponding res_colvars entries are not made till later,
1391 : * but doing this later would require an additional traversal
1392 : * of the usingClause list.
1393 : */
1394 1658 : res_colnames = lappend(res_colnames, lfirst(ucol));
1395 : }
1396 :
1397 : /* Construct the generated JOIN ON clause */
1398 1458 : j->quals = transformJoinUsingClause(pstate,
1399 : l_usingvars,
1400 : r_usingvars);
1401 : }
1402 73438 : else if (j->quals)
1403 : {
1404 : /* User-written ON-condition; transform it */
1405 73140 : j->quals = transformJoinOnClause(pstate, j, my_namespace);
1406 : }
1407 : else
1408 : {
1409 : /* CROSS JOIN: no quals */
1410 : }
1411 :
1412 : /*
1413 : * If this is an outer join, now mark the appropriate child RTEs as
1414 : * being nulled by this join. We have finished processing the child
1415 : * join expressions as well as the current join's quals, which deal in
1416 : * non-nulled input columns. All future references to those RTEs will
1417 : * see possibly-nulled values, and we should mark generated Vars to
1418 : * account for that. In particular, the join alias Vars that we're
1419 : * about to build should reflect the nulling effects of this join.
1420 : *
1421 : * A difficulty with doing this is that we need the join's RT index,
1422 : * which we don't officially have yet. However, no other RTE can get
1423 : * made between here and the addRangeTableEntryForJoin call, so we can
1424 : * predict what the assignment will be. (Alternatively, we could call
1425 : * addRangeTableEntryForJoin before we have all the data computed, but
1426 : * this seems less ugly.)
1427 : */
1428 74878 : j->rtindex = list_length(pstate->p_rtable) + 1;
1429 :
1430 74878 : switch (j->jointype)
1431 : {
1432 35750 : case JOIN_INNER:
1433 35750 : break;
1434 37784 : case JOIN_LEFT:
1435 37784 : markRelsAsNulledBy(pstate, j->rarg, j->rtindex);
1436 37784 : break;
1437 1006 : case JOIN_FULL:
1438 1006 : markRelsAsNulledBy(pstate, j->larg, j->rtindex);
1439 1006 : markRelsAsNulledBy(pstate, j->rarg, j->rtindex);
1440 1006 : break;
1441 338 : case JOIN_RIGHT:
1442 338 : markRelsAsNulledBy(pstate, j->larg, j->rtindex);
1443 338 : break;
1444 0 : default:
1445 : /* shouldn't see any other types here */
1446 0 : elog(ERROR, "unrecognized join type: %d",
1447 : (int) j->jointype);
1448 : break;
1449 : }
1450 :
1451 : /*
1452 : * Now we can construct join alias expressions for the USING columns.
1453 : */
1454 74878 : if (j->usingClause)
1455 : {
1456 : ListCell *lc1,
1457 : *lc2;
1458 :
1459 : /* Scan the colnos lists to recover info from the previous loop */
1460 3116 : forboth(lc1, l_colnos, lc2, r_colnos)
1461 : {
1462 1658 : int l_index = lfirst_int(lc1) - 1;
1463 1658 : int r_index = lfirst_int(lc2) - 1;
1464 : Var *l_colvar,
1465 : *r_colvar;
1466 : Node *u_colvar;
1467 : ParseNamespaceColumn *res_nscolumn;
1468 :
1469 : /*
1470 : * Note we re-build these Vars: they might have different
1471 : * varnullingrels than the ones made in the previous loop.
1472 : */
1473 1658 : l_colvar = buildVarFromNSColumn(pstate, l_nscolumns + l_index);
1474 1658 : r_colvar = buildVarFromNSColumn(pstate, r_nscolumns + r_index);
1475 :
1476 : /* Construct the join alias Var for this column */
1477 1658 : u_colvar = buildMergedJoinVar(pstate,
1478 : j->jointype,
1479 : l_colvar,
1480 : r_colvar);
1481 1658 : res_colvars = lappend(res_colvars, u_colvar);
1482 :
1483 : /* Construct column's res_nscolumns[] entry */
1484 1658 : res_nscolumn = res_nscolumns + res_colindex;
1485 1658 : res_colindex++;
1486 1658 : if (u_colvar == (Node *) l_colvar)
1487 : {
1488 : /* Merged column is equivalent to left input */
1489 1214 : *res_nscolumn = l_nscolumns[l_index];
1490 : }
1491 444 : else if (u_colvar == (Node *) r_colvar)
1492 : {
1493 : /* Merged column is equivalent to right input */
1494 42 : *res_nscolumn = r_nscolumns[r_index];
1495 : }
1496 : else
1497 : {
1498 : /*
1499 : * Merged column is not semantically equivalent to either
1500 : * input, so it needs to be referenced as the join output
1501 : * column.
1502 : */
1503 402 : res_nscolumn->p_varno = j->rtindex;
1504 402 : res_nscolumn->p_varattno = res_colindex;
1505 402 : res_nscolumn->p_vartype = exprType(u_colvar);
1506 402 : res_nscolumn->p_vartypmod = exprTypmod(u_colvar);
1507 402 : res_nscolumn->p_varcollid = exprCollation(u_colvar);
1508 402 : res_nscolumn->p_varnosyn = j->rtindex;
1509 402 : res_nscolumn->p_varattnosyn = res_colindex;
1510 : }
1511 : }
1512 : }
1513 :
1514 : /* Add remaining columns from each side to the output columns */
1515 74878 : res_colindex +=
1516 74878 : extractRemainingColumns(pstate,
1517 : l_nscolumns, l_colnames, &l_colnos,
1518 : &res_colnames, &res_colvars,
1519 74878 : res_nscolumns + res_colindex);
1520 74878 : res_colindex +=
1521 74878 : extractRemainingColumns(pstate,
1522 : r_nscolumns, r_colnames, &r_colnos,
1523 : &res_colnames, &res_colvars,
1524 74878 : res_nscolumns + res_colindex);
1525 :
1526 : /* If join has an alias, it syntactically hides all inputs */
1527 74878 : if (j->alias)
1528 : {
1529 978 : for (k = 0; k < res_colindex; k++)
1530 : {
1531 804 : ParseNamespaceColumn *nscol = res_nscolumns + k;
1532 :
1533 804 : nscol->p_varnosyn = j->rtindex;
1534 804 : nscol->p_varattnosyn = k + 1;
1535 : }
1536 : }
1537 :
1538 : /*
1539 : * Now build an RTE and nsitem for the result of the join.
1540 : */
1541 74878 : nsitem = addRangeTableEntryForJoin(pstate,
1542 : res_colnames,
1543 : res_nscolumns,
1544 : j->jointype,
1545 74878 : list_length(j->usingClause),
1546 : res_colvars,
1547 : l_colnos,
1548 : r_colnos,
1549 : j->join_using_alias,
1550 : j->alias,
1551 : true);
1552 :
1553 : /* Verify that we correctly predicted the join's RT index */
1554 : Assert(j->rtindex == nsitem->p_rtindex);
1555 : /* Cross-check number of columns, too */
1556 : Assert(res_colindex == list_length(nsitem->p_names->colnames));
1557 :
1558 : /*
1559 : * Save a link to the JoinExpr in the proper element of p_joinexprs.
1560 : * Since we maintain that list lazily, it may be necessary to fill in
1561 : * empty entries before we can add the JoinExpr in the right place.
1562 : */
1563 195806 : for (k = list_length(pstate->p_joinexprs) + 1; k < j->rtindex; k++)
1564 120934 : pstate->p_joinexprs = lappend(pstate->p_joinexprs, NULL);
1565 74872 : pstate->p_joinexprs = lappend(pstate->p_joinexprs, j);
1566 : Assert(list_length(pstate->p_joinexprs) == j->rtindex);
1567 :
1568 : /*
1569 : * If the join has a USING alias, build a ParseNamespaceItem for that
1570 : * and add it to the list of nsitems in the join's input.
1571 : */
1572 74872 : if (j->join_using_alias)
1573 : {
1574 : ParseNamespaceItem *jnsitem;
1575 :
1576 84 : jnsitem = (ParseNamespaceItem *) palloc(sizeof(ParseNamespaceItem));
1577 84 : jnsitem->p_names = j->join_using_alias;
1578 84 : jnsitem->p_rte = nsitem->p_rte;
1579 84 : jnsitem->p_rtindex = nsitem->p_rtindex;
1580 84 : jnsitem->p_perminfo = NULL;
1581 : /* no need to copy the first N columns, just use res_nscolumns */
1582 84 : jnsitem->p_nscolumns = res_nscolumns;
1583 : /* set default visibility flags; might get changed later */
1584 84 : jnsitem->p_rel_visible = true;
1585 84 : jnsitem->p_cols_visible = true;
1586 84 : jnsitem->p_lateral_only = false;
1587 84 : jnsitem->p_lateral_ok = true;
1588 : /* Per SQL, we must check for alias conflicts */
1589 84 : checkNameSpaceConflicts(pstate, list_make1(jnsitem), my_namespace);
1590 78 : my_namespace = lappend(my_namespace, jnsitem);
1591 : }
1592 :
1593 : /*
1594 : * Prepare returned namespace list. If the JOIN has an alias then it
1595 : * hides the contained RTEs completely; otherwise, the contained RTEs
1596 : * are still visible as table names, but are not visible for
1597 : * unqualified column-name access.
1598 : *
1599 : * Note: if there are nested alias-less JOINs, the lower-level ones
1600 : * will remain in the list although they have neither p_rel_visible
1601 : * nor p_cols_visible set. We could delete such list items, but it's
1602 : * unclear that it's worth expending cycles to do so.
1603 : */
1604 74866 : if (j->alias != NULL)
1605 168 : my_namespace = NIL;
1606 : else
1607 74698 : setNamespaceColumnVisibility(my_namespace, false);
1608 :
1609 : /*
1610 : * The join RTE itself is always made visible for unqualified column
1611 : * names. It's visible as a relation name only if it has an alias.
1612 : */
1613 74866 : nsitem->p_rel_visible = (j->alias != NULL);
1614 74866 : nsitem->p_cols_visible = true;
1615 74866 : nsitem->p_lateral_only = false;
1616 74866 : nsitem->p_lateral_ok = true;
1617 :
1618 74866 : *top_nsitem = nsitem;
1619 74866 : *namespace = lappend(my_namespace, nsitem);
1620 :
1621 74866 : return (Node *) j;
1622 : }
1623 : else
1624 0 : elog(ERROR, "unrecognized node type: %d", (int) nodeTag(n));
1625 : return NULL; /* can't get here, keep compiler quiet */
1626 : }
1627 :
1628 : /*
1629 : * buildVarFromNSColumn -
1630 : * build a Var node using ParseNamespaceColumn data
1631 : *
1632 : * This is used to construct joinaliasvars entries.
1633 : * We can assume varlevelsup should be 0, and no location is specified.
1634 : * Note also that no column SELECT privilege is requested here; that would
1635 : * happen only if the column is actually referenced in the query.
1636 : */
1637 : static Var *
1638 2822820 : buildVarFromNSColumn(ParseState *pstate, ParseNamespaceColumn *nscol)
1639 : {
1640 : Var *var;
1641 :
1642 : Assert(nscol->p_varno > 0); /* i.e., not deleted column */
1643 2822820 : var = makeVar(nscol->p_varno,
1644 2822820 : nscol->p_varattno,
1645 : nscol->p_vartype,
1646 : nscol->p_vartypmod,
1647 : nscol->p_varcollid,
1648 : 0);
1649 : /* makeVar doesn't offer parameters for these, so set by hand: */
1650 2822820 : var->varnosyn = nscol->p_varnosyn;
1651 2822820 : var->varattnosyn = nscol->p_varattnosyn;
1652 :
1653 : /* ... and update varnullingrels */
1654 2822820 : markNullableIfNeeded(pstate, var);
1655 :
1656 2822820 : return var;
1657 : }
1658 :
1659 : /*
1660 : * buildMergedJoinVar -
1661 : * generate a suitable replacement expression for a merged join column
1662 : */
1663 : static Node *
1664 1658 : buildMergedJoinVar(ParseState *pstate, JoinType jointype,
1665 : Var *l_colvar, Var *r_colvar)
1666 : {
1667 : Oid outcoltype;
1668 : int32 outcoltypmod;
1669 : Node *l_node,
1670 : *r_node,
1671 : *res_node;
1672 :
1673 1658 : outcoltype = select_common_type(pstate,
1674 1658 : list_make2(l_colvar, r_colvar),
1675 : "JOIN/USING",
1676 : NULL);
1677 1658 : outcoltypmod = select_common_typmod(pstate,
1678 1658 : list_make2(l_colvar, r_colvar),
1679 : outcoltype);
1680 :
1681 : /*
1682 : * Insert coercion functions if needed. Note that a difference in typmod
1683 : * can only happen if input has typmod but outcoltypmod is -1. In that
1684 : * case we insert a RelabelType to clearly mark that result's typmod is
1685 : * not same as input. We never need coerce_type_typmod.
1686 : */
1687 1658 : if (l_colvar->vartype != outcoltype)
1688 84 : l_node = coerce_type(pstate, (Node *) l_colvar, l_colvar->vartype,
1689 : outcoltype, outcoltypmod,
1690 : COERCION_IMPLICIT, COERCE_IMPLICIT_CAST, -1);
1691 1574 : else if (l_colvar->vartypmod != outcoltypmod)
1692 0 : l_node = (Node *) makeRelabelType((Expr *) l_colvar,
1693 : outcoltype, outcoltypmod,
1694 : InvalidOid, /* fixed below */
1695 : COERCE_IMPLICIT_CAST);
1696 : else
1697 1574 : l_node = (Node *) l_colvar;
1698 :
1699 1658 : if (r_colvar->vartype != outcoltype)
1700 30 : r_node = coerce_type(pstate, (Node *) r_colvar, r_colvar->vartype,
1701 : outcoltype, outcoltypmod,
1702 : COERCION_IMPLICIT, COERCE_IMPLICIT_CAST, -1);
1703 1628 : else if (r_colvar->vartypmod != outcoltypmod)
1704 0 : r_node = (Node *) makeRelabelType((Expr *) r_colvar,
1705 : outcoltype, outcoltypmod,
1706 : InvalidOid, /* fixed below */
1707 : COERCE_IMPLICIT_CAST);
1708 : else
1709 1628 : r_node = (Node *) r_colvar;
1710 :
1711 : /*
1712 : * Choose what to emit
1713 : */
1714 1658 : switch (jointype)
1715 : {
1716 1088 : case JOIN_INNER:
1717 :
1718 : /*
1719 : * We can use either var; prefer non-coerced one if available.
1720 : */
1721 1088 : if (IsA(l_node, Var))
1722 1058 : res_node = l_node;
1723 30 : else if (IsA(r_node, Var))
1724 30 : res_node = r_node;
1725 : else
1726 0 : res_node = l_node;
1727 1088 : break;
1728 210 : case JOIN_LEFT:
1729 : /* Always use left var */
1730 210 : res_node = l_node;
1731 210 : break;
1732 12 : case JOIN_RIGHT:
1733 : /* Always use right var */
1734 12 : res_node = r_node;
1735 12 : break;
1736 348 : case JOIN_FULL:
1737 : {
1738 : /*
1739 : * Here we must build a COALESCE expression to ensure that the
1740 : * join output is non-null if either input is.
1741 : */
1742 348 : CoalesceExpr *c = makeNode(CoalesceExpr);
1743 :
1744 348 : c->coalescetype = outcoltype;
1745 : /* coalescecollid will get set below */
1746 348 : c->args = list_make2(l_node, r_node);
1747 348 : c->location = -1;
1748 348 : res_node = (Node *) c;
1749 348 : break;
1750 : }
1751 0 : default:
1752 0 : elog(ERROR, "unrecognized join type: %d", (int) jointype);
1753 : res_node = NULL; /* keep compiler quiet */
1754 : break;
1755 : }
1756 :
1757 : /*
1758 : * Apply assign_expr_collations to fix up the collation info in the
1759 : * coercion and CoalesceExpr nodes, if we made any. This must be done now
1760 : * so that the join node's alias vars show correct collation info.
1761 : */
1762 1658 : assign_expr_collations(pstate, res_node);
1763 :
1764 1658 : return res_node;
1765 : }
1766 :
1767 : /*
1768 : * markRelsAsNulledBy -
1769 : * Mark the given jointree node and its children as nulled by join jindex
1770 : */
1771 : static void
1772 43234 : markRelsAsNulledBy(ParseState *pstate, Node *n, int jindex)
1773 : {
1774 : int varno;
1775 : ListCell *lc;
1776 :
1777 : /* Note: we can't see FromExpr here */
1778 43234 : if (IsA(n, RangeTblRef))
1779 : {
1780 41684 : varno = ((RangeTblRef *) n)->rtindex;
1781 : }
1782 1550 : else if (IsA(n, JoinExpr))
1783 : {
1784 1550 : JoinExpr *j = (JoinExpr *) n;
1785 :
1786 : /* recurse to children */
1787 1550 : markRelsAsNulledBy(pstate, j->larg, jindex);
1788 1550 : markRelsAsNulledBy(pstate, j->rarg, jindex);
1789 1550 : varno = j->rtindex;
1790 : }
1791 : else
1792 : {
1793 0 : elog(ERROR, "unrecognized node type: %d", (int) nodeTag(n));
1794 : varno = 0; /* keep compiler quiet */
1795 : }
1796 :
1797 : /*
1798 : * Now add jindex to the p_nullingrels set for relation varno. Since we
1799 : * maintain the p_nullingrels list lazily, we might need to extend it to
1800 : * make the varno'th entry exist.
1801 : */
1802 141038 : while (list_length(pstate->p_nullingrels) < varno)
1803 97804 : pstate->p_nullingrels = lappend(pstate->p_nullingrels, NULL);
1804 43234 : lc = list_nth_cell(pstate->p_nullingrels, varno - 1);
1805 43234 : lfirst(lc) = bms_add_member((Bitmapset *) lfirst(lc), jindex);
1806 43234 : }
1807 :
1808 : /*
1809 : * setNamespaceColumnVisibility -
1810 : * Convenience subroutine to update cols_visible flags in a namespace list.
1811 : */
1812 : static void
1813 74698 : setNamespaceColumnVisibility(List *namespace, bool cols_visible)
1814 : {
1815 : ListCell *lc;
1816 :
1817 317356 : foreach(lc, namespace)
1818 : {
1819 242658 : ParseNamespaceItem *nsitem = (ParseNamespaceItem *) lfirst(lc);
1820 :
1821 242658 : nsitem->p_cols_visible = cols_visible;
1822 : }
1823 74698 : }
1824 :
1825 : /*
1826 : * setNamespaceLateralState -
1827 : * Convenience subroutine to update LATERAL flags in a namespace list.
1828 : */
1829 : static void
1830 976664 : setNamespaceLateralState(List *namespace, bool lateral_only, bool lateral_ok)
1831 : {
1832 : ListCell *lc;
1833 :
1834 2383222 : foreach(lc, namespace)
1835 : {
1836 1406558 : ParseNamespaceItem *nsitem = (ParseNamespaceItem *) lfirst(lc);
1837 :
1838 1406558 : nsitem->p_lateral_only = lateral_only;
1839 1406558 : nsitem->p_lateral_ok = lateral_ok;
1840 : }
1841 976664 : }
1842 :
1843 :
1844 : /*
1845 : * transformWhereClause -
1846 : * Transform the qualification and make sure it is of type boolean.
1847 : * Used for WHERE and allied clauses.
1848 : *
1849 : * constructName does not affect the semantics, but is used in error messages
1850 : */
1851 : Node *
1852 1022390 : transformWhereClause(ParseState *pstate, Node *clause,
1853 : ParseExprKind exprKind, const char *constructName)
1854 : {
1855 : Node *qual;
1856 :
1857 1022390 : if (clause == NULL)
1858 740660 : return NULL;
1859 :
1860 281730 : qual = transformExpr(pstate, clause, exprKind);
1861 :
1862 281520 : qual = coerce_to_boolean(pstate, qual, constructName);
1863 :
1864 281514 : return qual;
1865 : }
1866 :
1867 :
1868 : /*
1869 : * transformLimitClause -
1870 : * Transform the expression and make sure it is of type bigint.
1871 : * Used for LIMIT and allied clauses.
1872 : *
1873 : * Note: as of Postgres 8.2, LIMIT expressions are expected to yield int8,
1874 : * rather than int4 as before.
1875 : *
1876 : * constructName does not affect the semantics, but is used in error messages
1877 : */
1878 : Node *
1879 952132 : transformLimitClause(ParseState *pstate, Node *clause,
1880 : ParseExprKind exprKind, const char *constructName,
1881 : LimitOption limitOption)
1882 : {
1883 : Node *qual;
1884 :
1885 952132 : if (clause == NULL)
1886 946894 : return NULL;
1887 :
1888 5238 : qual = transformExpr(pstate, clause, exprKind);
1889 :
1890 5232 : qual = coerce_to_specific_type(pstate, qual, INT8OID, constructName);
1891 :
1892 : /* LIMIT can't refer to any variables of the current query */
1893 5232 : checkExprIsVarFree(pstate, qual, constructName);
1894 :
1895 : /*
1896 : * Don't allow NULLs in FETCH FIRST .. WITH TIES. This test is ugly and
1897 : * extremely simplistic, in that you can pass a NULL anyway by hiding it
1898 : * inside an expression -- but this protects ruleutils against emitting an
1899 : * unadorned NULL that's not accepted back by the grammar.
1900 : */
1901 5232 : if (exprKind == EXPR_KIND_LIMIT && limitOption == LIMIT_OPTION_WITH_TIES &&
1902 52 : IsA(clause, A_Const) && castNode(A_Const, clause)->isnull)
1903 6 : ereport(ERROR,
1904 : (errcode(ERRCODE_INVALID_ROW_COUNT_IN_LIMIT_CLAUSE),
1905 : errmsg("row count cannot be null in FETCH FIRST ... WITH TIES clause")));
1906 :
1907 5226 : return qual;
1908 : }
1909 :
1910 : /*
1911 : * checkExprIsVarFree
1912 : * Check that given expr has no Vars of the current query level
1913 : * (aggregates and window functions should have been rejected already).
1914 : *
1915 : * This is used to check expressions that have to have a consistent value
1916 : * across all rows of the query, such as a LIMIT. Arguably it should reject
1917 : * volatile functions, too, but we don't do that --- whatever value the
1918 : * function gives on first execution is what you get.
1919 : *
1920 : * constructName does not affect the semantics, but is used in error messages
1921 : */
1922 : static void
1923 7038 : checkExprIsVarFree(ParseState *pstate, Node *n, const char *constructName)
1924 : {
1925 7038 : if (contain_vars_of_level(n, 0))
1926 : {
1927 6 : ereport(ERROR,
1928 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
1929 : /* translator: %s is name of a SQL construct, eg LIMIT */
1930 : errmsg("argument of %s must not contain variables",
1931 : constructName),
1932 : parser_errposition(pstate,
1933 : locate_var_of_level(n, 0))));
1934 : }
1935 7032 : }
1936 :
1937 :
1938 : /*
1939 : * checkTargetlistEntrySQL92 -
1940 : * Validate a targetlist entry found by findTargetlistEntrySQL92
1941 : *
1942 : * When we select a pre-existing tlist entry as a result of syntax such
1943 : * as "GROUP BY 1", we have to make sure it is acceptable for use in the
1944 : * indicated clause type; transformExpr() will have treated it as a regular
1945 : * targetlist item.
1946 : */
1947 : static void
1948 60212 : checkTargetlistEntrySQL92(ParseState *pstate, TargetEntry *tle,
1949 : ParseExprKind exprKind)
1950 : {
1951 60212 : switch (exprKind)
1952 : {
1953 794 : case EXPR_KIND_GROUP_BY:
1954 : /* reject aggregates and window functions */
1955 1454 : if (pstate->p_hasAggs &&
1956 660 : contain_aggs_of_level((Node *) tle->expr, 0))
1957 0 : ereport(ERROR,
1958 : (errcode(ERRCODE_GROUPING_ERROR),
1959 : /* translator: %s is name of a SQL construct, eg GROUP BY */
1960 : errmsg("aggregate functions are not allowed in %s",
1961 : ParseExprKindName(exprKind)),
1962 : parser_errposition(pstate,
1963 : locate_agg_of_level((Node *) tle->expr, 0))));
1964 806 : if (pstate->p_hasWindowFuncs &&
1965 12 : contain_windowfuncs((Node *) tle->expr))
1966 6 : ereport(ERROR,
1967 : (errcode(ERRCODE_WINDOWING_ERROR),
1968 : /* translator: %s is name of a SQL construct, eg GROUP BY */
1969 : errmsg("window functions are not allowed in %s",
1970 : ParseExprKindName(exprKind)),
1971 : parser_errposition(pstate,
1972 : locate_windowfunc((Node *) tle->expr))));
1973 788 : break;
1974 59274 : case EXPR_KIND_ORDER_BY:
1975 : /* no extra checks needed */
1976 59274 : break;
1977 144 : case EXPR_KIND_DISTINCT_ON:
1978 : /* no extra checks needed */
1979 144 : break;
1980 0 : default:
1981 0 : elog(ERROR, "unexpected exprKind in checkTargetlistEntrySQL92");
1982 : break;
1983 : }
1984 60206 : }
1985 :
1986 : /*
1987 : * findTargetlistEntrySQL92 -
1988 : * Returns the targetlist entry matching the given (untransformed) node.
1989 : * If no matching entry exists, one is created and appended to the target
1990 : * list as a "resjunk" node.
1991 : *
1992 : * This function supports the old SQL92 ORDER BY interpretation, where the
1993 : * expression is an output column name or number. If we fail to find a
1994 : * match of that sort, we fall through to the SQL99 rules. For historical
1995 : * reasons, Postgres also allows this interpretation for GROUP BY, though
1996 : * the standard never did. However, for GROUP BY we prefer a SQL99 match.
1997 : * This function is *not* used for WINDOW definitions.
1998 : *
1999 : * node the ORDER BY, GROUP BY, or DISTINCT ON expression to be matched
2000 : * tlist the target list (passed by reference so we can append to it)
2001 : * exprKind identifies clause type being processed
2002 : */
2003 : static TargetEntry *
2004 95956 : findTargetlistEntrySQL92(ParseState *pstate, Node *node, List **tlist,
2005 : ParseExprKind exprKind)
2006 : {
2007 : ListCell *tl;
2008 :
2009 : /*----------
2010 : * Handle two special cases as mandated by the SQL92 spec:
2011 : *
2012 : * 1. Bare ColumnName (no qualifier or subscripts)
2013 : * For a bare identifier, we search for a matching column name
2014 : * in the existing target list. Multiple matches are an error
2015 : * unless they refer to identical values; for example,
2016 : * we allow SELECT a, a FROM table ORDER BY a
2017 : * but not SELECT a AS b, b FROM table ORDER BY b
2018 : * If no match is found, we fall through and treat the identifier
2019 : * as an expression.
2020 : * For GROUP BY, it is incorrect to match the grouping item against
2021 : * targetlist entries: according to SQL92, an identifier in GROUP BY
2022 : * is a reference to a column name exposed by FROM, not to a target
2023 : * list column. However, many implementations (including pre-7.0
2024 : * PostgreSQL) accept this anyway. So for GROUP BY, we look first
2025 : * to see if the identifier matches any FROM column name, and only
2026 : * try for a targetlist name if it doesn't. This ensures that we
2027 : * adhere to the spec in the case where the name could be both.
2028 : * DISTINCT ON isn't in the standard, so we can do what we like there;
2029 : * we choose to make it work like ORDER BY, on the rather flimsy
2030 : * grounds that ordinary DISTINCT works on targetlist entries.
2031 : *
2032 : * 2. IntegerConstant
2033 : * This means to use the n'th item in the existing target list.
2034 : * Note that it would make no sense to order/group/distinct by an
2035 : * actual constant, so this does not create a conflict with SQL99.
2036 : * GROUP BY column-number is not allowed by SQL92, but since
2037 : * the standard has no other behavior defined for this syntax,
2038 : * we may as well accept this common extension.
2039 : *
2040 : * Note that pre-existing resjunk targets must not be used in either case,
2041 : * since the user didn't write them in his SELECT list.
2042 : *
2043 : * If neither special case applies, fall through to treat the item as
2044 : * an expression per SQL99.
2045 : *----------
2046 : */
2047 151022 : if (IsA(node, ColumnRef) &&
2048 55066 : list_length(((ColumnRef *) node)->fields) == 1 &&
2049 38106 : IsA(linitial(((ColumnRef *) node)->fields), String))
2050 : {
2051 38106 : char *name = strVal(linitial(((ColumnRef *) node)->fields));
2052 38106 : int location = ((ColumnRef *) node)->location;
2053 :
2054 38106 : if (exprKind == EXPR_KIND_GROUP_BY)
2055 : {
2056 : /*
2057 : * In GROUP BY, we must prefer a match against a FROM-clause
2058 : * column to one against the targetlist. Look to see if there is
2059 : * a matching column. If so, fall through to use SQL99 rules.
2060 : * NOTE: if name could refer ambiguously to more than one column
2061 : * name exposed by FROM, colNameToVar will ereport(ERROR). That's
2062 : * just what we want here.
2063 : *
2064 : * Small tweak for 7.4.3: ignore matches in upper query levels.
2065 : * This effectively changes the search order for bare names to (1)
2066 : * local FROM variables, (2) local targetlist aliases, (3) outer
2067 : * FROM variables, whereas before it was (1) (3) (2). SQL92 and
2068 : * SQL99 do not allow GROUPing BY an outer reference, so this
2069 : * breaks no cases that are legal per spec, and it seems a more
2070 : * self-consistent behavior.
2071 : */
2072 4594 : if (colNameToVar(pstate, name, true, location) != NULL)
2073 4484 : name = NULL;
2074 : }
2075 :
2076 38106 : if (name != NULL)
2077 : {
2078 33622 : TargetEntry *target_result = NULL;
2079 :
2080 186922 : foreach(tl, *tlist)
2081 : {
2082 153300 : TargetEntry *tle = (TargetEntry *) lfirst(tl);
2083 :
2084 153300 : if (!tle->resjunk &&
2085 152700 : strcmp(tle->resname, name) == 0)
2086 : {
2087 28598 : if (target_result != NULL)
2088 : {
2089 6 : if (!equal(target_result->expr, tle->expr))
2090 0 : ereport(ERROR,
2091 : (errcode(ERRCODE_AMBIGUOUS_COLUMN),
2092 :
2093 : /*------
2094 : translator: first %s is name of a SQL construct, eg ORDER BY */
2095 : errmsg("%s \"%s\" is ambiguous",
2096 : ParseExprKindName(exprKind),
2097 : name),
2098 : parser_errposition(pstate, location)));
2099 : }
2100 : else
2101 28592 : target_result = tle;
2102 : /* Stay in loop to check for ambiguity */
2103 : }
2104 : }
2105 33622 : if (target_result != NULL)
2106 : {
2107 : /* return the first match, after suitable validation */
2108 28592 : checkTargetlistEntrySQL92(pstate, target_result, exprKind);
2109 28592 : return target_result;
2110 : }
2111 : }
2112 : }
2113 67364 : if (IsA(node, A_Const))
2114 : {
2115 31626 : A_Const *aconst = castNode(A_Const, node);
2116 31626 : int targetlist_pos = 0;
2117 : int target_pos;
2118 :
2119 31626 : if (!IsA(&aconst->val, Integer))
2120 0 : ereport(ERROR,
2121 : (errcode(ERRCODE_SYNTAX_ERROR),
2122 : /* translator: %s is name of a SQL construct, eg ORDER BY */
2123 : errmsg("non-integer constant in %s",
2124 : ParseExprKindName(exprKind)),
2125 : parser_errposition(pstate, aconst->location)));
2126 :
2127 31626 : target_pos = intVal(&aconst->val);
2128 55812 : foreach(tl, *tlist)
2129 : {
2130 55806 : TargetEntry *tle = (TargetEntry *) lfirst(tl);
2131 :
2132 55806 : if (!tle->resjunk)
2133 : {
2134 55806 : if (++targetlist_pos == target_pos)
2135 : {
2136 : /* return the unique match, after suitable validation */
2137 31620 : checkTargetlistEntrySQL92(pstate, tle, exprKind);
2138 31614 : return tle;
2139 : }
2140 : }
2141 : }
2142 6 : ereport(ERROR,
2143 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
2144 : /* translator: %s is name of a SQL construct, eg ORDER BY */
2145 : errmsg("%s position %d is not in select list",
2146 : ParseExprKindName(exprKind), target_pos),
2147 : parser_errposition(pstate, aconst->location)));
2148 : }
2149 :
2150 : /*
2151 : * Otherwise, we have an expression, so process it per SQL99 rules.
2152 : */
2153 35738 : return findTargetlistEntrySQL99(pstate, node, tlist, exprKind);
2154 : }
2155 :
2156 : /*
2157 : * findTargetlistEntrySQL99 -
2158 : * Returns the targetlist entry matching the given (untransformed) node.
2159 : * If no matching entry exists, one is created and appended to the target
2160 : * list as a "resjunk" node.
2161 : *
2162 : * This function supports the SQL99 interpretation, wherein the expression
2163 : * is just an ordinary expression referencing input column names.
2164 : *
2165 : * node the ORDER BY, GROUP BY, etc expression to be matched
2166 : * tlist the target list (passed by reference so we can append to it)
2167 : * exprKind identifies clause type being processed
2168 : */
2169 : static TargetEntry *
2170 41088 : findTargetlistEntrySQL99(ParseState *pstate, Node *node, List **tlist,
2171 : ParseExprKind exprKind)
2172 : {
2173 : TargetEntry *target_result;
2174 : ListCell *tl;
2175 : Node *expr;
2176 :
2177 : /*
2178 : * Convert the untransformed node to a transformed expression, and search
2179 : * for a match in the tlist. NOTE: it doesn't really matter whether there
2180 : * is more than one match. Also, we are willing to match an existing
2181 : * resjunk target here, though the SQL92 cases above must ignore resjunk
2182 : * targets.
2183 : */
2184 41088 : expr = transformExpr(pstate, node, exprKind);
2185 :
2186 152300 : foreach(tl, *tlist)
2187 : {
2188 127812 : TargetEntry *tle = (TargetEntry *) lfirst(tl);
2189 : Node *texpr;
2190 :
2191 : /*
2192 : * Ignore any implicit cast on the existing tlist expression.
2193 : *
2194 : * This essentially allows the ORDER/GROUP/etc item to adopt the same
2195 : * datatype previously selected for a textually-equivalent tlist item.
2196 : * There can't be any implicit cast at top level in an ordinary SELECT
2197 : * tlist at this stage, but the case does arise with ORDER BY in an
2198 : * aggregate function.
2199 : */
2200 127812 : texpr = strip_implicit_coercions((Node *) tle->expr);
2201 :
2202 127812 : if (equal(expr, texpr))
2203 16546 : return tle;
2204 : }
2205 :
2206 : /*
2207 : * If no matches, construct a new target entry which is appended to the
2208 : * end of the target list. This target is given resjunk = true so that it
2209 : * will not be projected into the final tuple.
2210 : */
2211 24488 : target_result = transformTargetEntry(pstate, node, expr, exprKind,
2212 : NULL, true);
2213 :
2214 24488 : *tlist = lappend(*tlist, target_result);
2215 :
2216 24488 : return target_result;
2217 : }
2218 :
2219 : /*-------------------------------------------------------------------------
2220 : * Flatten out parenthesized sublists in grouping lists, and some cases
2221 : * of nested grouping sets.
2222 : *
2223 : * Inside a grouping set (ROLLUP, CUBE, or GROUPING SETS), we expect the
2224 : * content to be nested no more than 2 deep: i.e. ROLLUP((a,b),(c,d)) is
2225 : * ok, but ROLLUP((a,(b,c)),d) is flattened to ((a,b,c),d), which we then
2226 : * (later) normalize to ((a,b,c),(d)).
2227 : *
2228 : * CUBE or ROLLUP can be nested inside GROUPING SETS (but not the reverse),
2229 : * and we leave that alone if we find it. But if we see GROUPING SETS inside
2230 : * GROUPING SETS, we can flatten and normalize as follows:
2231 : * GROUPING SETS (a, (b,c), GROUPING SETS ((c,d),(e)), (f,g))
2232 : * becomes
2233 : * GROUPING SETS ((a), (b,c), (c,d), (e), (f,g))
2234 : *
2235 : * This is per the spec's syntax transformations, but these are the only such
2236 : * transformations we do in parse analysis, so that queries retain the
2237 : * originally specified grouping set syntax for CUBE and ROLLUP as much as
2238 : * possible when deparsed. (Full expansion of the result into a list of
2239 : * grouping sets is left to the planner.)
2240 : *
2241 : * When we're done, the resulting list should contain only these possible
2242 : * elements:
2243 : * - an expression
2244 : * - a CUBE or ROLLUP with a list of expressions nested 2 deep
2245 : * - a GROUPING SET containing any of:
2246 : * - expression lists
2247 : * - empty grouping sets
2248 : * - CUBE or ROLLUP nodes with lists nested 2 deep
2249 : * The return is a new list, but doesn't deep-copy the old nodes except for
2250 : * GroupingSet nodes.
2251 : *
2252 : * As a side effect, flag whether the list has any GroupingSet nodes.
2253 : *-------------------------------------------------------------------------
2254 : */
2255 : static Node *
2256 475256 : flatten_grouping_sets(Node *expr, bool toplevel, bool *hasGroupingSets)
2257 : {
2258 : /* just in case of pathological input */
2259 475256 : check_stack_depth();
2260 :
2261 475256 : if (expr == (Node *) NIL)
2262 459190 : return (Node *) NIL;
2263 :
2264 16066 : switch (expr->type)
2265 : {
2266 352 : case T_RowExpr:
2267 : {
2268 352 : RowExpr *r = (RowExpr *) expr;
2269 :
2270 352 : if (r->row_format == COERCE_IMPLICIT_CAST)
2271 352 : return flatten_grouping_sets((Node *) r->args,
2272 : false, NULL);
2273 : }
2274 0 : break;
2275 1248 : case T_GroupingSet:
2276 : {
2277 1248 : GroupingSet *gset = (GroupingSet *) expr;
2278 : ListCell *l2;
2279 1248 : List *result_set = NIL;
2280 :
2281 1248 : if (hasGroupingSets)
2282 924 : *hasGroupingSets = true;
2283 :
2284 : /*
2285 : * at the top level, we skip over all empty grouping sets; the
2286 : * caller can supply the canonical GROUP BY () if nothing is
2287 : * left.
2288 : */
2289 :
2290 1248 : if (toplevel && gset->kind == GROUPING_SET_EMPTY)
2291 18 : return (Node *) NIL;
2292 :
2293 3292 : foreach(l2, gset->content)
2294 : {
2295 2062 : Node *n1 = lfirst(l2);
2296 2062 : Node *n2 = flatten_grouping_sets(n1, false, NULL);
2297 :
2298 2062 : if (IsA(n1, GroupingSet) &&
2299 324 : ((GroupingSet *) n1)->kind == GROUPING_SET_SETS)
2300 96 : result_set = list_concat(result_set, (List *) n2);
2301 : else
2302 1966 : result_set = lappend(result_set, n2);
2303 : }
2304 :
2305 : /*
2306 : * At top level, keep the grouping set node; but if we're in a
2307 : * nested grouping set, then we need to concat the flattened
2308 : * result into the outer list if it's simply nested.
2309 : */
2310 :
2311 1230 : if (toplevel || (gset->kind != GROUPING_SET_SETS))
2312 : {
2313 1134 : return (Node *) makeGroupingSet(gset->kind, result_set, gset->location);
2314 : }
2315 : else
2316 96 : return (Node *) result_set;
2317 : }
2318 5534 : case T_List:
2319 : {
2320 5534 : List *result = NIL;
2321 : ListCell *l;
2322 :
2323 14004 : foreach(l, (List *) expr)
2324 : {
2325 8470 : Node *n = flatten_grouping_sets(lfirst(l), toplevel, hasGroupingSets);
2326 :
2327 8470 : if (n != (Node *) NIL)
2328 : {
2329 8452 : if (IsA(n, List))
2330 46 : result = list_concat(result, (List *) n);
2331 : else
2332 8406 : result = lappend(result, n);
2333 : }
2334 : }
2335 :
2336 5534 : return (Node *) result;
2337 : }
2338 8932 : default:
2339 8932 : break;
2340 : }
2341 :
2342 8932 : return expr;
2343 : }
2344 :
2345 : /*
2346 : * Transform a single expression within a GROUP BY clause or grouping set.
2347 : *
2348 : * The expression is added to the targetlist if not already present, and to the
2349 : * flatresult list (which will become the groupClause) if not already present
2350 : * there. The sortClause is consulted for operator and sort order hints.
2351 : *
2352 : * Returns the ressortgroupref of the expression.
2353 : *
2354 : * flatresult reference to flat list of SortGroupClause nodes
2355 : * seen_local bitmapset of sortgrouprefs already seen at the local level
2356 : * pstate ParseState
2357 : * gexpr node to transform
2358 : * targetlist reference to TargetEntry list
2359 : * sortClause ORDER BY clause (SortGroupClause nodes)
2360 : * exprKind expression kind
2361 : * useSQL99 SQL99 rather than SQL92 syntax
2362 : * toplevel false if within any grouping set
2363 : */
2364 : static Index
2365 8932 : transformGroupClauseExpr(List **flatresult, Bitmapset *seen_local,
2366 : ParseState *pstate, Node *gexpr,
2367 : List **targetlist, List *sortClause,
2368 : ParseExprKind exprKind, bool useSQL99, bool toplevel)
2369 : {
2370 : TargetEntry *tle;
2371 8932 : bool found = false;
2372 :
2373 8932 : if (useSQL99)
2374 1148 : tle = findTargetlistEntrySQL99(pstate, gexpr,
2375 : targetlist, exprKind);
2376 : else
2377 7784 : tle = findTargetlistEntrySQL92(pstate, gexpr,
2378 : targetlist, exprKind);
2379 :
2380 8908 : if (tle->ressortgroupref > 0)
2381 : {
2382 : ListCell *sl;
2383 :
2384 : /*
2385 : * Eliminate duplicates (GROUP BY x, x) but only at local level.
2386 : * (Duplicates in grouping sets can affect the number of returned
2387 : * rows, so can't be dropped indiscriminately.)
2388 : *
2389 : * Since we don't care about anything except the sortgroupref, we can
2390 : * use a bitmapset rather than scanning lists.
2391 : */
2392 2640 : if (bms_is_member(tle->ressortgroupref, seen_local))
2393 24 : return 0;
2394 :
2395 : /*
2396 : * If we're already in the flat clause list, we don't need to consider
2397 : * adding ourselves again.
2398 : */
2399 2616 : found = targetIsInSortList(tle, InvalidOid, *flatresult);
2400 2616 : if (found)
2401 262 : return tle->ressortgroupref;
2402 :
2403 : /*
2404 : * If the GROUP BY tlist entry also appears in ORDER BY, copy operator
2405 : * info from the (first) matching ORDER BY item. This means that if
2406 : * you write something like "GROUP BY foo ORDER BY foo USING <<<", the
2407 : * GROUP BY operation silently takes on the equality semantics implied
2408 : * by the ORDER BY. There are two reasons to do this: it improves the
2409 : * odds that we can implement both GROUP BY and ORDER BY with a single
2410 : * sort step, and it allows the user to choose the equality semantics
2411 : * used by GROUP BY, should she be working with a datatype that has
2412 : * more than one equality operator.
2413 : *
2414 : * If we're in a grouping set, though, we force our requested ordering
2415 : * to be NULLS LAST, because if we have any hope of using a sorted agg
2416 : * for the job, we're going to be tacking on generated NULL values
2417 : * after the corresponding groups. If the user demands nulls first,
2418 : * another sort step is going to be inevitable, but that's the
2419 : * planner's problem.
2420 : */
2421 :
2422 3288 : foreach(sl, sortClause)
2423 : {
2424 3098 : SortGroupClause *sc = (SortGroupClause *) lfirst(sl);
2425 :
2426 3098 : if (sc->tleSortGroupRef == tle->ressortgroupref)
2427 : {
2428 2164 : SortGroupClause *grpc = copyObject(sc);
2429 :
2430 2164 : if (!toplevel)
2431 592 : grpc->nulls_first = false;
2432 2164 : *flatresult = lappend(*flatresult, grpc);
2433 2164 : found = true;
2434 2164 : break;
2435 : }
2436 : }
2437 : }
2438 :
2439 : /*
2440 : * If no match in ORDER BY, just add it to the result using default
2441 : * sort/group semantics.
2442 : */
2443 8622 : if (!found)
2444 6458 : *flatresult = addTargetToGroupList(pstate, tle,
2445 : *flatresult, *targetlist,
2446 : exprLocation(gexpr));
2447 :
2448 : /*
2449 : * _something_ must have assigned us a sortgroupref by now...
2450 : */
2451 :
2452 8622 : return tle->ressortgroupref;
2453 : }
2454 :
2455 : /*
2456 : * Transform a list of expressions within a GROUP BY clause or grouping set.
2457 : *
2458 : * The list of expressions belongs to a single clause within which duplicates
2459 : * can be safely eliminated.
2460 : *
2461 : * Returns an integer list of ressortgroupref values.
2462 : *
2463 : * flatresult reference to flat list of SortGroupClause nodes
2464 : * pstate ParseState
2465 : * list nodes to transform
2466 : * targetlist reference to TargetEntry list
2467 : * sortClause ORDER BY clause (SortGroupClause nodes)
2468 : * exprKind expression kind
2469 : * useSQL99 SQL99 rather than SQL92 syntax
2470 : * toplevel false if within any grouping set
2471 : */
2472 : static List *
2473 306 : transformGroupClauseList(List **flatresult,
2474 : ParseState *pstate, List *list,
2475 : List **targetlist, List *sortClause,
2476 : ParseExprKind exprKind, bool useSQL99, bool toplevel)
2477 : {
2478 306 : Bitmapset *seen_local = NULL;
2479 306 : List *result = NIL;
2480 : ListCell *gl;
2481 :
2482 942 : foreach(gl, list)
2483 : {
2484 636 : Node *gexpr = (Node *) lfirst(gl);
2485 :
2486 636 : Index ref = transformGroupClauseExpr(flatresult,
2487 : seen_local,
2488 : pstate,
2489 : gexpr,
2490 : targetlist,
2491 : sortClause,
2492 : exprKind,
2493 : useSQL99,
2494 : toplevel);
2495 :
2496 636 : if (ref > 0)
2497 : {
2498 624 : seen_local = bms_add_member(seen_local, ref);
2499 624 : result = lappend_int(result, ref);
2500 : }
2501 : }
2502 :
2503 306 : return result;
2504 : }
2505 :
2506 : /*
2507 : * Transform a grouping set and (recursively) its content.
2508 : *
2509 : * The grouping set might be a GROUPING SETS node with other grouping sets
2510 : * inside it, but SETS within SETS have already been flattened out before
2511 : * reaching here.
2512 : *
2513 : * Returns the transformed node, which now contains SIMPLE nodes with lists
2514 : * of ressortgrouprefs rather than expressions.
2515 : *
2516 : * flatresult reference to flat list of SortGroupClause nodes
2517 : * pstate ParseState
2518 : * gset grouping set to transform
2519 : * targetlist reference to TargetEntry list
2520 : * sortClause ORDER BY clause (SortGroupClause nodes)
2521 : * exprKind expression kind
2522 : * useSQL99 SQL99 rather than SQL92 syntax
2523 : * toplevel false if within any grouping set
2524 : */
2525 : static Node *
2526 1134 : transformGroupingSet(List **flatresult,
2527 : ParseState *pstate, GroupingSet *gset,
2528 : List **targetlist, List *sortClause,
2529 : ParseExprKind exprKind, bool useSQL99, bool toplevel)
2530 : {
2531 : ListCell *gl;
2532 1134 : List *content = NIL;
2533 :
2534 : Assert(toplevel || gset->kind != GROUPING_SET_SETS);
2535 :
2536 3100 : foreach(gl, gset->content)
2537 : {
2538 1966 : Node *n = lfirst(gl);
2539 :
2540 1966 : if (IsA(n, List))
2541 : {
2542 306 : List *l = transformGroupClauseList(flatresult,
2543 : pstate, (List *) n,
2544 : targetlist, sortClause,
2545 : exprKind, useSQL99, false);
2546 :
2547 306 : content = lappend(content, makeGroupingSet(GROUPING_SET_SIMPLE,
2548 : l,
2549 : exprLocation(n)));
2550 : }
2551 1660 : else if (IsA(n, GroupingSet))
2552 : {
2553 228 : GroupingSet *gset2 = (GroupingSet *) lfirst(gl);
2554 :
2555 228 : content = lappend(content, transformGroupingSet(flatresult,
2556 : pstate, gset2,
2557 : targetlist, sortClause,
2558 : exprKind, useSQL99, false));
2559 : }
2560 : else
2561 : {
2562 1432 : Index ref = transformGroupClauseExpr(flatresult,
2563 : NULL,
2564 : pstate,
2565 : n,
2566 : targetlist,
2567 : sortClause,
2568 : exprKind,
2569 : useSQL99,
2570 : false);
2571 :
2572 2864 : content = lappend(content, makeGroupingSet(GROUPING_SET_SIMPLE,
2573 1432 : list_make1_int(ref),
2574 : exprLocation(n)));
2575 : }
2576 : }
2577 :
2578 : /* Arbitrarily cap the size of CUBE, which has exponential growth */
2579 1134 : if (gset->kind == GROUPING_SET_CUBE)
2580 : {
2581 184 : if (list_length(content) > 12)
2582 0 : ereport(ERROR,
2583 : (errcode(ERRCODE_TOO_MANY_COLUMNS),
2584 : errmsg("CUBE is limited to 12 elements"),
2585 : parser_errposition(pstate, gset->location)));
2586 : }
2587 :
2588 1134 : return (Node *) makeGroupingSet(gset->kind, content, gset->location);
2589 : }
2590 :
2591 :
2592 : /*
2593 : * transformGroupClause -
2594 : * transform a GROUP BY clause
2595 : *
2596 : * GROUP BY items will be added to the targetlist (as resjunk columns)
2597 : * if not already present, so the targetlist must be passed by reference.
2598 : *
2599 : * This is also used for window PARTITION BY clauses (which act almost the
2600 : * same, but are always interpreted per SQL99 rules).
2601 : *
2602 : * Grouping sets make this a lot more complex than it was. Our goal here is
2603 : * twofold: we make a flat list of SortGroupClause nodes referencing each
2604 : * distinct expression used for grouping, with those expressions added to the
2605 : * targetlist if needed. At the same time, we build the groupingSets tree,
2606 : * which stores only ressortgrouprefs as integer lists inside GroupingSet nodes
2607 : * (possibly nested, but limited in depth: a GROUPING_SET_SETS node can contain
2608 : * nested SIMPLE, CUBE or ROLLUP nodes, but not more sets - we flatten that
2609 : * out; while CUBE and ROLLUP can contain only SIMPLE nodes).
2610 : *
2611 : * We skip much of the hard work if there are no grouping sets.
2612 : *
2613 : * One subtlety is that the groupClause list can end up empty while the
2614 : * groupingSets list is not; this happens if there are only empty grouping
2615 : * sets, or an explicit GROUP BY (). This has the same effect as specifying
2616 : * aggregates or a HAVING clause with no GROUP BY; the output is one row per
2617 : * grouping set even if the input is empty.
2618 : *
2619 : * Returns the transformed (flat) groupClause.
2620 : *
2621 : * pstate ParseState
2622 : * grouplist clause to transform
2623 : * groupingSets reference to list to contain the grouping set tree
2624 : * targetlist reference to TargetEntry list
2625 : * sortClause ORDER BY clause (SortGroupClause nodes)
2626 : * exprKind expression kind
2627 : * useSQL99 SQL99 rather than SQL92 syntax
2628 : */
2629 : List *
2630 464372 : transformGroupClause(ParseState *pstate, List *grouplist, List **groupingSets,
2631 : List **targetlist, List *sortClause,
2632 : ParseExprKind exprKind, bool useSQL99)
2633 : {
2634 464372 : List *result = NIL;
2635 : List *flat_grouplist;
2636 464372 : List *gsets = NIL;
2637 : ListCell *gl;
2638 464372 : bool hasGroupingSets = false;
2639 464372 : Bitmapset *seen_local = NULL;
2640 :
2641 : /*
2642 : * Recursively flatten implicit RowExprs. (Technically this is only needed
2643 : * for GROUP BY, per the syntax rules for grouping sets, but we do it
2644 : * anyway.)
2645 : */
2646 464372 : flat_grouplist = (List *) flatten_grouping_sets((Node *) grouplist,
2647 : true,
2648 : &hasGroupingSets);
2649 :
2650 : /*
2651 : * If the list is now empty, but hasGroupingSets is true, it's because we
2652 : * elided redundant empty grouping sets. Restore a single empty grouping
2653 : * set to leave a canonical form: GROUP BY ()
2654 : */
2655 :
2656 464372 : if (flat_grouplist == NIL && hasGroupingSets)
2657 : {
2658 18 : flat_grouplist = list_make1(makeGroupingSet(GROUPING_SET_EMPTY,
2659 : NIL,
2660 : exprLocation((Node *) grouplist)));
2661 : }
2662 :
2663 472136 : foreach(gl, flat_grouplist)
2664 : {
2665 7788 : Node *gexpr = (Node *) lfirst(gl);
2666 :
2667 7788 : if (IsA(gexpr, GroupingSet))
2668 : {
2669 924 : GroupingSet *gset = (GroupingSet *) gexpr;
2670 :
2671 924 : switch (gset->kind)
2672 : {
2673 18 : case GROUPING_SET_EMPTY:
2674 18 : gsets = lappend(gsets, gset);
2675 18 : break;
2676 0 : case GROUPING_SET_SIMPLE:
2677 : /* can't happen */
2678 : Assert(false);
2679 0 : break;
2680 906 : case GROUPING_SET_SETS:
2681 : case GROUPING_SET_CUBE:
2682 : case GROUPING_SET_ROLLUP:
2683 906 : gsets = lappend(gsets,
2684 906 : transformGroupingSet(&result,
2685 : pstate, gset,
2686 : targetlist, sortClause,
2687 : exprKind, useSQL99, true));
2688 906 : break;
2689 : }
2690 924 : }
2691 : else
2692 : {
2693 6864 : Index ref = transformGroupClauseExpr(&result, seen_local,
2694 : pstate, gexpr,
2695 : targetlist, sortClause,
2696 : exprKind, useSQL99, true);
2697 :
2698 6840 : if (ref > 0)
2699 : {
2700 6828 : seen_local = bms_add_member(seen_local, ref);
2701 6828 : if (hasGroupingSets)
2702 36 : gsets = lappend(gsets,
2703 72 : makeGroupingSet(GROUPING_SET_SIMPLE,
2704 36 : list_make1_int(ref),
2705 : exprLocation(gexpr)));
2706 : }
2707 : }
2708 : }
2709 :
2710 : /* parser should prevent this */
2711 : Assert(gsets == NIL || groupingSets != NULL);
2712 :
2713 464348 : if (groupingSets)
2714 461638 : *groupingSets = gsets;
2715 :
2716 464348 : return result;
2717 : }
2718 :
2719 : /*
2720 : * transformSortClause -
2721 : * transform an ORDER BY clause
2722 : *
2723 : * ORDER BY items will be added to the targetlist (as resjunk columns)
2724 : * if not already present, so the targetlist must be passed by reference.
2725 : *
2726 : * This is also used for window and aggregate ORDER BY clauses (which act
2727 : * almost the same, but are always interpreted per SQL99 rules).
2728 : */
2729 : List *
2730 522868 : transformSortClause(ParseState *pstate,
2731 : List *orderlist,
2732 : List **targetlist,
2733 : ParseExprKind exprKind,
2734 : bool useSQL99)
2735 : {
2736 522868 : List *sortlist = NIL;
2737 : ListCell *olitem;
2738 :
2739 614906 : foreach(olitem, orderlist)
2740 : {
2741 92080 : SortBy *sortby = (SortBy *) lfirst(olitem);
2742 : TargetEntry *tle;
2743 :
2744 92080 : if (useSQL99)
2745 4202 : tle = findTargetlistEntrySQL99(pstate, sortby->node,
2746 : targetlist, exprKind);
2747 : else
2748 87878 : tle = findTargetlistEntrySQL92(pstate, sortby->node,
2749 : targetlist, exprKind);
2750 :
2751 92044 : sortlist = addTargetToSortList(pstate, tle,
2752 : sortlist, *targetlist, sortby);
2753 : }
2754 :
2755 522826 : return sortlist;
2756 : }
2757 :
2758 : /*
2759 : * transformWindowDefinitions -
2760 : * transform window definitions (WindowDef to WindowClause)
2761 : */
2762 : List *
2763 461614 : transformWindowDefinitions(ParseState *pstate,
2764 : List *windowdefs,
2765 : List **targetlist)
2766 : {
2767 461614 : List *result = NIL;
2768 461614 : Index winref = 0;
2769 : ListCell *lc;
2770 :
2771 464270 : foreach(lc, windowdefs)
2772 : {
2773 2722 : WindowDef *windef = (WindowDef *) lfirst(lc);
2774 2722 : WindowClause *refwc = NULL;
2775 : List *partitionClause;
2776 : List *orderClause;
2777 2722 : Oid rangeopfamily = InvalidOid;
2778 2722 : Oid rangeopcintype = InvalidOid;
2779 : WindowClause *wc;
2780 :
2781 2722 : winref++;
2782 :
2783 : /*
2784 : * Check for duplicate window names.
2785 : */
2786 3262 : if (windef->name &&
2787 540 : findWindowClause(result, windef->name) != NULL)
2788 6 : ereport(ERROR,
2789 : (errcode(ERRCODE_WINDOWING_ERROR),
2790 : errmsg("window \"%s\" is already defined", windef->name),
2791 : parser_errposition(pstate, windef->location)));
2792 :
2793 : /*
2794 : * If it references a previous window, look that up.
2795 : */
2796 2716 : if (windef->refname)
2797 : {
2798 24 : refwc = findWindowClause(result, windef->refname);
2799 24 : if (refwc == NULL)
2800 0 : ereport(ERROR,
2801 : (errcode(ERRCODE_UNDEFINED_OBJECT),
2802 : errmsg("window \"%s\" does not exist",
2803 : windef->refname),
2804 : parser_errposition(pstate, windef->location)));
2805 : }
2806 :
2807 : /*
2808 : * Transform PARTITION and ORDER specs, if any. These are treated
2809 : * almost exactly like top-level GROUP BY and ORDER BY clauses,
2810 : * including the special handling of nondefault operator semantics.
2811 : */
2812 2716 : orderClause = transformSortClause(pstate,
2813 : windef->orderClause,
2814 : targetlist,
2815 : EXPR_KIND_WINDOW_ORDER,
2816 : true /* force SQL99 rules */ );
2817 2710 : partitionClause = transformGroupClause(pstate,
2818 : windef->partitionClause,
2819 : NULL,
2820 : targetlist,
2821 : orderClause,
2822 : EXPR_KIND_WINDOW_PARTITION,
2823 : true /* force SQL99 rules */ );
2824 :
2825 : /*
2826 : * And prepare the new WindowClause.
2827 : */
2828 2710 : wc = makeNode(WindowClause);
2829 2710 : wc->name = windef->name;
2830 2710 : wc->refname = windef->refname;
2831 :
2832 : /*
2833 : * Per spec, a windowdef that references a previous one copies the
2834 : * previous partition clause (and mustn't specify its own). It can
2835 : * specify its own ordering clause, but only if the previous one had
2836 : * none. It always specifies its own frame clause, and the previous
2837 : * one must not have a frame clause. Yeah, it's bizarre that each of
2838 : * these cases works differently, but SQL:2008 says so; see 7.11
2839 : * <window clause> syntax rule 10 and general rule 1. The frame
2840 : * clause rule is especially bizarre because it makes "OVER foo"
2841 : * different from "OVER (foo)", and requires the latter to throw an
2842 : * error if foo has a nondefault frame clause. Well, ours not to
2843 : * reason why, but we do go out of our way to throw a useful error
2844 : * message for such cases.
2845 : */
2846 2710 : if (refwc)
2847 : {
2848 24 : if (partitionClause)
2849 0 : ereport(ERROR,
2850 : (errcode(ERRCODE_WINDOWING_ERROR),
2851 : errmsg("cannot override PARTITION BY clause of window \"%s\"",
2852 : windef->refname),
2853 : parser_errposition(pstate, windef->location)));
2854 24 : wc->partitionClause = copyObject(refwc->partitionClause);
2855 : }
2856 : else
2857 2686 : wc->partitionClause = partitionClause;
2858 2710 : if (refwc)
2859 : {
2860 24 : if (orderClause && refwc->orderClause)
2861 0 : ereport(ERROR,
2862 : (errcode(ERRCODE_WINDOWING_ERROR),
2863 : errmsg("cannot override ORDER BY clause of window \"%s\"",
2864 : windef->refname),
2865 : parser_errposition(pstate, windef->location)));
2866 24 : if (orderClause)
2867 : {
2868 0 : wc->orderClause = orderClause;
2869 0 : wc->copiedOrder = false;
2870 : }
2871 : else
2872 : {
2873 24 : wc->orderClause = copyObject(refwc->orderClause);
2874 24 : wc->copiedOrder = true;
2875 : }
2876 : }
2877 : else
2878 : {
2879 2686 : wc->orderClause = orderClause;
2880 2686 : wc->copiedOrder = false;
2881 : }
2882 2710 : if (refwc && refwc->frameOptions != FRAMEOPTION_DEFAULTS)
2883 : {
2884 : /*
2885 : * Use this message if this is a WINDOW clause, or if it's an OVER
2886 : * clause that includes ORDER BY or framing clauses. (We already
2887 : * rejected PARTITION BY above, so no need to check that.)
2888 : */
2889 0 : if (windef->name ||
2890 0 : orderClause || windef->frameOptions != FRAMEOPTION_DEFAULTS)
2891 0 : ereport(ERROR,
2892 : (errcode(ERRCODE_WINDOWING_ERROR),
2893 : errmsg("cannot copy window \"%s\" because it has a frame clause",
2894 : windef->refname),
2895 : parser_errposition(pstate, windef->location)));
2896 : /* Else this clause is just OVER (foo), so say this: */
2897 0 : ereport(ERROR,
2898 : (errcode(ERRCODE_WINDOWING_ERROR),
2899 : errmsg("cannot copy window \"%s\" because it has a frame clause",
2900 : windef->refname),
2901 : errhint("Omit the parentheses in this OVER clause."),
2902 : parser_errposition(pstate, windef->location)));
2903 : }
2904 2710 : wc->frameOptions = windef->frameOptions;
2905 :
2906 : /*
2907 : * RANGE offset PRECEDING/FOLLOWING requires exactly one ORDER BY
2908 : * column; check that and get its sort opfamily info.
2909 : */
2910 2710 : if ((wc->frameOptions & FRAMEOPTION_RANGE) &&
2911 1936 : (wc->frameOptions & (FRAMEOPTION_START_OFFSET |
2912 : FRAMEOPTION_END_OFFSET)))
2913 : {
2914 : SortGroupClause *sortcl;
2915 : Node *sortkey;
2916 : int16 rangestrategy;
2917 :
2918 636 : if (list_length(wc->orderClause) != 1)
2919 18 : ereport(ERROR,
2920 : (errcode(ERRCODE_WINDOWING_ERROR),
2921 : errmsg("RANGE with offset PRECEDING/FOLLOWING requires exactly one ORDER BY column"),
2922 : parser_errposition(pstate, windef->location)));
2923 618 : sortcl = linitial_node(SortGroupClause, wc->orderClause);
2924 618 : sortkey = get_sortgroupclause_expr(sortcl, *targetlist);
2925 : /* Find the sort operator in pg_amop */
2926 618 : if (!get_ordering_op_properties(sortcl->sortop,
2927 : &rangeopfamily,
2928 : &rangeopcintype,
2929 : &rangestrategy))
2930 0 : elog(ERROR, "operator %u is not a valid ordering operator",
2931 : sortcl->sortop);
2932 : /* Record properties of sort ordering */
2933 618 : wc->inRangeColl = exprCollation(sortkey);
2934 618 : wc->inRangeAsc = !sortcl->reverse_sort;
2935 618 : wc->inRangeNullsFirst = sortcl->nulls_first;
2936 : }
2937 :
2938 : /* Per spec, GROUPS mode requires an ORDER BY clause */
2939 2692 : if (wc->frameOptions & FRAMEOPTION_GROUPS)
2940 : {
2941 174 : if (wc->orderClause == NIL)
2942 6 : ereport(ERROR,
2943 : (errcode(ERRCODE_WINDOWING_ERROR),
2944 : errmsg("GROUPS mode requires an ORDER BY clause"),
2945 : parser_errposition(pstate, windef->location)));
2946 : }
2947 :
2948 : /* Process frame offset expressions */
2949 2686 : wc->startOffset = transformFrameOffset(pstate, wc->frameOptions,
2950 : rangeopfamily, rangeopcintype,
2951 : &wc->startInRangeFunc,
2952 : windef->startOffset);
2953 2662 : wc->endOffset = transformFrameOffset(pstate, wc->frameOptions,
2954 : rangeopfamily, rangeopcintype,
2955 : &wc->endInRangeFunc,
2956 : windef->endOffset);
2957 2656 : wc->winref = winref;
2958 :
2959 2656 : result = lappend(result, wc);
2960 : }
2961 :
2962 461548 : return result;
2963 : }
2964 :
2965 : /*
2966 : * transformDistinctClause -
2967 : * transform a DISTINCT clause
2968 : *
2969 : * Since we may need to add items to the query's targetlist, that list
2970 : * is passed by reference.
2971 : *
2972 : * As with GROUP BY, we absorb the sorting semantics of ORDER BY as much as
2973 : * possible into the distinctClause. This avoids a possible need to re-sort,
2974 : * and allows the user to choose the equality semantics used by DISTINCT,
2975 : * should she be working with a datatype that has more than one equality
2976 : * operator.
2977 : *
2978 : * is_agg is true if we are transforming an aggregate(DISTINCT ...)
2979 : * function call. This does not affect any behavior, only the phrasing
2980 : * of error messages.
2981 : */
2982 : List *
2983 3524 : transformDistinctClause(ParseState *pstate,
2984 : List **targetlist, List *sortClause, bool is_agg)
2985 : {
2986 3524 : List *result = NIL;
2987 : ListCell *slitem;
2988 : ListCell *tlitem;
2989 :
2990 : /*
2991 : * The distinctClause should consist of all ORDER BY items followed by all
2992 : * other non-resjunk targetlist items. There must not be any resjunk
2993 : * ORDER BY items --- that would imply that we are sorting by a value that
2994 : * isn't necessarily unique within a DISTINCT group, so the results
2995 : * wouldn't be well-defined. This construction ensures we follow the rule
2996 : * that sortClause and distinctClause match; in fact the sortClause will
2997 : * always be a prefix of distinctClause.
2998 : *
2999 : * Note a corner case: the same TLE could be in the ORDER BY list multiple
3000 : * times with different sortops. We have to include it in the
3001 : * distinctClause the same way to preserve the prefix property. The net
3002 : * effect will be that the TLE value will be made unique according to both
3003 : * sortops.
3004 : */
3005 4138 : foreach(slitem, sortClause)
3006 : {
3007 650 : SortGroupClause *scl = (SortGroupClause *) lfirst(slitem);
3008 650 : TargetEntry *tle = get_sortgroupclause_tle(scl, *targetlist);
3009 :
3010 650 : if (tle->resjunk)
3011 36 : ereport(ERROR,
3012 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3013 : is_agg ?
3014 : errmsg("in an aggregate with DISTINCT, ORDER BY expressions must appear in argument list") :
3015 : errmsg("for SELECT DISTINCT, ORDER BY expressions must appear in select list"),
3016 : parser_errposition(pstate,
3017 : exprLocation((Node *) tle->expr))));
3018 614 : result = lappend(result, copyObject(scl));
3019 : }
3020 :
3021 : /*
3022 : * Now add any remaining non-resjunk tlist items, using default sort/group
3023 : * semantics for their data types.
3024 : */
3025 13772 : foreach(tlitem, *targetlist)
3026 : {
3027 10284 : TargetEntry *tle = (TargetEntry *) lfirst(tlitem);
3028 :
3029 10284 : if (tle->resjunk)
3030 4 : continue; /* ignore junk */
3031 10280 : result = addTargetToGroupList(pstate, tle,
3032 : result, *targetlist,
3033 10280 : exprLocation((Node *) tle->expr));
3034 : }
3035 :
3036 : /*
3037 : * Complain if we found nothing to make DISTINCT. Returning an empty list
3038 : * would cause the parsed Query to look like it didn't have DISTINCT, with
3039 : * results that would probably surprise the user. Note: this case is
3040 : * presently impossible for aggregates because of grammar restrictions,
3041 : * but we check anyway.
3042 : */
3043 3488 : if (result == NIL)
3044 0 : ereport(ERROR,
3045 : (errcode(ERRCODE_SYNTAX_ERROR),
3046 : is_agg ?
3047 : errmsg("an aggregate with DISTINCT must have at least one argument") :
3048 : errmsg("SELECT DISTINCT must have at least one column")));
3049 :
3050 3488 : return result;
3051 : }
3052 :
3053 : /*
3054 : * transformDistinctOnClause -
3055 : * transform a DISTINCT ON clause
3056 : *
3057 : * Since we may need to add items to the query's targetlist, that list
3058 : * is passed by reference.
3059 : *
3060 : * As with GROUP BY, we absorb the sorting semantics of ORDER BY as much as
3061 : * possible into the distinctClause. This avoids a possible need to re-sort,
3062 : * and allows the user to choose the equality semantics used by DISTINCT,
3063 : * should she be working with a datatype that has more than one equality
3064 : * operator.
3065 : */
3066 : List *
3067 204 : transformDistinctOnClause(ParseState *pstate, List *distinctlist,
3068 : List **targetlist, List *sortClause)
3069 : {
3070 204 : List *result = NIL;
3071 204 : List *sortgrouprefs = NIL;
3072 : bool skipped_sortitem;
3073 : ListCell *lc;
3074 : ListCell *lc2;
3075 :
3076 : /*
3077 : * Add all the DISTINCT ON expressions to the tlist (if not already
3078 : * present, they are added as resjunk items). Assign sortgroupref numbers
3079 : * to them, and make a list of these numbers. (NB: we rely below on the
3080 : * sortgrouprefs list being one-for-one with the original distinctlist.
3081 : * Also notice that we could have duplicate DISTINCT ON expressions and
3082 : * hence duplicate entries in sortgrouprefs.)
3083 : */
3084 492 : foreach(lc, distinctlist)
3085 : {
3086 294 : Node *dexpr = (Node *) lfirst(lc);
3087 : int sortgroupref;
3088 : TargetEntry *tle;
3089 :
3090 294 : tle = findTargetlistEntrySQL92(pstate, dexpr, targetlist,
3091 : EXPR_KIND_DISTINCT_ON);
3092 288 : sortgroupref = assignSortGroupRef(tle, *targetlist);
3093 288 : sortgrouprefs = lappend_int(sortgrouprefs, sortgroupref);
3094 : }
3095 :
3096 : /*
3097 : * If the user writes both DISTINCT ON and ORDER BY, adopt the sorting
3098 : * semantics from ORDER BY items that match DISTINCT ON items, and also
3099 : * adopt their column sort order. We insist that the distinctClause and
3100 : * sortClause match, so throw error if we find the need to add any more
3101 : * distinctClause items after we've skipped an ORDER BY item that wasn't
3102 : * in DISTINCT ON.
3103 : */
3104 198 : skipped_sortitem = false;
3105 480 : foreach(lc, sortClause)
3106 : {
3107 288 : SortGroupClause *scl = (SortGroupClause *) lfirst(lc);
3108 :
3109 288 : if (list_member_int(sortgrouprefs, scl->tleSortGroupRef))
3110 : {
3111 228 : if (skipped_sortitem)
3112 6 : ereport(ERROR,
3113 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3114 : errmsg("SELECT DISTINCT ON expressions must match initial ORDER BY expressions"),
3115 : parser_errposition(pstate,
3116 : get_matching_location(scl->tleSortGroupRef,
3117 : sortgrouprefs,
3118 : distinctlist))));
3119 : else
3120 222 : result = lappend(result, copyObject(scl));
3121 : }
3122 : else
3123 60 : skipped_sortitem = true;
3124 : }
3125 :
3126 : /*
3127 : * Now add any remaining DISTINCT ON items, using default sort/group
3128 : * semantics for their data types. (Note: this is pretty questionable; if
3129 : * the ORDER BY list doesn't include all the DISTINCT ON items and more
3130 : * besides, you certainly aren't using DISTINCT ON in the intended way,
3131 : * and you probably aren't going to get consistent results. It might be
3132 : * better to throw an error or warning here. But historically we've
3133 : * allowed it, so keep doing so.)
3134 : */
3135 468 : forboth(lc, distinctlist, lc2, sortgrouprefs)
3136 : {
3137 276 : Node *dexpr = (Node *) lfirst(lc);
3138 276 : int sortgroupref = lfirst_int(lc2);
3139 276 : TargetEntry *tle = get_sortgroupref_tle(sortgroupref, *targetlist);
3140 :
3141 276 : if (targetIsInSortList(tle, InvalidOid, result))
3142 216 : continue; /* already in list (with some semantics) */
3143 60 : if (skipped_sortitem)
3144 0 : ereport(ERROR,
3145 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3146 : errmsg("SELECT DISTINCT ON expressions must match initial ORDER BY expressions"),
3147 : parser_errposition(pstate, exprLocation(dexpr))));
3148 60 : result = addTargetToGroupList(pstate, tle,
3149 : result, *targetlist,
3150 : exprLocation(dexpr));
3151 : }
3152 :
3153 : /*
3154 : * An empty result list is impossible here because of grammar
3155 : * restrictions.
3156 : */
3157 : Assert(result != NIL);
3158 :
3159 192 : return result;
3160 : }
3161 :
3162 : /*
3163 : * get_matching_location
3164 : * Get the exprLocation of the exprs member corresponding to the
3165 : * (first) member of sortgrouprefs that equals sortgroupref.
3166 : *
3167 : * This is used so that we can point at a troublesome DISTINCT ON entry.
3168 : * (Note that we need to use the original untransformed DISTINCT ON list
3169 : * item, as whatever TLE it corresponds to will very possibly have a
3170 : * parse location pointing to some matching entry in the SELECT list
3171 : * or ORDER BY list.)
3172 : */
3173 : static int
3174 6 : get_matching_location(int sortgroupref, List *sortgrouprefs, List *exprs)
3175 : {
3176 : ListCell *lcs;
3177 : ListCell *lce;
3178 :
3179 12 : forboth(lcs, sortgrouprefs, lce, exprs)
3180 : {
3181 12 : if (lfirst_int(lcs) == sortgroupref)
3182 6 : return exprLocation((Node *) lfirst(lce));
3183 : }
3184 : /* if no match, caller blew it */
3185 0 : elog(ERROR, "get_matching_location: no matching sortgroupref");
3186 : return -1; /* keep compiler quiet */
3187 : }
3188 :
3189 : /*
3190 : * resolve_unique_index_expr
3191 : * Infer a unique index from a list of indexElems, for ON
3192 : * CONFLICT clause
3193 : *
3194 : * Perform parse analysis of expressions and columns appearing within ON
3195 : * CONFLICT clause. During planning, the returned list of expressions is used
3196 : * to infer which unique index to use.
3197 : */
3198 : static List *
3199 1418 : resolve_unique_index_expr(ParseState *pstate, InferClause *infer,
3200 : Relation heapRel)
3201 : {
3202 1418 : List *result = NIL;
3203 : ListCell *l;
3204 :
3205 3290 : foreach(l, infer->indexElems)
3206 : {
3207 1878 : IndexElem *ielem = (IndexElem *) lfirst(l);
3208 1878 : InferenceElem *pInfer = makeNode(InferenceElem);
3209 : Node *parse;
3210 :
3211 : /*
3212 : * Raw grammar re-uses CREATE INDEX infrastructure for unique index
3213 : * inference clause, and so will accept opclasses by name and so on.
3214 : *
3215 : * Make no attempt to match ASC or DESC ordering or NULLS FIRST/NULLS
3216 : * LAST ordering, since those are not significant for inference
3217 : * purposes (any unique index matching the inference specification in
3218 : * other regards is accepted indifferently). Actively reject this as
3219 : * wrong-headed.
3220 : */
3221 1878 : if (ielem->ordering != SORTBY_DEFAULT)
3222 0 : ereport(ERROR,
3223 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3224 : errmsg("ASC/DESC is not allowed in ON CONFLICT clause"),
3225 : parser_errposition(pstate,
3226 : exprLocation((Node *) infer))));
3227 1878 : if (ielem->nulls_ordering != SORTBY_NULLS_DEFAULT)
3228 0 : ereport(ERROR,
3229 : (errcode(ERRCODE_INVALID_COLUMN_REFERENCE),
3230 : errmsg("NULLS FIRST/LAST is not allowed in ON CONFLICT clause"),
3231 : parser_errposition(pstate,
3232 : exprLocation((Node *) infer))));
3233 :
3234 1878 : if (!ielem->expr)
3235 : {
3236 : /* Simple index attribute */
3237 : ColumnRef *n;
3238 :
3239 : /*
3240 : * Grammar won't have built raw expression for us in event of
3241 : * plain column reference. Create one directly, and perform
3242 : * expression transformation. Planner expects this, and performs
3243 : * its own normalization for the purposes of matching against
3244 : * pg_index.
3245 : */
3246 1704 : n = makeNode(ColumnRef);
3247 1704 : n->fields = list_make1(makeString(ielem->name));
3248 : /* Location is approximately that of inference specification */
3249 1704 : n->location = infer->location;
3250 1704 : parse = (Node *) n;
3251 : }
3252 : else
3253 : {
3254 : /* Do parse transformation of the raw expression */
3255 174 : parse = (Node *) ielem->expr;
3256 : }
3257 :
3258 : /*
3259 : * transformExpr() will reject subqueries, aggregates, window
3260 : * functions, and SRFs, based on being passed
3261 : * EXPR_KIND_INDEX_EXPRESSION. So we needn't worry about those
3262 : * further ... not that they would match any available index
3263 : * expression anyway.
3264 : */
3265 1878 : pInfer->expr = transformExpr(pstate, parse, EXPR_KIND_INDEX_EXPRESSION);
3266 :
3267 : /* Perform lookup of collation and operator class as required */
3268 1872 : if (!ielem->collation)
3269 1830 : pInfer->infercollid = InvalidOid;
3270 : else
3271 42 : pInfer->infercollid = LookupCollation(pstate, ielem->collation,
3272 42 : exprLocation(pInfer->expr));
3273 :
3274 1872 : if (!ielem->opclass)
3275 1830 : pInfer->inferopclass = InvalidOid;
3276 : else
3277 42 : pInfer->inferopclass = get_opclass_oid(BTREE_AM_OID,
3278 : ielem->opclass, false);
3279 :
3280 1872 : result = lappend(result, pInfer);
3281 : }
3282 :
3283 1412 : return result;
3284 : }
3285 :
3286 : /*
3287 : * transformOnConflictArbiter -
3288 : * transform arbiter expressions in an ON CONFLICT clause.
3289 : *
3290 : * Transformed expressions used to infer one unique index relation to serve as
3291 : * an ON CONFLICT arbiter. Partial unique indexes may be inferred using WHERE
3292 : * clause from inference specification clause.
3293 : */
3294 : void
3295 1844 : transformOnConflictArbiter(ParseState *pstate,
3296 : OnConflictClause *onConflictClause,
3297 : List **arbiterExpr, Node **arbiterWhere,
3298 : Oid *constraint)
3299 : {
3300 1844 : InferClause *infer = onConflictClause->infer;
3301 :
3302 1844 : *arbiterExpr = NIL;
3303 1844 : *arbiterWhere = NULL;
3304 1844 : *constraint = InvalidOid;
3305 :
3306 1844 : if (onConflictClause->action == ONCONFLICT_UPDATE && !infer)
3307 6 : ereport(ERROR,
3308 : (errcode(ERRCODE_SYNTAX_ERROR),
3309 : errmsg("ON CONFLICT DO UPDATE requires inference specification or constraint name"),
3310 : errhint("For example, ON CONFLICT (column_name)."),
3311 : parser_errposition(pstate,
3312 : exprLocation((Node *) onConflictClause))));
3313 :
3314 : /*
3315 : * To simplify certain aspects of its design, speculative insertion into
3316 : * system catalogs is disallowed
3317 : */
3318 1838 : if (IsCatalogRelation(pstate->p_target_relation))
3319 0 : ereport(ERROR,
3320 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3321 : errmsg("ON CONFLICT is not supported with system catalog tables"),
3322 : parser_errposition(pstate,
3323 : exprLocation((Node *) onConflictClause))));
3324 :
3325 : /* Same applies to table used by logical decoding as catalog table */
3326 1838 : if (RelationIsUsedAsCatalogTable(pstate->p_target_relation))
3327 0 : ereport(ERROR,
3328 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3329 : errmsg("ON CONFLICT is not supported on table \"%s\" used as a catalog table",
3330 : RelationGetRelationName(pstate->p_target_relation)),
3331 : parser_errposition(pstate,
3332 : exprLocation((Node *) onConflictClause))));
3333 :
3334 : /* ON CONFLICT DO NOTHING does not require an inference clause */
3335 1838 : if (infer)
3336 : {
3337 1610 : if (infer->indexElems)
3338 1418 : *arbiterExpr = resolve_unique_index_expr(pstate, infer,
3339 : pstate->p_target_relation);
3340 :
3341 : /*
3342 : * Handling inference WHERE clause (for partial unique index
3343 : * inference)
3344 : */
3345 1604 : if (infer->whereClause)
3346 48 : *arbiterWhere = transformExpr(pstate, infer->whereClause,
3347 : EXPR_KIND_INDEX_PREDICATE);
3348 :
3349 : /*
3350 : * If the arbiter is specified by constraint name, get the constraint
3351 : * OID and mark the constrained columns as requiring SELECT privilege,
3352 : * in the same way as would have happened if the arbiter had been
3353 : * specified by explicit reference to the constraint's index columns.
3354 : */
3355 1604 : if (infer->conname)
3356 : {
3357 192 : Oid relid = RelationGetRelid(pstate->p_target_relation);
3358 192 : RTEPermissionInfo *perminfo = pstate->p_target_nsitem->p_perminfo;
3359 : Bitmapset *conattnos;
3360 :
3361 192 : conattnos = get_relation_constraint_attnos(relid, infer->conname,
3362 : false, constraint);
3363 :
3364 : /* Make sure the rel as a whole is marked for SELECT access */
3365 192 : perminfo->requiredPerms |= ACL_SELECT;
3366 : /* Mark the constrained columns as requiring SELECT access */
3367 192 : perminfo->selectedCols = bms_add_members(perminfo->selectedCols,
3368 : conattnos);
3369 : }
3370 : }
3371 :
3372 : /*
3373 : * It's convenient to form a list of expressions based on the
3374 : * representation used by CREATE INDEX, since the same restrictions are
3375 : * appropriate (e.g. on subqueries). However, from here on, a dedicated
3376 : * primnode representation is used for inference elements, and so
3377 : * assign_query_collations() can be trusted to do the right thing with the
3378 : * post parse analysis query tree inference clause representation.
3379 : */
3380 1832 : }
3381 :
3382 : /*
3383 : * addTargetToSortList
3384 : * If the given targetlist entry isn't already in the SortGroupClause
3385 : * list, add it to the end of the list, using the given sort ordering
3386 : * info.
3387 : *
3388 : * Returns the updated SortGroupClause list.
3389 : */
3390 : List *
3391 92416 : addTargetToSortList(ParseState *pstate, TargetEntry *tle,
3392 : List *sortlist, List *targetlist, SortBy *sortby)
3393 : {
3394 92416 : Oid restype = exprType((Node *) tle->expr);
3395 : Oid sortop;
3396 : Oid eqop;
3397 : bool hashable;
3398 : bool reverse;
3399 : int location;
3400 : ParseCallbackState pcbstate;
3401 :
3402 : /* if tlist item is an UNKNOWN literal, change it to TEXT */
3403 92416 : if (restype == UNKNOWNOID)
3404 : {
3405 12 : tle->expr = (Expr *) coerce_type(pstate, (Node *) tle->expr,
3406 : restype, TEXTOID, -1,
3407 : COERCION_IMPLICIT,
3408 : COERCE_IMPLICIT_CAST,
3409 : -1);
3410 12 : restype = TEXTOID;
3411 : }
3412 :
3413 : /*
3414 : * Rather than clutter the API of get_sort_group_operators and the other
3415 : * functions we're about to use, make use of error context callback to
3416 : * mark any error reports with a parse position. We point to the operator
3417 : * location if present, else to the expression being sorted. (NB: use the
3418 : * original untransformed expression here; the TLE entry might well point
3419 : * at a duplicate expression in the regular SELECT list.)
3420 : */
3421 92416 : location = sortby->location;
3422 92416 : if (location < 0)
3423 92196 : location = exprLocation(sortby->node);
3424 92416 : setup_parser_errposition_callback(&pcbstate, pstate, location);
3425 :
3426 : /* determine the sortop, eqop, and directionality */
3427 92416 : switch (sortby->sortby_dir)
3428 : {
3429 89430 : case SORTBY_DEFAULT:
3430 : case SORTBY_ASC:
3431 89430 : get_sort_group_operators(restype,
3432 : true, true, false,
3433 : &sortop, &eqop, NULL,
3434 : &hashable);
3435 89424 : reverse = false;
3436 89424 : break;
3437 2766 : case SORTBY_DESC:
3438 2766 : get_sort_group_operators(restype,
3439 : false, true, true,
3440 : NULL, &eqop, &sortop,
3441 : &hashable);
3442 2766 : reverse = true;
3443 2766 : break;
3444 220 : case SORTBY_USING:
3445 : Assert(sortby->useOp != NIL);
3446 220 : sortop = compatible_oper_opid(sortby->useOp,
3447 : restype,
3448 : restype,
3449 : false);
3450 :
3451 : /*
3452 : * Verify it's a valid ordering operator, fetch the corresponding
3453 : * equality operator, and determine whether to consider it like
3454 : * ASC or DESC.
3455 : */
3456 220 : eqop = get_equality_op_for_ordering_op(sortop, &reverse);
3457 220 : if (!OidIsValid(eqop))
3458 0 : ereport(ERROR,
3459 : (errcode(ERRCODE_WRONG_OBJECT_TYPE),
3460 : errmsg("operator %s is not a valid ordering operator",
3461 : strVal(llast(sortby->useOp))),
3462 : errhint("Ordering operators must be \"<\" or \">\" members of btree operator families.")));
3463 :
3464 : /*
3465 : * Also see if the equality operator is hashable.
3466 : */
3467 220 : hashable = op_hashjoinable(eqop, restype);
3468 220 : break;
3469 0 : default:
3470 0 : elog(ERROR, "unrecognized sortby_dir: %d", sortby->sortby_dir);
3471 : sortop = InvalidOid; /* keep compiler quiet */
3472 : eqop = InvalidOid;
3473 : hashable = false;
3474 : reverse = false;
3475 : break;
3476 : }
3477 :
3478 92410 : cancel_parser_errposition_callback(&pcbstate);
3479 :
3480 : /* avoid making duplicate sortlist entries */
3481 92410 : if (!targetIsInSortList(tle, sortop, sortlist))
3482 : {
3483 92410 : SortGroupClause *sortcl = makeNode(SortGroupClause);
3484 :
3485 92410 : sortcl->tleSortGroupRef = assignSortGroupRef(tle, targetlist);
3486 :
3487 92410 : sortcl->eqop = eqop;
3488 92410 : sortcl->sortop = sortop;
3489 92410 : sortcl->hashable = hashable;
3490 92410 : sortcl->reverse_sort = reverse;
3491 :
3492 92410 : switch (sortby->sortby_nulls)
3493 : {
3494 90406 : case SORTBY_NULLS_DEFAULT:
3495 : /* NULLS FIRST is default for DESC; other way for ASC */
3496 90406 : sortcl->nulls_first = reverse;
3497 90406 : break;
3498 336 : case SORTBY_NULLS_FIRST:
3499 336 : sortcl->nulls_first = true;
3500 336 : break;
3501 1668 : case SORTBY_NULLS_LAST:
3502 1668 : sortcl->nulls_first = false;
3503 1668 : break;
3504 0 : default:
3505 0 : elog(ERROR, "unrecognized sortby_nulls: %d",
3506 : sortby->sortby_nulls);
3507 : break;
3508 : }
3509 :
3510 92410 : sortlist = lappend(sortlist, sortcl);
3511 : }
3512 :
3513 92410 : return sortlist;
3514 : }
3515 :
3516 : /*
3517 : * addTargetToGroupList
3518 : * If the given targetlist entry isn't already in the SortGroupClause
3519 : * list, add it to the end of the list, using default sort/group
3520 : * semantics.
3521 : *
3522 : * This is very similar to addTargetToSortList, except that we allow the
3523 : * case where only a grouping (equality) operator can be found, and that
3524 : * the TLE is considered "already in the list" if it appears there with any
3525 : * sorting semantics.
3526 : *
3527 : * location is the parse location to be fingered in event of trouble. Note
3528 : * that we can't rely on exprLocation(tle->expr), because that might point
3529 : * to a SELECT item that matches the GROUP BY item; it'd be pretty confusing
3530 : * to report such a location.
3531 : *
3532 : * Returns the updated SortGroupClause list.
3533 : */
3534 : static List *
3535 16798 : addTargetToGroupList(ParseState *pstate, TargetEntry *tle,
3536 : List *grouplist, List *targetlist, int location)
3537 : {
3538 16798 : Oid restype = exprType((Node *) tle->expr);
3539 :
3540 : /* if tlist item is an UNKNOWN literal, change it to TEXT */
3541 16798 : if (restype == UNKNOWNOID)
3542 : {
3543 16 : tle->expr = (Expr *) coerce_type(pstate, (Node *) tle->expr,
3544 : restype, TEXTOID, -1,
3545 : COERCION_IMPLICIT,
3546 : COERCE_IMPLICIT_CAST,
3547 : -1);
3548 16 : restype = TEXTOID;
3549 : }
3550 :
3551 : /* avoid making duplicate grouplist entries */
3552 16798 : if (!targetIsInSortList(tle, InvalidOid, grouplist))
3553 : {
3554 16208 : SortGroupClause *grpcl = makeNode(SortGroupClause);
3555 : Oid sortop;
3556 : Oid eqop;
3557 : bool hashable;
3558 : ParseCallbackState pcbstate;
3559 :
3560 16208 : setup_parser_errposition_callback(&pcbstate, pstate, location);
3561 :
3562 : /* determine the eqop and optional sortop */
3563 16208 : get_sort_group_operators(restype,
3564 : false, true, false,
3565 : &sortop, &eqop, NULL,
3566 : &hashable);
3567 :
3568 16208 : cancel_parser_errposition_callback(&pcbstate);
3569 :
3570 16208 : grpcl->tleSortGroupRef = assignSortGroupRef(tle, targetlist);
3571 16208 : grpcl->eqop = eqop;
3572 16208 : grpcl->sortop = sortop;
3573 16208 : grpcl->reverse_sort = false; /* sortop is "less than", or
3574 : * InvalidOid */
3575 16208 : grpcl->nulls_first = false; /* OK with or without sortop */
3576 16208 : grpcl->hashable = hashable;
3577 :
3578 16208 : grouplist = lappend(grouplist, grpcl);
3579 : }
3580 :
3581 16798 : return grouplist;
3582 : }
3583 :
3584 : /*
3585 : * assignSortGroupRef
3586 : * Assign the targetentry an unused ressortgroupref, if it doesn't
3587 : * already have one. Return the assigned or pre-existing refnumber.
3588 : *
3589 : * 'tlist' is the targetlist containing (or to contain) the given targetentry.
3590 : */
3591 : Index
3592 141612 : assignSortGroupRef(TargetEntry *tle, List *tlist)
3593 : {
3594 : Index maxRef;
3595 : ListCell *l;
3596 :
3597 141612 : if (tle->ressortgroupref) /* already has one? */
3598 820 : return tle->ressortgroupref;
3599 :
3600 : /* easiest way to pick an unused refnumber: max used + 1 */
3601 140792 : maxRef = 0;
3602 796300 : foreach(l, tlist)
3603 : {
3604 655508 : Index ref = ((TargetEntry *) lfirst(l))->ressortgroupref;
3605 :
3606 655508 : if (ref > maxRef)
3607 95442 : maxRef = ref;
3608 : }
3609 140792 : tle->ressortgroupref = maxRef + 1;
3610 140792 : return tle->ressortgroupref;
3611 : }
3612 :
3613 : /*
3614 : * targetIsInSortList
3615 : * Is the given target item already in the sortlist?
3616 : * If sortop is not InvalidOid, also test for a match to the sortop.
3617 : *
3618 : * It is not an oversight that this function ignores the nulls_first flag.
3619 : * We check sortop when determining if an ORDER BY item is redundant with
3620 : * earlier ORDER BY items, because it's conceivable that "ORDER BY
3621 : * foo USING <, foo USING <<<" is not redundant, if <<< distinguishes
3622 : * values that < considers equal. We need not check nulls_first
3623 : * however, because a lower-order column with the same sortop but
3624 : * opposite nulls direction is redundant. Also, we can consider
3625 : * ORDER BY foo ASC, foo DESC redundant, so check for a commutator match.
3626 : *
3627 : * Works for both ordering and grouping lists (sortop would normally be
3628 : * InvalidOid when considering grouping). Note that the main reason we need
3629 : * this routine (and not just a quick test for nonzeroness of ressortgroupref)
3630 : * is that a TLE might be in only one of the lists.
3631 : */
3632 : bool
3633 113854 : targetIsInSortList(TargetEntry *tle, Oid sortop, List *sortList)
3634 : {
3635 113854 : Index ref = tle->ressortgroupref;
3636 : ListCell *l;
3637 :
3638 : /* no need to scan list if tle has no marker */
3639 113854 : if (ref == 0)
3640 109212 : return false;
3641 :
3642 5874 : foreach(l, sortList)
3643 : {
3644 2692 : SortGroupClause *scl = (SortGroupClause *) lfirst(l);
3645 :
3646 2692 : if (scl->tleSortGroupRef == ref &&
3647 0 : (sortop == InvalidOid ||
3648 0 : sortop == scl->sortop ||
3649 0 : sortop == get_commutator(scl->sortop)))
3650 1460 : return true;
3651 : }
3652 3182 : return false;
3653 : }
3654 :
3655 : /*
3656 : * findWindowClause
3657 : * Find the named WindowClause in the list, or return NULL if not there
3658 : */
3659 : static WindowClause *
3660 564 : findWindowClause(List *wclist, const char *name)
3661 : {
3662 : ListCell *l;
3663 :
3664 570 : foreach(l, wclist)
3665 : {
3666 36 : WindowClause *wc = (WindowClause *) lfirst(l);
3667 :
3668 36 : if (wc->name && strcmp(wc->name, name) == 0)
3669 30 : return wc;
3670 : }
3671 :
3672 534 : return NULL;
3673 : }
3674 :
3675 : /*
3676 : * transformFrameOffset
3677 : * Process a window frame offset expression
3678 : *
3679 : * In RANGE mode, rangeopfamily is the sort opfamily for the input ORDER BY
3680 : * column, and rangeopcintype is the input data type the sort operator is
3681 : * registered with. We expect the in_range function to be registered with
3682 : * that same type. (In binary-compatible cases, it might be different from
3683 : * the input column's actual type, so we can't use that for the lookups.)
3684 : * We'll return the OID of the in_range function to *inRangeFunc.
3685 : */
3686 : static Node *
3687 5348 : transformFrameOffset(ParseState *pstate, int frameOptions,
3688 : Oid rangeopfamily, Oid rangeopcintype, Oid *inRangeFunc,
3689 : Node *clause)
3690 : {
3691 5348 : const char *constructName = NULL;
3692 : Node *node;
3693 :
3694 5348 : *inRangeFunc = InvalidOid; /* default result */
3695 :
3696 : /* Quick exit if no offset expression */
3697 5348 : if (clause == NULL)
3698 3518 : return NULL;
3699 :
3700 1830 : if (frameOptions & FRAMEOPTION_ROWS)
3701 : {
3702 : /* Transform the raw expression tree */
3703 396 : node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_ROWS);
3704 :
3705 : /*
3706 : * Like LIMIT clause, simply coerce to int8
3707 : */
3708 396 : constructName = "ROWS";
3709 396 : node = coerce_to_specific_type(pstate, node, INT8OID, constructName);
3710 : }
3711 1434 : else if (frameOptions & FRAMEOPTION_RANGE)
3712 : {
3713 : /*
3714 : * We must look up the in_range support function that's to be used,
3715 : * possibly choosing one of several, and coerce the "offset" value to
3716 : * the appropriate input type.
3717 : */
3718 : Oid nodeType;
3719 : Oid preferredType;
3720 1152 : int nfuncs = 0;
3721 1152 : int nmatches = 0;
3722 1152 : Oid selectedType = InvalidOid;
3723 1152 : Oid selectedFunc = InvalidOid;
3724 : CatCList *proclist;
3725 : int i;
3726 :
3727 : /* Transform the raw expression tree */
3728 1152 : node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_RANGE);
3729 1152 : nodeType = exprType(node);
3730 :
3731 : /*
3732 : * If there are multiple candidates, we'll prefer the one that exactly
3733 : * matches nodeType; or if nodeType is as yet unknown, prefer the one
3734 : * that exactly matches the sort column type. (The second rule is
3735 : * like what we do for "known_type operator unknown".)
3736 : */
3737 1152 : preferredType = (nodeType != UNKNOWNOID) ? nodeType : rangeopcintype;
3738 :
3739 : /* Find the in_range support functions applicable to this case */
3740 1152 : proclist = SearchSysCacheList2(AMPROCNUM,
3741 : ObjectIdGetDatum(rangeopfamily),
3742 : ObjectIdGetDatum(rangeopcintype));
3743 7344 : for (i = 0; i < proclist->n_members; i++)
3744 : {
3745 6192 : HeapTuple proctup = &proclist->members[i]->tuple;
3746 6192 : Form_pg_amproc procform = (Form_pg_amproc) GETSTRUCT(proctup);
3747 :
3748 : /* The search will find all support proc types; ignore others */
3749 6192 : if (procform->amprocnum != BTINRANGE_PROC)
3750 4434 : continue;
3751 1758 : nfuncs++;
3752 :
3753 : /* Ignore function if given value can't be coerced to that type */
3754 1758 : if (!can_coerce_type(1, &nodeType, &procform->amprocrighttype,
3755 : COERCION_IMPLICIT))
3756 330 : continue;
3757 1428 : nmatches++;
3758 :
3759 : /* Remember preferred match, or any match if didn't find that */
3760 1428 : if (selectedType != preferredType)
3761 : {
3762 1368 : selectedType = procform->amprocrighttype;
3763 1368 : selectedFunc = procform->amproc;
3764 : }
3765 : }
3766 1152 : ReleaseCatCacheList(proclist);
3767 :
3768 : /*
3769 : * Throw error if needed. It seems worth taking the trouble to
3770 : * distinguish "no support at all" from "you didn't match any
3771 : * available offset type".
3772 : */
3773 1152 : if (nfuncs == 0)
3774 6 : ereport(ERROR,
3775 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3776 : errmsg("RANGE with offset PRECEDING/FOLLOWING is not supported for column type %s",
3777 : format_type_be(rangeopcintype)),
3778 : parser_errposition(pstate, exprLocation(node))));
3779 1146 : if (nmatches == 0)
3780 18 : ereport(ERROR,
3781 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3782 : errmsg("RANGE with offset PRECEDING/FOLLOWING is not supported for column type %s and offset type %s",
3783 : format_type_be(rangeopcintype),
3784 : format_type_be(nodeType)),
3785 : errhint("Cast the offset value to an appropriate type."),
3786 : parser_errposition(pstate, exprLocation(node))));
3787 1128 : if (nmatches != 1 && selectedType != preferredType)
3788 0 : ereport(ERROR,
3789 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3790 : errmsg("RANGE with offset PRECEDING/FOLLOWING has multiple interpretations for column type %s and offset type %s",
3791 : format_type_be(rangeopcintype),
3792 : format_type_be(nodeType)),
3793 : errhint("Cast the offset value to the exact intended type."),
3794 : parser_errposition(pstate, exprLocation(node))));
3795 :
3796 : /* OK, coerce the offset to the right type */
3797 1128 : constructName = "RANGE";
3798 1128 : node = coerce_to_specific_type(pstate, node,
3799 : selectedType, constructName);
3800 1128 : *inRangeFunc = selectedFunc;
3801 : }
3802 282 : else if (frameOptions & FRAMEOPTION_GROUPS)
3803 : {
3804 : /* Transform the raw expression tree */
3805 282 : node = transformExpr(pstate, clause, EXPR_KIND_WINDOW_FRAME_GROUPS);
3806 :
3807 : /*
3808 : * Like LIMIT clause, simply coerce to int8
3809 : */
3810 282 : constructName = "GROUPS";
3811 282 : node = coerce_to_specific_type(pstate, node, INT8OID, constructName);
3812 : }
3813 : else
3814 : {
3815 : Assert(false);
3816 0 : node = NULL;
3817 : }
3818 :
3819 : /* Disallow variables in frame offsets */
3820 1806 : checkExprIsVarFree(pstate, node, constructName);
3821 :
3822 1800 : return node;
3823 : }
|
