1 /*-------------------------------------------------------------------------
2 *
3 * clauses.c
4 * routines to manipulate qualification clauses
5 *
6 * Portions Copyright (c) 1996-2016, PostgreSQL Global Development Group
7 * Portions Copyright (c) 1994, Regents of the University of California
8 *
9 *
10 * IDENTIFICATION
11 * src/backend/optimizer/util/clauses.c
12 *
13 * HISTORY
14 * AUTHOR DATE MAJOR EVENT
15 * Andrew Yu Nov 3, 1994 clause.c and clauses.c combined
16 *
17 *-------------------------------------------------------------------------
18 */
19
20 #include "postgres.h"
21
22 #include "access/htup_details.h"
23 #include "catalog/pg_aggregate.h"
24 #include "catalog/pg_class.h"
25 #include "catalog/pg_language.h"
26 #include "catalog/pg_operator.h"
27 #include "catalog/pg_proc.h"
28 #include "catalog/pg_type.h"
29 #include "executor/executor.h"
30 #include "executor/functions.h"
31 #include "funcapi.h"
32 #include "miscadmin.h"
33 #include "nodes/makefuncs.h"
34 #include "nodes/nodeFuncs.h"
35 #include "optimizer/clauses.h"
36 #include "optimizer/cost.h"
37 #include "optimizer/planmain.h"
38 #include "optimizer/prep.h"
39 #include "optimizer/var.h"
40 #include "parser/analyze.h"
41 #include "parser/parse_agg.h"
42 #include "parser/parse_coerce.h"
43 #include "parser/parse_func.h"
44 #include "rewrite/rewriteManip.h"
45 #include "tcop/tcopprot.h"
46 #include "utils/acl.h"
47 #include "utils/builtins.h"
48 #include "utils/datum.h"
49 #include "utils/fmgroids.h"
50 #include "utils/lsyscache.h"
51 #include "utils/memutils.h"
52 #include "utils/syscache.h"
53 #include "utils/typcache.h"
54
55
56 typedef struct
57 {
58 PlannerInfo *root;
59 AggSplit aggsplit;
60 AggClauseCosts *costs;
61 } get_agg_clause_costs_context;
62
63 typedef struct
64 {
65 ParamListInfo boundParams;
66 PlannerInfo *root;
67 List *active_fns;
68 Node *case_val;
69 bool estimate;
70 } eval_const_expressions_context;
71
72 typedef struct
73 {
74 int nargs;
75 List *args;
76 int *usecounts;
77 } substitute_actual_parameters_context;
78
79 typedef struct
80 {
81 int nargs;
82 List *args;
83 int sublevels_up;
84 } substitute_actual_srf_parameters_context;
85
86 typedef struct
87 {
88 char *proname;
89 char *prosrc;
90 } inline_error_callback_arg;
91
92 typedef struct
93 {
94 bool allow_restricted;
95 } has_parallel_hazard_arg;
96
97 static bool contain_agg_clause_walker(Node *node, void *context);
98 static bool get_agg_clause_costs_walker(Node *node,
99 get_agg_clause_costs_context *context);
100 static bool find_window_functions_walker(Node *node, WindowFuncLists *lists);
101 static bool expression_returns_set_rows_walker(Node *node, double *count);
102 static bool contain_subplans_walker(Node *node, void *context);
103 static bool contain_mutable_functions_walker(Node *node, void *context);
104 static bool contain_volatile_functions_walker(Node *node, void *context);
105 static bool contain_volatile_functions_not_nextval_walker(Node *node, void *context);
106 static bool has_parallel_hazard_walker(Node *node,
107 has_parallel_hazard_arg *context);
108 static bool contain_nonstrict_functions_walker(Node *node, void *context);
109 static bool contain_exec_param_walker(Node *node, List *param_ids);
110 static bool contain_context_dependent_node(Node *clause);
111 static bool contain_context_dependent_node_walker(Node *node, int *flags);
112 static bool contain_leaked_vars_walker(Node *node, void *context);
113 static Relids find_nonnullable_rels_walker(Node *node, bool top_level);
114 static List *find_nonnullable_vars_walker(Node *node, bool top_level);
115 static bool is_strict_saop(ScalarArrayOpExpr *expr, bool falseOK);
116 static Node *eval_const_expressions_mutator(Node *node,
117 eval_const_expressions_context *context);
118 static List *simplify_or_arguments(List *args,
119 eval_const_expressions_context *context,
120 bool *haveNull, bool *forceTrue);
121 static List *simplify_and_arguments(List *args,
122 eval_const_expressions_context *context,
123 bool *haveNull, bool *forceFalse);
124 static Node *simplify_boolean_equality(Oid opno, List *args);
125 static Expr *simplify_function(Oid funcid,
126 Oid result_type, int32 result_typmod,
127 Oid result_collid, Oid input_collid, List **args_p,
128 bool funcvariadic, bool process_args, bool allow_non_const,
129 eval_const_expressions_context *context);
130 static List *expand_function_arguments(List *args, Oid result_type,
131 HeapTuple func_tuple);
132 static List *reorder_function_arguments(List *args, HeapTuple func_tuple);
133 static List *add_function_defaults(List *args, HeapTuple func_tuple);
134 static List *fetch_function_defaults(HeapTuple func_tuple);
135 static void recheck_cast_function_args(List *args, Oid result_type,
136 HeapTuple func_tuple);
137 static Expr *evaluate_function(Oid funcid, Oid result_type, int32 result_typmod,
138 Oid result_collid, Oid input_collid, List *args,
139 bool funcvariadic,
140 HeapTuple func_tuple,
141 eval_const_expressions_context *context);
142 static Expr *inline_function(Oid funcid, Oid result_type, Oid result_collid,
143 Oid input_collid, List *args,
144 bool funcvariadic,
145 HeapTuple func_tuple,
146 eval_const_expressions_context *context);
147 static Node *substitute_actual_parameters(Node *expr, int nargs, List *args,
148 int *usecounts);
149 static Node *substitute_actual_parameters_mutator(Node *node,
150 substitute_actual_parameters_context *context);
151 static void sql_inline_error_callback(void *arg);
152 static Expr *evaluate_expr(Expr *expr, Oid result_type, int32 result_typmod,
153 Oid result_collation);
154 static Query *substitute_actual_srf_parameters(Query *expr,
155 int nargs, List *args);
156 static Node *substitute_actual_srf_parameters_mutator(Node *node,
157 substitute_actual_srf_parameters_context *context);
158 static bool tlist_matches_coltypelist(List *tlist, List *coltypelist);
159
160
161 /*****************************************************************************
162 * OPERATOR clause functions
163 *****************************************************************************/
164
165 /*
166 * make_opclause
167 * Creates an operator clause given its operator info, left operand
168 * and right operand (pass NULL to create single-operand clause),
169 * and collation info.
170 */
171 Expr *
make_opclause(Oid opno,Oid opresulttype,bool opretset,Expr * leftop,Expr * rightop,Oid opcollid,Oid inputcollid)172 make_opclause(Oid opno, Oid opresulttype, bool opretset,
173 Expr *leftop, Expr *rightop,
174 Oid opcollid, Oid inputcollid)
175 {
176 OpExpr *expr = makeNode(OpExpr);
177
178 expr->opno = opno;
179 expr->opfuncid = InvalidOid;
180 expr->opresulttype = opresulttype;
181 expr->opretset = opretset;
182 expr->opcollid = opcollid;
183 expr->inputcollid = inputcollid;
184 if (rightop)
185 expr->args = list_make2(leftop, rightop);
186 else
187 expr->args = list_make1(leftop);
188 expr->location = -1;
189 return (Expr *) expr;
190 }
191
192 /*
193 * get_leftop
194 *
195 * Returns the left operand of a clause of the form (op expr expr)
196 * or (op expr)
197 */
198 Node *
get_leftop(const Expr * clause)199 get_leftop(const Expr *clause)
200 {
201 const OpExpr *expr = (const OpExpr *) clause;
202
203 if (expr->args != NIL)
204 return linitial(expr->args);
205 else
206 return NULL;
207 }
208
209 /*
210 * get_rightop
211 *
212 * Returns the right operand in a clause of the form (op expr expr).
213 * NB: result will be NULL if applied to a unary op clause.
214 */
215 Node *
get_rightop(const Expr * clause)216 get_rightop(const Expr *clause)
217 {
218 const OpExpr *expr = (const OpExpr *) clause;
219
220 if (list_length(expr->args) >= 2)
221 return lsecond(expr->args);
222 else
223 return NULL;
224 }
225
226 /*****************************************************************************
227 * NOT clause functions
228 *****************************************************************************/
229
230 /*
231 * not_clause
232 *
233 * Returns t iff this is a 'not' clause: (NOT expr).
234 */
235 bool
not_clause(Node * clause)236 not_clause(Node *clause)
237 {
238 return (clause != NULL &&
239 IsA(clause, BoolExpr) &&
240 ((BoolExpr *) clause)->boolop == NOT_EXPR);
241 }
242
243 /*
244 * make_notclause
245 *
246 * Create a 'not' clause given the expression to be negated.
247 */
248 Expr *
make_notclause(Expr * notclause)249 make_notclause(Expr *notclause)
250 {
251 BoolExpr *expr = makeNode(BoolExpr);
252
253 expr->boolop = NOT_EXPR;
254 expr->args = list_make1(notclause);
255 expr->location = -1;
256 return (Expr *) expr;
257 }
258
259 /*
260 * get_notclausearg
261 *
262 * Retrieve the clause within a 'not' clause
263 */
264 Expr *
get_notclausearg(Expr * notclause)265 get_notclausearg(Expr *notclause)
266 {
267 return linitial(((BoolExpr *) notclause)->args);
268 }
269
270 /*****************************************************************************
271 * OR clause functions
272 *****************************************************************************/
273
274 /*
275 * or_clause
276 *
277 * Returns t iff the clause is an 'or' clause: (OR { expr }).
278 */
279 bool
or_clause(Node * clause)280 or_clause(Node *clause)
281 {
282 return (clause != NULL &&
283 IsA(clause, BoolExpr) &&
284 ((BoolExpr *) clause)->boolop == OR_EXPR);
285 }
286
287 /*
288 * make_orclause
289 *
290 * Creates an 'or' clause given a list of its subclauses.
291 */
292 Expr *
make_orclause(List * orclauses)293 make_orclause(List *orclauses)
294 {
295 BoolExpr *expr = makeNode(BoolExpr);
296
297 expr->boolop = OR_EXPR;
298 expr->args = orclauses;
299 expr->location = -1;
300 return (Expr *) expr;
301 }
302
303 /*****************************************************************************
304 * AND clause functions
305 *****************************************************************************/
306
307
308 /*
309 * and_clause
310 *
311 * Returns t iff its argument is an 'and' clause: (AND { expr }).
312 */
313 bool
and_clause(Node * clause)314 and_clause(Node *clause)
315 {
316 return (clause != NULL &&
317 IsA(clause, BoolExpr) &&
318 ((BoolExpr *) clause)->boolop == AND_EXPR);
319 }
320
321 /*
322 * make_andclause
323 *
324 * Creates an 'and' clause given a list of its subclauses.
325 */
326 Expr *
make_andclause(List * andclauses)327 make_andclause(List *andclauses)
328 {
329 BoolExpr *expr = makeNode(BoolExpr);
330
331 expr->boolop = AND_EXPR;
332 expr->args = andclauses;
333 expr->location = -1;
334 return (Expr *) expr;
335 }
336
337 /*
338 * make_and_qual
339 *
340 * Variant of make_andclause for ANDing two qual conditions together.
341 * Qual conditions have the property that a NULL nodetree is interpreted
342 * as 'true'.
343 *
344 * NB: this makes no attempt to preserve AND/OR flatness; so it should not
345 * be used on a qual that has already been run through prepqual.c.
346 */
347 Node *
make_and_qual(Node * qual1,Node * qual2)348 make_and_qual(Node *qual1, Node *qual2)
349 {
350 if (qual1 == NULL)
351 return qual2;
352 if (qual2 == NULL)
353 return qual1;
354 return (Node *) make_andclause(list_make2(qual1, qual2));
355 }
356
357 /*
358 * Sometimes (such as in the input of ExecQual), we use lists of expression
359 * nodes with implicit AND semantics.
360 *
361 * These functions convert between an AND-semantics expression list and the
362 * ordinary representation of a boolean expression.
363 *
364 * Note that an empty list is considered equivalent to TRUE.
365 */
366 Expr *
make_ands_explicit(List * andclauses)367 make_ands_explicit(List *andclauses)
368 {
369 if (andclauses == NIL)
370 return (Expr *) makeBoolConst(true, false);
371 else if (list_length(andclauses) == 1)
372 return (Expr *) linitial(andclauses);
373 else
374 return make_andclause(andclauses);
375 }
376
377 List *
make_ands_implicit(Expr * clause)378 make_ands_implicit(Expr *clause)
379 {
380 /*
381 * NB: because the parser sets the qual field to NULL in a query that has
382 * no WHERE clause, we must consider a NULL input clause as TRUE, even
383 * though one might more reasonably think it FALSE. Grumble. If this
384 * causes trouble, consider changing the parser's behavior.
385 */
386 if (clause == NULL)
387 return NIL; /* NULL -> NIL list == TRUE */
388 else if (and_clause((Node *) clause))
389 return ((BoolExpr *) clause)->args;
390 else if (IsA(clause, Const) &&
391 !((Const *) clause)->constisnull &&
392 DatumGetBool(((Const *) clause)->constvalue))
393 return NIL; /* constant TRUE input -> NIL list */
394 else
395 return list_make1(clause);
396 }
397
398
399 /*****************************************************************************
400 * Aggregate-function clause manipulation
401 *****************************************************************************/
402
403 /*
404 * contain_agg_clause
405 * Recursively search for Aggref/GroupingFunc nodes within a clause.
406 *
407 * Returns true if any aggregate found.
408 *
409 * This does not descend into subqueries, and so should be used only after
410 * reduction of sublinks to subplans, or in contexts where it's known there
411 * are no subqueries. There mustn't be outer-aggregate references either.
412 *
413 * (If you want something like this but able to deal with subqueries,
414 * see rewriteManip.c's contain_aggs_of_level().)
415 */
416 bool
contain_agg_clause(Node * clause)417 contain_agg_clause(Node *clause)
418 {
419 return contain_agg_clause_walker(clause, NULL);
420 }
421
422 static bool
contain_agg_clause_walker(Node * node,void * context)423 contain_agg_clause_walker(Node *node, void *context)
424 {
425 if (node == NULL)
426 return false;
427 if (IsA(node, Aggref))
428 {
429 Assert(((Aggref *) node)->agglevelsup == 0);
430 return true; /* abort the tree traversal and return true */
431 }
432 if (IsA(node, GroupingFunc))
433 {
434 Assert(((GroupingFunc *) node)->agglevelsup == 0);
435 return true; /* abort the tree traversal and return true */
436 }
437 Assert(!IsA(node, SubLink));
438 return expression_tree_walker(node, contain_agg_clause_walker, context);
439 }
440
441 /*
442 * get_agg_clause_costs
443 * Recursively find the Aggref nodes in an expression tree, and
444 * accumulate cost information about them.
445 *
446 * 'aggsplit' tells us the expected partial-aggregation mode, which affects
447 * the cost estimates.
448 *
449 * NOTE that the counts/costs are ADDED to those already in *costs ... so
450 * the caller is responsible for zeroing the struct initially.
451 *
452 * We count the nodes, estimate their execution costs, and estimate the total
453 * space needed for their transition state values if all are evaluated in
454 * parallel (as would be done in a HashAgg plan). Also, we check whether
455 * partial aggregation is feasible. See AggClauseCosts for the exact set
456 * of statistics collected.
457 *
458 * In addition, we mark Aggref nodes with the correct aggtranstype, so
459 * that that doesn't need to be done repeatedly. (That makes this function's
460 * name a bit of a misnomer.)
461 *
462 * This does not descend into subqueries, and so should be used only after
463 * reduction of sublinks to subplans, or in contexts where it's known there
464 * are no subqueries. There mustn't be outer-aggregate references either.
465 */
466 void
get_agg_clause_costs(PlannerInfo * root,Node * clause,AggSplit aggsplit,AggClauseCosts * costs)467 get_agg_clause_costs(PlannerInfo *root, Node *clause, AggSplit aggsplit,
468 AggClauseCosts *costs)
469 {
470 get_agg_clause_costs_context context;
471
472 context.root = root;
473 context.aggsplit = aggsplit;
474 context.costs = costs;
475 (void) get_agg_clause_costs_walker(clause, &context);
476 }
477
478 static bool
get_agg_clause_costs_walker(Node * node,get_agg_clause_costs_context * context)479 get_agg_clause_costs_walker(Node *node, get_agg_clause_costs_context *context)
480 {
481 if (node == NULL)
482 return false;
483 if (IsA(node, Aggref))
484 {
485 Aggref *aggref = (Aggref *) node;
486 AggClauseCosts *costs = context->costs;
487 HeapTuple aggTuple;
488 Form_pg_aggregate aggform;
489 Oid aggtransfn;
490 Oid aggfinalfn;
491 Oid aggcombinefn;
492 Oid aggserialfn;
493 Oid aggdeserialfn;
494 Oid aggtranstype;
495 int32 aggtransspace;
496 QualCost argcosts;
497
498 Assert(aggref->agglevelsup == 0);
499
500 /*
501 * Fetch info about aggregate from pg_aggregate. Note it's correct to
502 * ignore the moving-aggregate variant, since what we're concerned
503 * with here is aggregates not window functions.
504 */
505 aggTuple = SearchSysCache1(AGGFNOID,
506 ObjectIdGetDatum(aggref->aggfnoid));
507 if (!HeapTupleIsValid(aggTuple))
508 elog(ERROR, "cache lookup failed for aggregate %u",
509 aggref->aggfnoid);
510 aggform = (Form_pg_aggregate) GETSTRUCT(aggTuple);
511 aggtransfn = aggform->aggtransfn;
512 aggfinalfn = aggform->aggfinalfn;
513 aggcombinefn = aggform->aggcombinefn;
514 aggserialfn = aggform->aggserialfn;
515 aggdeserialfn = aggform->aggdeserialfn;
516 aggtranstype = aggform->aggtranstype;
517 aggtransspace = aggform->aggtransspace;
518 ReleaseSysCache(aggTuple);
519
520 /*
521 * Resolve the possibly-polymorphic aggregate transition type, unless
522 * already done in a previous pass over the expression.
523 */
524 if (OidIsValid(aggref->aggtranstype))
525 aggtranstype = aggref->aggtranstype;
526 else
527 {
528 Oid inputTypes[FUNC_MAX_ARGS];
529 int numArguments;
530
531 /* extract argument types (ignoring any ORDER BY expressions) */
532 numArguments = get_aggregate_argtypes(aggref, inputTypes);
533
534 /* resolve actual type of transition state, if polymorphic */
535 aggtranstype = resolve_aggregate_transtype(aggref->aggfnoid,
536 aggtranstype,
537 inputTypes,
538 numArguments);
539 aggref->aggtranstype = aggtranstype;
540 }
541
542 /*
543 * Count it, and check for cases requiring ordered input. Note that
544 * ordered-set aggs always have nonempty aggorder. Any ordered-input
545 * case also defeats partial aggregation.
546 */
547 costs->numAggs++;
548 if (aggref->aggorder != NIL || aggref->aggdistinct != NIL)
549 {
550 costs->numOrderedAggs++;
551 costs->hasNonPartial = true;
552 }
553
554 /*
555 * Check whether partial aggregation is feasible, unless we already
556 * found out that we can't do it.
557 */
558 if (!costs->hasNonPartial)
559 {
560 /*
561 * If there is no combine function, then partial aggregation is
562 * not possible.
563 */
564 if (!OidIsValid(aggcombinefn))
565 costs->hasNonPartial = true;
566
567 /*
568 * If we have any aggs with transtype INTERNAL then we must check
569 * whether they have serialization/deserialization functions; if
570 * not, we can't serialize partial-aggregation results.
571 */
572 else if (aggtranstype == INTERNALOID &&
573 (!OidIsValid(aggserialfn) || !OidIsValid(aggdeserialfn)))
574 costs->hasNonSerial = true;
575 }
576
577 /*
578 * Add the appropriate component function execution costs to
579 * appropriate totals.
580 */
581 if (DO_AGGSPLIT_COMBINE(context->aggsplit))
582 {
583 /* charge for combining previously aggregated states */
584 costs->transCost.per_tuple += get_func_cost(aggcombinefn) * cpu_operator_cost;
585 }
586 else
587 costs->transCost.per_tuple += get_func_cost(aggtransfn) * cpu_operator_cost;
588 if (DO_AGGSPLIT_DESERIALIZE(context->aggsplit) &&
589 OidIsValid(aggdeserialfn))
590 costs->transCost.per_tuple += get_func_cost(aggdeserialfn) * cpu_operator_cost;
591 if (DO_AGGSPLIT_SERIALIZE(context->aggsplit) &&
592 OidIsValid(aggserialfn))
593 costs->finalCost += get_func_cost(aggserialfn) * cpu_operator_cost;
594 if (!DO_AGGSPLIT_SKIPFINAL(context->aggsplit) &&
595 OidIsValid(aggfinalfn))
596 costs->finalCost += get_func_cost(aggfinalfn) * cpu_operator_cost;
597
598 /*
599 * These costs are incurred only by the initial aggregate node, so we
600 * mustn't include them again at upper levels.
601 */
602 if (!DO_AGGSPLIT_COMBINE(context->aggsplit))
603 {
604 /* add the input expressions' cost to per-input-row costs */
605 cost_qual_eval_node(&argcosts, (Node *) aggref->args, context->root);
606 costs->transCost.startup += argcosts.startup;
607 costs->transCost.per_tuple += argcosts.per_tuple;
608
609 /*
610 * Add any filter's cost to per-input-row costs.
611 *
612 * XXX Ideally we should reduce input expression costs according
613 * to filter selectivity, but it's not clear it's worth the
614 * trouble.
615 */
616 if (aggref->aggfilter)
617 {
618 cost_qual_eval_node(&argcosts, (Node *) aggref->aggfilter,
619 context->root);
620 costs->transCost.startup += argcosts.startup;
621 costs->transCost.per_tuple += argcosts.per_tuple;
622 }
623 }
624
625 /*
626 * If there are direct arguments, treat their evaluation cost like the
627 * cost of the finalfn.
628 */
629 if (aggref->aggdirectargs)
630 {
631 cost_qual_eval_node(&argcosts, (Node *) aggref->aggdirectargs,
632 context->root);
633 costs->transCost.startup += argcosts.startup;
634 costs->finalCost += argcosts.per_tuple;
635 }
636
637 /*
638 * If the transition type is pass-by-value then it doesn't add
639 * anything to the required size of the hashtable. If it is
640 * pass-by-reference then we have to add the estimated size of the
641 * value itself, plus palloc overhead.
642 */
643 if (!get_typbyval(aggtranstype))
644 {
645 int32 avgwidth;
646
647 /* Use average width if aggregate definition gave one */
648 if (aggtransspace > 0)
649 avgwidth = aggtransspace;
650 else if (aggtransfn == F_ARRAY_APPEND)
651 {
652 /*
653 * If the transition function is array_append(), it'll use an
654 * expanded array as transvalue, which will occupy at least
655 * ALLOCSET_SMALL_INITSIZE and possibly more. Use that as the
656 * estimate for lack of a better idea.
657 */
658 avgwidth = ALLOCSET_SMALL_INITSIZE;
659 }
660 else
661 {
662 /*
663 * If transition state is of same type as first aggregated
664 * input, assume it's the same typmod (same width) as well.
665 * This works for cases like MAX/MIN and is probably somewhat
666 * reasonable otherwise.
667 */
668 int32 aggtranstypmod = -1;
669
670 if (aggref->args)
671 {
672 TargetEntry *tle = (TargetEntry *) linitial(aggref->args);
673
674 if (aggtranstype == exprType((Node *) tle->expr))
675 aggtranstypmod = exprTypmod((Node *) tle->expr);
676 }
677
678 avgwidth = get_typavgwidth(aggtranstype, aggtranstypmod);
679 }
680
681 avgwidth = MAXALIGN(avgwidth);
682 costs->transitionSpace += avgwidth + 2 * sizeof(void *);
683 }
684 else if (aggtranstype == INTERNALOID)
685 {
686 /*
687 * INTERNAL transition type is a special case: although INTERNAL
688 * is pass-by-value, it's almost certainly being used as a pointer
689 * to some large data structure. The aggregate definition can
690 * provide an estimate of the size. If it doesn't, then we assume
691 * ALLOCSET_DEFAULT_INITSIZE, which is a good guess if the data is
692 * being kept in a private memory context, as is done by
693 * array_agg() for instance.
694 */
695 if (aggtransspace > 0)
696 costs->transitionSpace += aggtransspace;
697 else
698 costs->transitionSpace += ALLOCSET_DEFAULT_INITSIZE;
699 }
700
701 /*
702 * We assume that the parser checked that there are no aggregates (of
703 * this level anyway) in the aggregated arguments, direct arguments,
704 * or filter clause. Hence, we need not recurse into any of them.
705 */
706 return false;
707 }
708 Assert(!IsA(node, SubLink));
709 return expression_tree_walker(node, get_agg_clause_costs_walker,
710 (void *) context);
711 }
712
713
714 /*****************************************************************************
715 * Window-function clause manipulation
716 *****************************************************************************/
717
718 /*
719 * contain_window_function
720 * Recursively search for WindowFunc nodes within a clause.
721 *
722 * Since window functions don't have level fields, but are hard-wired to
723 * be associated with the current query level, this is just the same as
724 * rewriteManip.c's function.
725 */
726 bool
contain_window_function(Node * clause)727 contain_window_function(Node *clause)
728 {
729 return contain_windowfuncs(clause);
730 }
731
732 /*
733 * find_window_functions
734 * Locate all the WindowFunc nodes in an expression tree, and organize
735 * them by winref ID number.
736 *
737 * Caller must provide an upper bound on the winref IDs expected in the tree.
738 */
739 WindowFuncLists *
find_window_functions(Node * clause,Index maxWinRef)740 find_window_functions(Node *clause, Index maxWinRef)
741 {
742 WindowFuncLists *lists = palloc(sizeof(WindowFuncLists));
743
744 lists->numWindowFuncs = 0;
745 lists->maxWinRef = maxWinRef;
746 lists->windowFuncs = (List **) palloc0((maxWinRef + 1) * sizeof(List *));
747 (void) find_window_functions_walker(clause, lists);
748 return lists;
749 }
750
751 static bool
find_window_functions_walker(Node * node,WindowFuncLists * lists)752 find_window_functions_walker(Node *node, WindowFuncLists *lists)
753 {
754 if (node == NULL)
755 return false;
756 if (IsA(node, WindowFunc))
757 {
758 WindowFunc *wfunc = (WindowFunc *) node;
759
760 /* winref is unsigned, so one-sided test is OK */
761 if (wfunc->winref > lists->maxWinRef)
762 elog(ERROR, "WindowFunc contains out-of-range winref %u",
763 wfunc->winref);
764 /* eliminate duplicates, so that we avoid repeated computation */
765 if (!list_member(lists->windowFuncs[wfunc->winref], wfunc))
766 {
767 lists->windowFuncs[wfunc->winref] =
768 lappend(lists->windowFuncs[wfunc->winref], wfunc);
769 lists->numWindowFuncs++;
770 }
771
772 /*
773 * We assume that the parser checked that there are no window
774 * functions in the arguments or filter clause. Hence, we need not
775 * recurse into them. (If either the parser or the planner screws up
776 * on this point, the executor will still catch it; see ExecInitExpr.)
777 */
778 return false;
779 }
780 Assert(!IsA(node, SubLink));
781 return expression_tree_walker(node, find_window_functions_walker,
782 (void *) lists);
783 }
784
785
786 /*****************************************************************************
787 * Support for expressions returning sets
788 *****************************************************************************/
789
790 /*
791 * expression_returns_set_rows
792 * Estimate the number of rows returned by a set-returning expression.
793 * The result is 1 if there are no set-returning functions.
794 *
795 * We use the product of the rowcount estimates of all the functions in
796 * the given tree (this corresponds to the behavior of ExecMakeFunctionResult
797 * for nested set-returning functions).
798 *
799 * Note: keep this in sync with expression_returns_set() in nodes/nodeFuncs.c.
800 */
801 double
expression_returns_set_rows(Node * clause)802 expression_returns_set_rows(Node *clause)
803 {
804 double result = 1;
805
806 (void) expression_returns_set_rows_walker(clause, &result);
807 return clamp_row_est(result);
808 }
809
810 static bool
expression_returns_set_rows_walker(Node * node,double * count)811 expression_returns_set_rows_walker(Node *node, double *count)
812 {
813 if (node == NULL)
814 return false;
815 if (IsA(node, FuncExpr))
816 {
817 FuncExpr *expr = (FuncExpr *) node;
818
819 if (expr->funcretset)
820 *count *= get_func_rows(expr->funcid);
821 }
822 if (IsA(node, OpExpr))
823 {
824 OpExpr *expr = (OpExpr *) node;
825
826 if (expr->opretset)
827 {
828 set_opfuncid(expr);
829 *count *= get_func_rows(expr->opfuncid);
830 }
831 }
832
833 /* Avoid recursion for some cases that can't return a set */
834 if (IsA(node, Aggref))
835 return false;
836 if (IsA(node, WindowFunc))
837 return false;
838 if (IsA(node, DistinctExpr))
839 return false;
840 if (IsA(node, NullIfExpr))
841 return false;
842 if (IsA(node, ScalarArrayOpExpr))
843 return false;
844 if (IsA(node, BoolExpr))
845 return false;
846 if (IsA(node, SubLink))
847 return false;
848 if (IsA(node, SubPlan))
849 return false;
850 if (IsA(node, AlternativeSubPlan))
851 return false;
852 if (IsA(node, ArrayExpr))
853 return false;
854 if (IsA(node, RowExpr))
855 return false;
856 if (IsA(node, RowCompareExpr))
857 return false;
858 if (IsA(node, CoalesceExpr))
859 return false;
860 if (IsA(node, MinMaxExpr))
861 return false;
862 if (IsA(node, XmlExpr))
863 return false;
864
865 return expression_tree_walker(node, expression_returns_set_rows_walker,
866 (void *) count);
867 }
868
869 /*
870 * tlist_returns_set_rows
871 * Estimate the number of rows returned by a set-returning targetlist.
872 * The result is 1 if there are no set-returning functions.
873 *
874 * Here, the result is the largest rowcount estimate of any of the tlist's
875 * expressions, not the product as you would get from naively applying
876 * expression_returns_set_rows() to the whole tlist. The behavior actually
877 * implemented by ExecTargetList produces a number of rows equal to the least
878 * common multiple of the expression rowcounts, so that the product would be
879 * a worst-case estimate that is typically not realistic. Taking the max as
880 * we do here is a best-case estimate that might not be realistic either,
881 * but it's probably closer for typical usages. We don't try to compute the
882 * actual LCM because we're working with very approximate estimates, so their
883 * LCM would be unduly noisy.
884 */
885 double
tlist_returns_set_rows(List * tlist)886 tlist_returns_set_rows(List *tlist)
887 {
888 double result = 1;
889 ListCell *lc;
890
891 foreach(lc, tlist)
892 {
893 TargetEntry *tle = (TargetEntry *) lfirst(lc);
894 double colresult;
895
896 colresult = expression_returns_set_rows((Node *) tle->expr);
897 if (result < colresult)
898 result = colresult;
899 }
900 return result;
901 }
902
903
904 /*****************************************************************************
905 * Subplan clause manipulation
906 *****************************************************************************/
907
908 /*
909 * contain_subplans
910 * Recursively search for subplan nodes within a clause.
911 *
912 * If we see a SubLink node, we will return TRUE. This is only possible if
913 * the expression tree hasn't yet been transformed by subselect.c. We do not
914 * know whether the node will produce a true subplan or just an initplan,
915 * but we make the conservative assumption that it will be a subplan.
916 *
917 * Returns true if any subplan found.
918 */
919 bool
contain_subplans(Node * clause)920 contain_subplans(Node *clause)
921 {
922 return contain_subplans_walker(clause, NULL);
923 }
924
925 static bool
contain_subplans_walker(Node * node,void * context)926 contain_subplans_walker(Node *node, void *context)
927 {
928 if (node == NULL)
929 return false;
930 if (IsA(node, SubPlan) ||
931 IsA(node, AlternativeSubPlan) ||
932 IsA(node, SubLink))
933 return true; /* abort the tree traversal and return true */
934 return expression_tree_walker(node, contain_subplans_walker, context);
935 }
936
937
938 /*****************************************************************************
939 * Check clauses for mutable functions
940 *****************************************************************************/
941
942 /*
943 * contain_mutable_functions
944 * Recursively search for mutable functions within a clause.
945 *
946 * Returns true if any mutable function (or operator implemented by a
947 * mutable function) is found. This test is needed so that we don't
948 * mistakenly think that something like "WHERE random() < 0.5" can be treated
949 * as a constant qualification.
950 *
951 * We will recursively look into Query nodes (i.e., SubLink sub-selects)
952 * but not into SubPlans. See comments for contain_volatile_functions().
953 */
954 bool
contain_mutable_functions(Node * clause)955 contain_mutable_functions(Node *clause)
956 {
957 return contain_mutable_functions_walker(clause, NULL);
958 }
959
960 static bool
contain_mutable_functions_checker(Oid func_id,void * context)961 contain_mutable_functions_checker(Oid func_id, void *context)
962 {
963 return (func_volatile(func_id) != PROVOLATILE_IMMUTABLE);
964 }
965
966 static bool
contain_mutable_functions_walker(Node * node,void * context)967 contain_mutable_functions_walker(Node *node, void *context)
968 {
969 if (node == NULL)
970 return false;
971 /* Check for mutable functions in node itself */
972 if (check_functions_in_node(node, contain_mutable_functions_checker,
973 context))
974 return true;
975
976 /*
977 * It should be safe to treat MinMaxExpr as immutable, because it will
978 * depend on a non-cross-type btree comparison function, and those should
979 * always be immutable. Treating XmlExpr as immutable is more dubious,
980 * and treating CoerceToDomain as immutable is outright dangerous. But we
981 * have done so historically, and changing this would probably cause more
982 * problems than it would fix. In practice, if you have a non-immutable
983 * domain constraint you are in for pain anyhow.
984 */
985
986 /* Recurse to check arguments */
987 if (IsA(node, Query))
988 {
989 /* Recurse into subselects */
990 return query_tree_walker((Query *) node,
991 contain_mutable_functions_walker,
992 context, 0);
993 }
994 return expression_tree_walker(node, contain_mutable_functions_walker,
995 context);
996 }
997
998
999 /*****************************************************************************
1000 * Check clauses for volatile functions
1001 *****************************************************************************/
1002
1003 /*
1004 * contain_volatile_functions
1005 * Recursively search for volatile functions within a clause.
1006 *
1007 * Returns true if any volatile function (or operator implemented by a
1008 * volatile function) is found. This test prevents, for example,
1009 * invalid conversions of volatile expressions into indexscan quals.
1010 *
1011 * We will recursively look into Query nodes (i.e., SubLink sub-selects)
1012 * but not into SubPlans. This is a bit odd, but intentional. If we are
1013 * looking at a SubLink, we are probably deciding whether a query tree
1014 * transformation is safe, and a contained sub-select should affect that;
1015 * for example, duplicating a sub-select containing a volatile function
1016 * would be bad. However, once we've got to the stage of having SubPlans,
1017 * subsequent planning need not consider volatility within those, since
1018 * the executor won't change its evaluation rules for a SubPlan based on
1019 * volatility.
1020 */
1021 bool
contain_volatile_functions(Node * clause)1022 contain_volatile_functions(Node *clause)
1023 {
1024 return contain_volatile_functions_walker(clause, NULL);
1025 }
1026
1027 static bool
contain_volatile_functions_checker(Oid func_id,void * context)1028 contain_volatile_functions_checker(Oid func_id, void *context)
1029 {
1030 return (func_volatile(func_id) == PROVOLATILE_VOLATILE);
1031 }
1032
1033 static bool
contain_volatile_functions_walker(Node * node,void * context)1034 contain_volatile_functions_walker(Node *node, void *context)
1035 {
1036 if (node == NULL)
1037 return false;
1038 /* Check for volatile functions in node itself */
1039 if (check_functions_in_node(node, contain_volatile_functions_checker,
1040 context))
1041 return true;
1042
1043 /*
1044 * See notes in contain_mutable_functions_walker about why we treat
1045 * MinMaxExpr, XmlExpr, and CoerceToDomain as immutable.
1046 */
1047
1048 /* Recurse to check arguments */
1049 if (IsA(node, Query))
1050 {
1051 /* Recurse into subselects */
1052 return query_tree_walker((Query *) node,
1053 contain_volatile_functions_walker,
1054 context, 0);
1055 }
1056 return expression_tree_walker(node, contain_volatile_functions_walker,
1057 context);
1058 }
1059
1060 /*
1061 * Special purpose version of contain_volatile_functions() for use in COPY:
1062 * ignore nextval(), but treat all other functions normally.
1063 */
1064 bool
contain_volatile_functions_not_nextval(Node * clause)1065 contain_volatile_functions_not_nextval(Node *clause)
1066 {
1067 return contain_volatile_functions_not_nextval_walker(clause, NULL);
1068 }
1069
1070 static bool
contain_volatile_functions_not_nextval_checker(Oid func_id,void * context)1071 contain_volatile_functions_not_nextval_checker(Oid func_id, void *context)
1072 {
1073 return (func_id != F_NEXTVAL_OID &&
1074 func_volatile(func_id) == PROVOLATILE_VOLATILE);
1075 }
1076
1077 static bool
contain_volatile_functions_not_nextval_walker(Node * node,void * context)1078 contain_volatile_functions_not_nextval_walker(Node *node, void *context)
1079 {
1080 if (node == NULL)
1081 return false;
1082 /* Check for volatile functions in node itself */
1083 if (check_functions_in_node(node,
1084 contain_volatile_functions_not_nextval_checker,
1085 context))
1086 return true;
1087
1088 /*
1089 * See notes in contain_mutable_functions_walker about why we treat
1090 * MinMaxExpr, XmlExpr, and CoerceToDomain as immutable.
1091 */
1092
1093 /* Recurse to check arguments */
1094 if (IsA(node, Query))
1095 {
1096 /* Recurse into subselects */
1097 return query_tree_walker((Query *) node,
1098 contain_volatile_functions_not_nextval_walker,
1099 context, 0);
1100 }
1101 return expression_tree_walker(node,
1102 contain_volatile_functions_not_nextval_walker,
1103 context);
1104 }
1105
1106 /*****************************************************************************
1107 * Check queries for parallel unsafe and/or restricted constructs
1108 *****************************************************************************/
1109
1110 /*
1111 * Check whether a node tree contains parallel hazards. This is used both on
1112 * the entire query tree, to see whether the query can be parallelized at all
1113 * (with allow_restricted = true), and also to evaluate whether a particular
1114 * expression is safe to run within a parallel worker (with allow_restricted =
1115 * false). We could separate these concerns into two different functions, but
1116 * there's enough overlap that it doesn't seem worthwhile.
1117 */
1118 bool
has_parallel_hazard(Node * node,bool allow_restricted)1119 has_parallel_hazard(Node *node, bool allow_restricted)
1120 {
1121 has_parallel_hazard_arg context;
1122
1123 context.allow_restricted = allow_restricted;
1124 return has_parallel_hazard_walker(node, &context);
1125 }
1126
1127 static bool
has_parallel_hazard_checker(Oid func_id,void * context)1128 has_parallel_hazard_checker(Oid func_id, void *context)
1129 {
1130 char proparallel = func_parallel(func_id);
1131
1132 if (((has_parallel_hazard_arg *) context)->allow_restricted)
1133 return (proparallel == PROPARALLEL_UNSAFE);
1134 else
1135 return (proparallel != PROPARALLEL_SAFE);
1136 }
1137
1138 static bool
has_parallel_hazard_walker(Node * node,has_parallel_hazard_arg * context)1139 has_parallel_hazard_walker(Node *node, has_parallel_hazard_arg *context)
1140 {
1141 if (node == NULL)
1142 return false;
1143
1144 /* Check for hazardous functions in node itself */
1145 if (check_functions_in_node(node, has_parallel_hazard_checker,
1146 context))
1147 return true;
1148
1149 /*
1150 * It should be OK to treat MinMaxExpr as parallel-safe, since btree
1151 * opclass support functions are generally parallel-safe. XmlExpr is a
1152 * bit more dubious but we can probably get away with it. We err on the
1153 * side of caution by treating CoerceToDomain as parallel-restricted.
1154 * (Note: in principle that's wrong because a domain constraint could
1155 * contain a parallel-unsafe function; but useful constraints probably
1156 * never would have such, and assuming they do would cripple use of
1157 * parallel query in the presence of domain types.)
1158 */
1159 if (IsA(node, CoerceToDomain))
1160 {
1161 if (!context->allow_restricted)
1162 return true;
1163 }
1164
1165 /*
1166 * Treat window functions as parallel-restricted because we aren't sure
1167 * whether the input row ordering is fully deterministic, and the output
1168 * of window functions might vary across workers if not. (In some cases,
1169 * like where the window frame orders by a primary key, we could relax
1170 * this restriction. But it doesn't currently seem worth expending extra
1171 * effort to do so.)
1172 */
1173 else if (IsA(node, WindowFunc))
1174 {
1175 if (!context->allow_restricted)
1176 return true;
1177 }
1178
1179 /*
1180 * As a notational convenience for callers, look through RestrictInfo.
1181 */
1182 else if (IsA(node, RestrictInfo))
1183 {
1184 RestrictInfo *rinfo = (RestrictInfo *) node;
1185
1186 return has_parallel_hazard_walker((Node *) rinfo->clause, context);
1187 }
1188
1189 /*
1190 * Since we don't have the ability to push subplans down to workers at
1191 * present, we treat subplan references as parallel-restricted. We need
1192 * not worry about examining their contents; if they are unsafe, we would
1193 * have found that out while examining the whole tree before reduction of
1194 * sublinks to subplans. (Really we should not see SubLink during a
1195 * not-allow_restricted scan, but if we do, return true.)
1196 */
1197 else if (IsA(node, SubLink) ||
1198 IsA(node, SubPlan) ||
1199 IsA(node, AlternativeSubPlan))
1200 {
1201 if (!context->allow_restricted)
1202 return true;
1203 }
1204
1205 /*
1206 * We can't pass Params to workers at the moment either, so they are also
1207 * parallel-restricted.
1208 */
1209 else if (IsA(node, Param))
1210 {
1211 if (!context->allow_restricted)
1212 return true;
1213 }
1214
1215 /*
1216 * When we're first invoked on a completely unplanned tree, we must
1217 * recurse into subqueries so to as to locate parallel-unsafe constructs
1218 * anywhere in the tree.
1219 */
1220 else if (IsA(node, Query))
1221 {
1222 Query *query = (Query *) node;
1223
1224 /* SELECT FOR UPDATE/SHARE must be treated as unsafe */
1225 if (query->rowMarks != NULL)
1226 return true;
1227
1228 /* Recurse into subselects */
1229 return query_tree_walker(query,
1230 has_parallel_hazard_walker,
1231 context, 0);
1232 }
1233
1234 /* Recurse to check arguments */
1235 return expression_tree_walker(node,
1236 has_parallel_hazard_walker,
1237 context);
1238 }
1239
1240 /*****************************************************************************
1241 * Check clauses for nonstrict functions
1242 *****************************************************************************/
1243
1244 /*
1245 * contain_nonstrict_functions
1246 * Recursively search for nonstrict functions within a clause.
1247 *
1248 * Returns true if any nonstrict construct is found --- ie, anything that
1249 * could produce non-NULL output with a NULL input.
1250 *
1251 * The idea here is that the caller has verified that the expression contains
1252 * one or more Var or Param nodes (as appropriate for the caller's need), and
1253 * now wishes to prove that the expression result will be NULL if any of these
1254 * inputs is NULL. If we return false, then the proof succeeded.
1255 */
1256 bool
contain_nonstrict_functions(Node * clause)1257 contain_nonstrict_functions(Node *clause)
1258 {
1259 return contain_nonstrict_functions_walker(clause, NULL);
1260 }
1261
1262 static bool
contain_nonstrict_functions_checker(Oid func_id,void * context)1263 contain_nonstrict_functions_checker(Oid func_id, void *context)
1264 {
1265 return !func_strict(func_id);
1266 }
1267
1268 static bool
contain_nonstrict_functions_walker(Node * node,void * context)1269 contain_nonstrict_functions_walker(Node *node, void *context)
1270 {
1271 if (node == NULL)
1272 return false;
1273 if (IsA(node, Aggref))
1274 {
1275 /* an aggregate could return non-null with null input */
1276 return true;
1277 }
1278 if (IsA(node, GroupingFunc))
1279 {
1280 /*
1281 * A GroupingFunc doesn't evaluate its arguments, and therefore must
1282 * be treated as nonstrict.
1283 */
1284 return true;
1285 }
1286 if (IsA(node, WindowFunc))
1287 {
1288 /* a window function could return non-null with null input */
1289 return true;
1290 }
1291 if (IsA(node, ArrayRef))
1292 {
1293 /* array assignment is nonstrict, but subscripting is strict */
1294 if (((ArrayRef *) node)->refassgnexpr != NULL)
1295 return true;
1296 /* else fall through to check args */
1297 }
1298 if (IsA(node, DistinctExpr))
1299 {
1300 /* IS DISTINCT FROM is inherently non-strict */
1301 return true;
1302 }
1303 if (IsA(node, NullIfExpr))
1304 {
1305 /* NULLIF is inherently non-strict */
1306 return true;
1307 }
1308 if (IsA(node, BoolExpr))
1309 {
1310 BoolExpr *expr = (BoolExpr *) node;
1311
1312 switch (expr->boolop)
1313 {
1314 case AND_EXPR:
1315 case OR_EXPR:
1316 /* AND, OR are inherently non-strict */
1317 return true;
1318 default:
1319 break;
1320 }
1321 }
1322 if (IsA(node, SubLink))
1323 {
1324 /* In some cases a sublink might be strict, but in general not */
1325 return true;
1326 }
1327 if (IsA(node, SubPlan))
1328 return true;
1329 if (IsA(node, AlternativeSubPlan))
1330 return true;
1331 if (IsA(node, FieldStore))
1332 return true;
1333 if (IsA(node, CaseExpr))
1334 return true;
1335 if (IsA(node, ArrayExpr))
1336 return true;
1337 if (IsA(node, RowExpr))
1338 return true;
1339 if (IsA(node, RowCompareExpr))
1340 return true;
1341 if (IsA(node, CoalesceExpr))
1342 return true;
1343 if (IsA(node, MinMaxExpr))
1344 return true;
1345 if (IsA(node, XmlExpr))
1346 return true;
1347 if (IsA(node, NullTest))
1348 return true;
1349 if (IsA(node, BooleanTest))
1350 return true;
1351
1352 /*
1353 * Check other function-containing nodes; but ArrayCoerceExpr is strict at
1354 * the array level, regardless of elemfunc.
1355 */
1356 if (!IsA(node, ArrayCoerceExpr) &&
1357 check_functions_in_node(node, contain_nonstrict_functions_checker,
1358 context))
1359 return true;
1360 return expression_tree_walker(node, contain_nonstrict_functions_walker,
1361 context);
1362 }
1363
1364 /*****************************************************************************
1365 * Check clauses for Params
1366 *****************************************************************************/
1367
1368 /*
1369 * contain_exec_param
1370 * Recursively search for PARAM_EXEC Params within a clause.
1371 *
1372 * Returns true if the clause contains any PARAM_EXEC Param with a paramid
1373 * appearing in the given list of Param IDs. Does not descend into
1374 * subqueries!
1375 */
1376 bool
contain_exec_param(Node * clause,List * param_ids)1377 contain_exec_param(Node *clause, List *param_ids)
1378 {
1379 return contain_exec_param_walker(clause, param_ids);
1380 }
1381
1382 static bool
contain_exec_param_walker(Node * node,List * param_ids)1383 contain_exec_param_walker(Node *node, List *param_ids)
1384 {
1385 if (node == NULL)
1386 return false;
1387 if (IsA(node, Param))
1388 {
1389 Param *p = (Param *) node;
1390
1391 if (p->paramkind == PARAM_EXEC &&
1392 list_member_int(param_ids, p->paramid))
1393 return true;
1394 }
1395 return expression_tree_walker(node, contain_exec_param_walker, param_ids);
1396 }
1397
1398 /*****************************************************************************
1399 * Check clauses for context-dependent nodes
1400 *****************************************************************************/
1401
1402 /*
1403 * contain_context_dependent_node
1404 * Recursively search for context-dependent nodes within a clause.
1405 *
1406 * CaseTestExpr nodes must appear directly within the corresponding CaseExpr,
1407 * not nested within another one, or they'll see the wrong test value. If one
1408 * appears "bare" in the arguments of a SQL function, then we can't inline the
1409 * SQL function for fear of creating such a situation.
1410 *
1411 * CoerceToDomainValue would have the same issue if domain CHECK expressions
1412 * could get inlined into larger expressions, but presently that's impossible.
1413 * Still, it might be allowed in future, or other node types with similar
1414 * issues might get invented. So give this function a generic name, and set
1415 * up the recursion state to allow multiple flag bits.
1416 */
1417 static bool
contain_context_dependent_node(Node * clause)1418 contain_context_dependent_node(Node *clause)
1419 {
1420 int flags = 0;
1421
1422 return contain_context_dependent_node_walker(clause, &flags);
1423 }
1424
1425 #define CCDN_IN_CASEEXPR 0x0001 /* CaseTestExpr okay here? */
1426
1427 static bool
contain_context_dependent_node_walker(Node * node,int * flags)1428 contain_context_dependent_node_walker(Node *node, int *flags)
1429 {
1430 if (node == NULL)
1431 return false;
1432 if (IsA(node, CaseTestExpr))
1433 return !(*flags & CCDN_IN_CASEEXPR);
1434 if (IsA(node, CaseExpr))
1435 {
1436 CaseExpr *caseexpr = (CaseExpr *) node;
1437
1438 /*
1439 * If this CASE doesn't have a test expression, then it doesn't create
1440 * a context in which CaseTestExprs should appear, so just fall
1441 * through and treat it as a generic expression node.
1442 */
1443 if (caseexpr->arg)
1444 {
1445 int save_flags = *flags;
1446 bool res;
1447
1448 /*
1449 * Note: in principle, we could distinguish the various sub-parts
1450 * of a CASE construct and set the flag bit only for some of them,
1451 * since we are only expecting CaseTestExprs to appear in the
1452 * "expr" subtree of the CaseWhen nodes. But it doesn't really
1453 * seem worth any extra code. If there are any bare CaseTestExprs
1454 * elsewhere in the CASE, something's wrong already.
1455 */
1456 *flags |= CCDN_IN_CASEEXPR;
1457 res = expression_tree_walker(node,
1458 contain_context_dependent_node_walker,
1459 (void *) flags);
1460 *flags = save_flags;
1461 return res;
1462 }
1463 }
1464 return expression_tree_walker(node, contain_context_dependent_node_walker,
1465 (void *) flags);
1466 }
1467
1468 /*****************************************************************************
1469 * Check clauses for Vars passed to non-leakproof functions
1470 *****************************************************************************/
1471
1472 /*
1473 * contain_leaked_vars
1474 * Recursively scan a clause to discover whether it contains any Var
1475 * nodes (of the current query level) that are passed as arguments to
1476 * leaky functions.
1477 *
1478 * Returns true if the clause contains any non-leakproof functions that are
1479 * passed Var nodes of the current query level, and which might therefore leak
1480 * data. Qualifiers from outside a security_barrier view that might leak data
1481 * in this way should not be pushed down into the view in case the contents of
1482 * tuples intended to be filtered out by the view are revealed by the leaky
1483 * functions.
1484 */
1485 bool
contain_leaked_vars(Node * clause)1486 contain_leaked_vars(Node *clause)
1487 {
1488 return contain_leaked_vars_walker(clause, NULL);
1489 }
1490
1491 static bool
contain_leaked_vars_checker(Oid func_id,void * context)1492 contain_leaked_vars_checker(Oid func_id, void *context)
1493 {
1494 return !get_func_leakproof(func_id);
1495 }
1496
1497 static bool
contain_leaked_vars_walker(Node * node,void * context)1498 contain_leaked_vars_walker(Node *node, void *context)
1499 {
1500 if (node == NULL)
1501 return false;
1502
1503 switch (nodeTag(node))
1504 {
1505 case T_Var:
1506 case T_Const:
1507 case T_Param:
1508 case T_ArrayExpr:
1509 case T_FieldSelect:
1510 case T_FieldStore:
1511 case T_NamedArgExpr:
1512 case T_BoolExpr:
1513 case T_RelabelType:
1514 case T_CollateExpr:
1515 case T_CaseExpr:
1516 case T_CaseTestExpr:
1517 case T_RowExpr:
1518 case T_NullTest:
1519 case T_BooleanTest:
1520 case T_List:
1521
1522 /*
1523 * We know these node types don't contain function calls; but
1524 * something further down in the node tree might.
1525 */
1526 break;
1527
1528 case T_FuncExpr:
1529 case T_OpExpr:
1530 case T_DistinctExpr:
1531 case T_NullIfExpr:
1532 case T_ScalarArrayOpExpr:
1533 case T_CoerceViaIO:
1534 case T_ArrayCoerceExpr:
1535
1536 /*
1537 * If node contains a leaky function call, and there's any Var
1538 * underneath it, reject.
1539 */
1540 if (check_functions_in_node(node, contain_leaked_vars_checker,
1541 context) &&
1542 contain_var_clause(node))
1543 return true;
1544 break;
1545
1546 case T_ArrayRef:
1547 {
1548 ArrayRef *aref = (ArrayRef *) node;
1549
1550 /*
1551 * array assignment is leaky, but subscripted fetches
1552 * are not
1553 */
1554 if (aref->refassgnexpr != NULL)
1555 {
1556 /* Node is leaky, so reject if it contains Vars */
1557 if (contain_var_clause(node))
1558 return true;
1559 }
1560 }
1561 break;
1562
1563 case T_RowCompareExpr:
1564 {
1565 /*
1566 * It's worth special-casing this because a leaky comparison
1567 * function only compromises one pair of row elements, which
1568 * might not contain Vars while others do.
1569 */
1570 RowCompareExpr *rcexpr = (RowCompareExpr *) node;
1571 ListCell *opid;
1572 ListCell *larg;
1573 ListCell *rarg;
1574
1575 forthree(opid, rcexpr->opnos,
1576 larg, rcexpr->largs,
1577 rarg, rcexpr->rargs)
1578 {
1579 Oid funcid = get_opcode(lfirst_oid(opid));
1580
1581 if (!get_func_leakproof(funcid) &&
1582 (contain_var_clause((Node *) lfirst(larg)) ||
1583 contain_var_clause((Node *) lfirst(rarg))))
1584 return true;
1585 }
1586 }
1587 break;
1588
1589 case T_MinMaxExpr:
1590 {
1591 /*
1592 * MinMaxExpr is leakproof if the comparison function it calls
1593 * is leakproof.
1594 */
1595 MinMaxExpr *minmaxexpr = (MinMaxExpr *) node;
1596 TypeCacheEntry *typentry;
1597 bool leakproof;
1598
1599 /* Look up the btree comparison function for the datatype */
1600 typentry = lookup_type_cache(minmaxexpr->minmaxtype,
1601 TYPECACHE_CMP_PROC);
1602 if (OidIsValid(typentry->cmp_proc))
1603 leakproof = get_func_leakproof(typentry->cmp_proc);
1604 else
1605 {
1606 /*
1607 * The executor will throw an error, but here we just
1608 * treat the missing function as leaky.
1609 */
1610 leakproof = false;
1611 }
1612
1613 if (!leakproof &&
1614 contain_var_clause((Node *) minmaxexpr->args))
1615 return true;
1616 }
1617 break;
1618
1619 case T_CurrentOfExpr:
1620
1621 /*
1622 * WHERE CURRENT OF doesn't contain function calls. Moreover, it
1623 * is important that this can be pushed down into a
1624 * security_barrier view, since the planner must always generate a
1625 * TID scan when CURRENT OF is present -- c.f. cost_tidscan.
1626 */
1627 return false;
1628
1629 default:
1630
1631 /*
1632 * If we don't recognize the node tag, assume it might be leaky.
1633 * This prevents an unexpected security hole if someone adds a new
1634 * node type that can call a function.
1635 */
1636 return true;
1637 }
1638 return expression_tree_walker(node, contain_leaked_vars_walker,
1639 context);
1640 }
1641
1642 /*
1643 * find_nonnullable_rels
1644 * Determine which base rels are forced nonnullable by given clause.
1645 *
1646 * Returns the set of all Relids that are referenced in the clause in such
1647 * a way that the clause cannot possibly return TRUE if any of these Relids
1648 * is an all-NULL row. (It is OK to err on the side of conservatism; hence
1649 * the analysis here is simplistic.)
1650 *
1651 * The semantics here are subtly different from contain_nonstrict_functions:
1652 * that function is concerned with NULL results from arbitrary expressions,
1653 * but here we assume that the input is a Boolean expression, and wish to
1654 * see if NULL inputs will provably cause a FALSE-or-NULL result. We expect
1655 * the expression to have been AND/OR flattened and converted to implicit-AND
1656 * format.
1657 *
1658 * Note: this function is largely duplicative of find_nonnullable_vars().
1659 * The reason not to simplify this function into a thin wrapper around
1660 * find_nonnullable_vars() is that the tested conditions really are different:
1661 * a clause like "t1.v1 IS NOT NULL OR t1.v2 IS NOT NULL" does not prove
1662 * that either v1 or v2 can't be NULL, but it does prove that the t1 row
1663 * as a whole can't be all-NULL.
1664 *
1665 * top_level is TRUE while scanning top-level AND/OR structure; here, showing
1666 * the result is either FALSE or NULL is good enough. top_level is FALSE when
1667 * we have descended below a NOT or a strict function: now we must be able to
1668 * prove that the subexpression goes to NULL.
1669 *
1670 * We don't use expression_tree_walker here because we don't want to descend
1671 * through very many kinds of nodes; only the ones we can be sure are strict.
1672 */
1673 Relids
find_nonnullable_rels(Node * clause)1674 find_nonnullable_rels(Node *clause)
1675 {
1676 return find_nonnullable_rels_walker(clause, true);
1677 }
1678
1679 static Relids
find_nonnullable_rels_walker(Node * node,bool top_level)1680 find_nonnullable_rels_walker(Node *node, bool top_level)
1681 {
1682 Relids result = NULL;
1683 ListCell *l;
1684
1685 if (node == NULL)
1686 return NULL;
1687 if (IsA(node, Var))
1688 {
1689 Var *var = (Var *) node;
1690
1691 if (var->varlevelsup == 0)
1692 result = bms_make_singleton(var->varno);
1693 }
1694 else if (IsA(node, List))
1695 {
1696 /*
1697 * At top level, we are examining an implicit-AND list: if any of the
1698 * arms produces FALSE-or-NULL then the result is FALSE-or-NULL. If
1699 * not at top level, we are examining the arguments of a strict
1700 * function: if any of them produce NULL then the result of the
1701 * function must be NULL. So in both cases, the set of nonnullable
1702 * rels is the union of those found in the arms, and we pass down the
1703 * top_level flag unmodified.
1704 */
1705 foreach(l, (List *) node)
1706 {
1707 result = bms_join(result,
1708 find_nonnullable_rels_walker(lfirst(l),
1709 top_level));
1710 }
1711 }
1712 else if (IsA(node, FuncExpr))
1713 {
1714 FuncExpr *expr = (FuncExpr *) node;
1715
1716 if (func_strict(expr->funcid))
1717 result = find_nonnullable_rels_walker((Node *) expr->args, false);
1718 }
1719 else if (IsA(node, OpExpr))
1720 {
1721 OpExpr *expr = (OpExpr *) node;
1722
1723 set_opfuncid(expr);
1724 if (func_strict(expr->opfuncid))
1725 result = find_nonnullable_rels_walker((Node *) expr->args, false);
1726 }
1727 else if (IsA(node, ScalarArrayOpExpr))
1728 {
1729 ScalarArrayOpExpr *expr = (ScalarArrayOpExpr *) node;
1730
1731 if (is_strict_saop(expr, true))
1732 result = find_nonnullable_rels_walker((Node *) expr->args, false);
1733 }
1734 else if (IsA(node, BoolExpr))
1735 {
1736 BoolExpr *expr = (BoolExpr *) node;
1737
1738 switch (expr->boolop)
1739 {
1740 case AND_EXPR:
1741 /* At top level we can just recurse (to the List case) */
1742 if (top_level)
1743 {
1744 result = find_nonnullable_rels_walker((Node *) expr->args,
1745 top_level);
1746 break;
1747 }
1748
1749 /*
1750 * Below top level, even if one arm produces NULL, the result
1751 * could be FALSE (hence not NULL). However, if *all* the
1752 * arms produce NULL then the result is NULL, so we can take
1753 * the intersection of the sets of nonnullable rels, just as
1754 * for OR. Fall through to share code.
1755 */
1756 /* FALL THRU */
1757 case OR_EXPR:
1758
1759 /*
1760 * OR is strict if all of its arms are, so we can take the
1761 * intersection of the sets of nonnullable rels for each arm.
1762 * This works for both values of top_level.
1763 */
1764 foreach(l, expr->args)
1765 {
1766 Relids subresult;
1767
1768 subresult = find_nonnullable_rels_walker(lfirst(l),
1769 top_level);
1770 if (result == NULL) /* first subresult? */
1771 result = subresult;
1772 else
1773 result = bms_int_members(result, subresult);
1774
1775 /*
1776 * If the intersection is empty, we can stop looking. This
1777 * also justifies the test for first-subresult above.
1778 */
1779 if (bms_is_empty(result))
1780 break;
1781 }
1782 break;
1783 case NOT_EXPR:
1784 /* NOT will return null if its arg is null */
1785 result = find_nonnullable_rels_walker((Node *) expr->args,
1786 false);
1787 break;
1788 default:
1789 elog(ERROR, "unrecognized boolop: %d", (int) expr->boolop);
1790 break;
1791 }
1792 }
1793 else if (IsA(node, RelabelType))
1794 {
1795 RelabelType *expr = (RelabelType *) node;
1796
1797 result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
1798 }
1799 else if (IsA(node, CoerceViaIO))
1800 {
1801 /* not clear this is useful, but it can't hurt */
1802 CoerceViaIO *expr = (CoerceViaIO *) node;
1803
1804 result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
1805 }
1806 else if (IsA(node, ArrayCoerceExpr))
1807 {
1808 /* ArrayCoerceExpr is strict at the array level */
1809 ArrayCoerceExpr *expr = (ArrayCoerceExpr *) node;
1810
1811 result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
1812 }
1813 else if (IsA(node, ConvertRowtypeExpr))
1814 {
1815 /* not clear this is useful, but it can't hurt */
1816 ConvertRowtypeExpr *expr = (ConvertRowtypeExpr *) node;
1817
1818 result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
1819 }
1820 else if (IsA(node, CollateExpr))
1821 {
1822 CollateExpr *expr = (CollateExpr *) node;
1823
1824 result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
1825 }
1826 else if (IsA(node, NullTest))
1827 {
1828 /* IS NOT NULL can be considered strict, but only at top level */
1829 NullTest *expr = (NullTest *) node;
1830
1831 if (top_level && expr->nulltesttype == IS_NOT_NULL && !expr->argisrow)
1832 result = find_nonnullable_rels_walker((Node *) expr->arg, false);
1833 }
1834 else if (IsA(node, BooleanTest))
1835 {
1836 /* Boolean tests that reject NULL are strict at top level */
1837 BooleanTest *expr = (BooleanTest *) node;
1838
1839 if (top_level &&
1840 (expr->booltesttype == IS_TRUE ||
1841 expr->booltesttype == IS_FALSE ||
1842 expr->booltesttype == IS_NOT_UNKNOWN))
1843 result = find_nonnullable_rels_walker((Node *) expr->arg, false);
1844 }
1845 else if (IsA(node, PlaceHolderVar))
1846 {
1847 PlaceHolderVar *phv = (PlaceHolderVar *) node;
1848
1849 result = find_nonnullable_rels_walker((Node *) phv->phexpr, top_level);
1850 }
1851 return result;
1852 }
1853
1854 /*
1855 * find_nonnullable_vars
1856 * Determine which Vars are forced nonnullable by given clause.
1857 *
1858 * Returns a list of all level-zero Vars that are referenced in the clause in
1859 * such a way that the clause cannot possibly return TRUE if any of these Vars
1860 * is NULL. (It is OK to err on the side of conservatism; hence the analysis
1861 * here is simplistic.)
1862 *
1863 * The semantics here are subtly different from contain_nonstrict_functions:
1864 * that function is concerned with NULL results from arbitrary expressions,
1865 * but here we assume that the input is a Boolean expression, and wish to
1866 * see if NULL inputs will provably cause a FALSE-or-NULL result. We expect
1867 * the expression to have been AND/OR flattened and converted to implicit-AND
1868 * format.
1869 *
1870 * The result is a palloc'd List, but we have not copied the member Var nodes.
1871 * Also, we don't bother trying to eliminate duplicate entries.
1872 *
1873 * top_level is TRUE while scanning top-level AND/OR structure; here, showing
1874 * the result is either FALSE or NULL is good enough. top_level is FALSE when
1875 * we have descended below a NOT or a strict function: now we must be able to
1876 * prove that the subexpression goes to NULL.
1877 *
1878 * We don't use expression_tree_walker here because we don't want to descend
1879 * through very many kinds of nodes; only the ones we can be sure are strict.
1880 */
1881 List *
find_nonnullable_vars(Node * clause)1882 find_nonnullable_vars(Node *clause)
1883 {
1884 return find_nonnullable_vars_walker(clause, true);
1885 }
1886
1887 static List *
find_nonnullable_vars_walker(Node * node,bool top_level)1888 find_nonnullable_vars_walker(Node *node, bool top_level)
1889 {
1890 List *result = NIL;
1891 ListCell *l;
1892
1893 if (node == NULL)
1894 return NIL;
1895 if (IsA(node, Var))
1896 {
1897 Var *var = (Var *) node;
1898
1899 if (var->varlevelsup == 0)
1900 result = list_make1(var);
1901 }
1902 else if (IsA(node, List))
1903 {
1904 /*
1905 * At top level, we are examining an implicit-AND list: if any of the
1906 * arms produces FALSE-or-NULL then the result is FALSE-or-NULL. If
1907 * not at top level, we are examining the arguments of a strict
1908 * function: if any of them produce NULL then the result of the
1909 * function must be NULL. So in both cases, the set of nonnullable
1910 * vars is the union of those found in the arms, and we pass down the
1911 * top_level flag unmodified.
1912 */
1913 foreach(l, (List *) node)
1914 {
1915 result = list_concat(result,
1916 find_nonnullable_vars_walker(lfirst(l),
1917 top_level));
1918 }
1919 }
1920 else if (IsA(node, FuncExpr))
1921 {
1922 FuncExpr *expr = (FuncExpr *) node;
1923
1924 if (func_strict(expr->funcid))
1925 result = find_nonnullable_vars_walker((Node *) expr->args, false);
1926 }
1927 else if (IsA(node, OpExpr))
1928 {
1929 OpExpr *expr = (OpExpr *) node;
1930
1931 set_opfuncid(expr);
1932 if (func_strict(expr->opfuncid))
1933 result = find_nonnullable_vars_walker((Node *) expr->args, false);
1934 }
1935 else if (IsA(node, ScalarArrayOpExpr))
1936 {
1937 ScalarArrayOpExpr *expr = (ScalarArrayOpExpr *) node;
1938
1939 if (is_strict_saop(expr, true))
1940 result = find_nonnullable_vars_walker((Node *) expr->args, false);
1941 }
1942 else if (IsA(node, BoolExpr))
1943 {
1944 BoolExpr *expr = (BoolExpr *) node;
1945
1946 switch (expr->boolop)
1947 {
1948 case AND_EXPR:
1949 /* At top level we can just recurse (to the List case) */
1950 if (top_level)
1951 {
1952 result = find_nonnullable_vars_walker((Node *) expr->args,
1953 top_level);
1954 break;
1955 }
1956
1957 /*
1958 * Below top level, even if one arm produces NULL, the result
1959 * could be FALSE (hence not NULL). However, if *all* the
1960 * arms produce NULL then the result is NULL, so we can take
1961 * the intersection of the sets of nonnullable vars, just as
1962 * for OR. Fall through to share code.
1963 */
1964 /* FALL THRU */
1965 case OR_EXPR:
1966
1967 /*
1968 * OR is strict if all of its arms are, so we can take the
1969 * intersection of the sets of nonnullable vars for each arm.
1970 * This works for both values of top_level.
1971 */
1972 foreach(l, expr->args)
1973 {
1974 List *subresult;
1975
1976 subresult = find_nonnullable_vars_walker(lfirst(l),
1977 top_level);
1978 if (result == NIL) /* first subresult? */
1979 result = subresult;
1980 else
1981 result = list_intersection(result, subresult);
1982
1983 /*
1984 * If the intersection is empty, we can stop looking. This
1985 * also justifies the test for first-subresult above.
1986 */
1987 if (result == NIL)
1988 break;
1989 }
1990 break;
1991 case NOT_EXPR:
1992 /* NOT will return null if its arg is null */
1993 result = find_nonnullable_vars_walker((Node *) expr->args,
1994 false);
1995 break;
1996 default:
1997 elog(ERROR, "unrecognized boolop: %d", (int) expr->boolop);
1998 break;
1999 }
2000 }
2001 else if (IsA(node, RelabelType))
2002 {
2003 RelabelType *expr = (RelabelType *) node;
2004
2005 result = find_nonnullable_vars_walker((Node *) expr->arg, top_level);
2006 }
2007 else if (IsA(node, CoerceViaIO))
2008 {
2009 /* not clear this is useful, but it can't hurt */
2010 CoerceViaIO *expr = (CoerceViaIO *) node;
2011
2012 result = find_nonnullable_vars_walker((Node *) expr->arg, false);
2013 }
2014 else if (IsA(node, ArrayCoerceExpr))
2015 {
2016 /* ArrayCoerceExpr is strict at the array level */
2017 ArrayCoerceExpr *expr = (ArrayCoerceExpr *) node;
2018
2019 result = find_nonnullable_vars_walker((Node *) expr->arg, top_level);
2020 }
2021 else if (IsA(node, ConvertRowtypeExpr))
2022 {
2023 /* not clear this is useful, but it can't hurt */
2024 ConvertRowtypeExpr *expr = (ConvertRowtypeExpr *) node;
2025
2026 result = find_nonnullable_vars_walker((Node *) expr->arg, top_level);
2027 }
2028 else if (IsA(node, CollateExpr))
2029 {
2030 CollateExpr *expr = (CollateExpr *) node;
2031
2032 result = find_nonnullable_vars_walker((Node *) expr->arg, top_level);
2033 }
2034 else if (IsA(node, NullTest))
2035 {
2036 /* IS NOT NULL can be considered strict, but only at top level */
2037 NullTest *expr = (NullTest *) node;
2038
2039 if (top_level && expr->nulltesttype == IS_NOT_NULL && !expr->argisrow)
2040 result = find_nonnullable_vars_walker((Node *) expr->arg, false);
2041 }
2042 else if (IsA(node, BooleanTest))
2043 {
2044 /* Boolean tests that reject NULL are strict at top level */
2045 BooleanTest *expr = (BooleanTest *) node;
2046
2047 if (top_level &&
2048 (expr->booltesttype == IS_TRUE ||
2049 expr->booltesttype == IS_FALSE ||
2050 expr->booltesttype == IS_NOT_UNKNOWN))
2051 result = find_nonnullable_vars_walker((Node *) expr->arg, false);
2052 }
2053 else if (IsA(node, PlaceHolderVar))
2054 {
2055 PlaceHolderVar *phv = (PlaceHolderVar *) node;
2056
2057 result = find_nonnullable_vars_walker((Node *) phv->phexpr, top_level);
2058 }
2059 return result;
2060 }
2061
2062 /*
2063 * find_forced_null_vars
2064 * Determine which Vars must be NULL for the given clause to return TRUE.
2065 *
2066 * This is the complement of find_nonnullable_vars: find the level-zero Vars
2067 * that must be NULL for the clause to return TRUE. (It is OK to err on the
2068 * side of conservatism; hence the analysis here is simplistic. In fact,
2069 * we only detect simple "var IS NULL" tests at the top level.)
2070 *
2071 * The result is a palloc'd List, but we have not copied the member Var nodes.
2072 * Also, we don't bother trying to eliminate duplicate entries.
2073 */
2074 List *
find_forced_null_vars(Node * node)2075 find_forced_null_vars(Node *node)
2076 {
2077 List *result = NIL;
2078 Var *var;
2079 ListCell *l;
2080
2081 if (node == NULL)
2082 return NIL;
2083 /* Check single-clause cases using subroutine */
2084 var = find_forced_null_var(node);
2085 if (var)
2086 {
2087 result = list_make1(var);
2088 }
2089 /* Otherwise, handle AND-conditions */
2090 else if (IsA(node, List))
2091 {
2092 /*
2093 * At top level, we are examining an implicit-AND list: if any of the
2094 * arms produces FALSE-or-NULL then the result is FALSE-or-NULL.
2095 */
2096 foreach(l, (List *) node)
2097 {
2098 result = list_concat(result,
2099 find_forced_null_vars(lfirst(l)));
2100 }
2101 }
2102 else if (IsA(node, BoolExpr))
2103 {
2104 BoolExpr *expr = (BoolExpr *) node;
2105
2106 /*
2107 * We don't bother considering the OR case, because it's fairly
2108 * unlikely anyone would write "v1 IS NULL OR v1 IS NULL". Likewise,
2109 * the NOT case isn't worth expending code on.
2110 */
2111 if (expr->boolop == AND_EXPR)
2112 {
2113 /* At top level we can just recurse (to the List case) */
2114 result = find_forced_null_vars((Node *) expr->args);
2115 }
2116 }
2117 return result;
2118 }
2119
2120 /*
2121 * find_forced_null_var
2122 * Return the Var forced null by the given clause, or NULL if it's
2123 * not an IS NULL-type clause. For success, the clause must enforce
2124 * *only* nullness of the particular Var, not any other conditions.
2125 *
2126 * This is just the single-clause case of find_forced_null_vars(), without
2127 * any allowance for AND conditions. It's used by initsplan.c on individual
2128 * qual clauses. The reason for not just applying find_forced_null_vars()
2129 * is that if an AND of an IS NULL clause with something else were to somehow
2130 * survive AND/OR flattening, initsplan.c might get fooled into discarding
2131 * the whole clause when only the IS NULL part of it had been proved redundant.
2132 */
2133 Var *
find_forced_null_var(Node * node)2134 find_forced_null_var(Node *node)
2135 {
2136 if (node == NULL)
2137 return NULL;
2138 if (IsA(node, NullTest))
2139 {
2140 /* check for var IS NULL */
2141 NullTest *expr = (NullTest *) node;
2142
2143 if (expr->nulltesttype == IS_NULL && !expr->argisrow)
2144 {
2145 Var *var = (Var *) expr->arg;
2146
2147 if (var && IsA(var, Var) &&
2148 var->varlevelsup == 0)
2149 return var;
2150 }
2151 }
2152 else if (IsA(node, BooleanTest))
2153 {
2154 /* var IS UNKNOWN is equivalent to var IS NULL */
2155 BooleanTest *expr = (BooleanTest *) node;
2156
2157 if (expr->booltesttype == IS_UNKNOWN)
2158 {
2159 Var *var = (Var *) expr->arg;
2160
2161 if (var && IsA(var, Var) &&
2162 var->varlevelsup == 0)
2163 return var;
2164 }
2165 }
2166 return NULL;
2167 }
2168
2169 /*
2170 * Can we treat a ScalarArrayOpExpr as strict?
2171 *
2172 * If "falseOK" is true, then a "false" result can be considered strict,
2173 * else we need to guarantee an actual NULL result for NULL input.
2174 *
2175 * "foo op ALL array" is strict if the op is strict *and* we can prove
2176 * that the array input isn't an empty array. We can check that
2177 * for the cases of an array constant and an ARRAY[] construct.
2178 *
2179 * "foo op ANY array" is strict in the falseOK sense if the op is strict.
2180 * If not falseOK, the test is the same as for "foo op ALL array".
2181 */
2182 static bool
is_strict_saop(ScalarArrayOpExpr * expr,bool falseOK)2183 is_strict_saop(ScalarArrayOpExpr *expr, bool falseOK)
2184 {
2185 Node *rightop;
2186
2187 /* The contained operator must be strict. */
2188 set_sa_opfuncid(expr);
2189 if (!func_strict(expr->opfuncid))
2190 return false;
2191 /* If ANY and falseOK, that's all we need to check. */
2192 if (expr->useOr && falseOK)
2193 return true;
2194 /* Else, we have to see if the array is provably non-empty. */
2195 Assert(list_length(expr->args) == 2);
2196 rightop = (Node *) lsecond(expr->args);
2197 if (rightop && IsA(rightop, Const))
2198 {
2199 Datum arraydatum = ((Const *) rightop)->constvalue;
2200 bool arrayisnull = ((Const *) rightop)->constisnull;
2201 ArrayType *arrayval;
2202 int nitems;
2203
2204 if (arrayisnull)
2205 return false;
2206 arrayval = DatumGetArrayTypeP(arraydatum);
2207 nitems = ArrayGetNItems(ARR_NDIM(arrayval), ARR_DIMS(arrayval));
2208 if (nitems > 0)
2209 return true;
2210 }
2211 else if (rightop && IsA(rightop, ArrayExpr))
2212 {
2213 ArrayExpr *arrayexpr = (ArrayExpr *) rightop;
2214
2215 if (arrayexpr->elements != NIL && !arrayexpr->multidims)
2216 return true;
2217 }
2218 return false;
2219 }
2220
2221
2222 /*****************************************************************************
2223 * Check for "pseudo-constant" clauses
2224 *****************************************************************************/
2225
2226 /*
2227 * is_pseudo_constant_clause
2228 * Detect whether an expression is "pseudo constant", ie, it contains no
2229 * variables of the current query level and no uses of volatile functions.
2230 * Such an expr is not necessarily a true constant: it can still contain
2231 * Params and outer-level Vars, not to mention functions whose results
2232 * may vary from one statement to the next. However, the expr's value
2233 * will be constant over any one scan of the current query, so it can be
2234 * used as, eg, an indexscan key.
2235 *
2236 * CAUTION: this function omits to test for one very important class of
2237 * not-constant expressions, namely aggregates (Aggrefs). In current usage
2238 * this is only applied to WHERE clauses and so a check for Aggrefs would be
2239 * a waste of cycles; but be sure to also check contain_agg_clause() if you
2240 * want to know about pseudo-constness in other contexts. The same goes
2241 * for window functions (WindowFuncs).
2242 */
2243 bool
is_pseudo_constant_clause(Node * clause)2244 is_pseudo_constant_clause(Node *clause)
2245 {
2246 /*
2247 * We could implement this check in one recursive scan. But since the
2248 * check for volatile functions is both moderately expensive and unlikely
2249 * to fail, it seems better to look for Vars first and only check for
2250 * volatile functions if we find no Vars.
2251 */
2252 if (!contain_var_clause(clause) &&
2253 !contain_volatile_functions(clause))
2254 return true;
2255 return false;
2256 }
2257
2258 /*
2259 * is_pseudo_constant_clause_relids
2260 * Same as above, except caller already has available the var membership
2261 * of the expression; this lets us avoid the contain_var_clause() scan.
2262 */
2263 bool
is_pseudo_constant_clause_relids(Node * clause,Relids relids)2264 is_pseudo_constant_clause_relids(Node *clause, Relids relids)
2265 {
2266 if (bms_is_empty(relids) &&
2267 !contain_volatile_functions(clause))
2268 return true;
2269 return false;
2270 }
2271
2272
2273 /*****************************************************************************
2274 * *
2275 * General clause-manipulating routines *
2276 * *
2277 *****************************************************************************/
2278
2279 /*
2280 * NumRelids
2281 * (formerly clause_relids)
2282 *
2283 * Returns the number of different relations referenced in 'clause'.
2284 */
2285 int
NumRelids(Node * clause)2286 NumRelids(Node *clause)
2287 {
2288 Relids varnos = pull_varnos(clause);
2289 int result = bms_num_members(varnos);
2290
2291 bms_free(varnos);
2292 return result;
2293 }
2294
2295 /*
2296 * CommuteOpExpr: commute a binary operator clause
2297 *
2298 * XXX the clause is destructively modified!
2299 */
2300 void
CommuteOpExpr(OpExpr * clause)2301 CommuteOpExpr(OpExpr *clause)
2302 {
2303 Oid opoid;
2304 Node *temp;
2305
2306 /* Sanity checks: caller is at fault if these fail */
2307 if (!is_opclause(clause) ||
2308 list_length(clause->args) != 2)
2309 elog(ERROR, "cannot commute non-binary-operator clause");
2310
2311 opoid = get_commutator(clause->opno);
2312
2313 if (!OidIsValid(opoid))
2314 elog(ERROR, "could not find commutator for operator %u",
2315 clause->opno);
2316
2317 /*
2318 * modify the clause in-place!
2319 */
2320 clause->opno = opoid;
2321 clause->opfuncid = InvalidOid;
2322 /* opresulttype, opretset, opcollid, inputcollid need not change */
2323
2324 temp = linitial(clause->args);
2325 linitial(clause->args) = lsecond(clause->args);
2326 lsecond(clause->args) = temp;
2327 }
2328
2329 /*
2330 * CommuteRowCompareExpr: commute a RowCompareExpr clause
2331 *
2332 * XXX the clause is destructively modified!
2333 */
2334 void
CommuteRowCompareExpr(RowCompareExpr * clause)2335 CommuteRowCompareExpr(RowCompareExpr *clause)
2336 {
2337 List *newops;
2338 List *temp;
2339 ListCell *l;
2340
2341 /* Sanity checks: caller is at fault if these fail */
2342 if (!IsA(clause, RowCompareExpr))
2343 elog(ERROR, "expected a RowCompareExpr");
2344
2345 /* Build list of commuted operators */
2346 newops = NIL;
2347 foreach(l, clause->opnos)
2348 {
2349 Oid opoid = lfirst_oid(l);
2350
2351 opoid = get_commutator(opoid);
2352 if (!OidIsValid(opoid))
2353 elog(ERROR, "could not find commutator for operator %u",
2354 lfirst_oid(l));
2355 newops = lappend_oid(newops, opoid);
2356 }
2357
2358 /*
2359 * modify the clause in-place!
2360 */
2361 switch (clause->rctype)
2362 {
2363 case ROWCOMPARE_LT:
2364 clause->rctype = ROWCOMPARE_GT;
2365 break;
2366 case ROWCOMPARE_LE:
2367 clause->rctype = ROWCOMPARE_GE;
2368 break;
2369 case ROWCOMPARE_GE:
2370 clause->rctype = ROWCOMPARE_LE;
2371 break;
2372 case ROWCOMPARE_GT:
2373 clause->rctype = ROWCOMPARE_LT;
2374 break;
2375 default:
2376 elog(ERROR, "unexpected RowCompare type: %d",
2377 (int) clause->rctype);
2378 break;
2379 }
2380
2381 clause->opnos = newops;
2382
2383 /*
2384 * Note: we need not change the opfamilies list; we assume any btree
2385 * opfamily containing an operator will also contain its commutator.
2386 * Collations don't change either.
2387 */
2388
2389 temp = clause->largs;
2390 clause->largs = clause->rargs;
2391 clause->rargs = temp;
2392 }
2393
2394 /*
2395 * Helper for eval_const_expressions: check that datatype of an attribute
2396 * is still what it was when the expression was parsed. This is needed to
2397 * guard against improper simplification after ALTER COLUMN TYPE. (XXX we
2398 * may well need to make similar checks elsewhere?)
2399 */
2400 static bool
rowtype_field_matches(Oid rowtypeid,int fieldnum,Oid expectedtype,int32 expectedtypmod,Oid expectedcollation)2401 rowtype_field_matches(Oid rowtypeid, int fieldnum,
2402 Oid expectedtype, int32 expectedtypmod,
2403 Oid expectedcollation)
2404 {
2405 TupleDesc tupdesc;
2406 Form_pg_attribute attr;
2407
2408 /* No issue for RECORD, since there is no way to ALTER such a type */
2409 if (rowtypeid == RECORDOID)
2410 return true;
2411 tupdesc = lookup_rowtype_tupdesc(rowtypeid, -1);
2412 if (fieldnum <= 0 || fieldnum > tupdesc->natts)
2413 {
2414 ReleaseTupleDesc(tupdesc);
2415 return false;
2416 }
2417 attr = tupdesc->attrs[fieldnum - 1];
2418 if (attr->attisdropped ||
2419 attr->atttypid != expectedtype ||
2420 attr->atttypmod != expectedtypmod ||
2421 attr->attcollation != expectedcollation)
2422 {
2423 ReleaseTupleDesc(tupdesc);
2424 return false;
2425 }
2426 ReleaseTupleDesc(tupdesc);
2427 return true;
2428 }
2429
2430
2431 /*--------------------
2432 * eval_const_expressions
2433 *
2434 * Reduce any recognizably constant subexpressions of the given
2435 * expression tree, for example "2 + 2" => "4". More interestingly,
2436 * we can reduce certain boolean expressions even when they contain
2437 * non-constant subexpressions: "x OR true" => "true" no matter what
2438 * the subexpression x is. (XXX We assume that no such subexpression
2439 * will have important side-effects, which is not necessarily a good
2440 * assumption in the presence of user-defined functions; do we need a
2441 * pg_proc flag that prevents discarding the execution of a function?)
2442 *
2443 * We do understand that certain functions may deliver non-constant
2444 * results even with constant inputs, "nextval()" being the classic
2445 * example. Functions that are not marked "immutable" in pg_proc
2446 * will not be pre-evaluated here, although we will reduce their
2447 * arguments as far as possible.
2448 *
2449 * Whenever a function is eliminated from the expression by means of
2450 * constant-expression evaluation or inlining, we add the function to
2451 * root->glob->invalItems. This ensures the plan is known to depend on
2452 * such functions, even though they aren't referenced anymore.
2453 *
2454 * We assume that the tree has already been type-checked and contains
2455 * only operators and functions that are reasonable to try to execute.
2456 *
2457 * NOTE: "root" can be passed as NULL if the caller never wants to do any
2458 * Param substitutions nor receive info about inlined functions.
2459 *
2460 * NOTE: the planner assumes that this will always flatten nested AND and
2461 * OR clauses into N-argument form. See comments in prepqual.c.
2462 *
2463 * NOTE: another critical effect is that any function calls that require
2464 * default arguments will be expanded, and named-argument calls will be
2465 * converted to positional notation. The executor won't handle either.
2466 *--------------------
2467 */
2468 Node *
eval_const_expressions(PlannerInfo * root,Node * node)2469 eval_const_expressions(PlannerInfo *root, Node *node)
2470 {
2471 eval_const_expressions_context context;
2472
2473 if (root)
2474 context.boundParams = root->glob->boundParams; /* bound Params */
2475 else
2476 context.boundParams = NULL;
2477 context.root = root; /* for inlined-function dependencies */
2478 context.active_fns = NIL; /* nothing being recursively simplified */
2479 context.case_val = NULL; /* no CASE being examined */
2480 context.estimate = false; /* safe transformations only */
2481 return eval_const_expressions_mutator(node, &context);
2482 }
2483
2484 /*--------------------
2485 * estimate_expression_value
2486 *
2487 * This function attempts to estimate the value of an expression for
2488 * planning purposes. It is in essence a more aggressive version of
2489 * eval_const_expressions(): we will perform constant reductions that are
2490 * not necessarily 100% safe, but are reasonable for estimation purposes.
2491 *
2492 * Currently the extra steps that are taken in this mode are:
2493 * 1. Substitute values for Params, where a bound Param value has been made
2494 * available by the caller of planner(), even if the Param isn't marked
2495 * constant. This effectively means that we plan using the first supplied
2496 * value of the Param.
2497 * 2. Fold stable, as well as immutable, functions to constants.
2498 * 3. Reduce PlaceHolderVar nodes to their contained expressions.
2499 *--------------------
2500 */
2501 Node *
estimate_expression_value(PlannerInfo * root,Node * node)2502 estimate_expression_value(PlannerInfo *root, Node *node)
2503 {
2504 eval_const_expressions_context context;
2505
2506 context.boundParams = root->glob->boundParams; /* bound Params */
2507 /* we do not need to mark the plan as depending on inlined functions */
2508 context.root = NULL;
2509 context.active_fns = NIL; /* nothing being recursively simplified */
2510 context.case_val = NULL; /* no CASE being examined */
2511 context.estimate = true; /* unsafe transformations OK */
2512 return eval_const_expressions_mutator(node, &context);
2513 }
2514
2515 static Node *
eval_const_expressions_mutator(Node * node,eval_const_expressions_context * context)2516 eval_const_expressions_mutator(Node *node,
2517 eval_const_expressions_context *context)
2518 {
2519 if (node == NULL)
2520 return NULL;
2521 switch (nodeTag(node))
2522 {
2523 case T_Param:
2524 {
2525 Param *param = (Param *) node;
2526
2527 /* Look to see if we've been given a value for this Param */
2528 if (param->paramkind == PARAM_EXTERN &&
2529 context->boundParams != NULL &&
2530 param->paramid > 0 &&
2531 param->paramid <= context->boundParams->numParams)
2532 {
2533 ParamExternData *prm = &context->boundParams->params[param->paramid - 1];
2534
2535 if (OidIsValid(prm->ptype))
2536 {
2537 /* OK to substitute parameter value? */
2538 if (context->estimate ||
2539 (prm->pflags & PARAM_FLAG_CONST))
2540 {
2541 /*
2542 * Return a Const representing the param value.
2543 * Must copy pass-by-ref datatypes, since the
2544 * Param might be in a memory context
2545 * shorter-lived than our output plan should be.
2546 */
2547 int16 typLen;
2548 bool typByVal;
2549 Datum pval;
2550
2551 Assert(prm->ptype == param->paramtype);
2552 get_typlenbyval(param->paramtype,
2553 &typLen, &typByVal);
2554 if (prm->isnull || typByVal)
2555 pval = prm->value;
2556 else
2557 pval = datumCopy(prm->value, typByVal, typLen);
2558 return (Node *) makeConst(param->paramtype,
2559 param->paramtypmod,
2560 param->paramcollid,
2561 (int) typLen,
2562 pval,
2563 prm->isnull,
2564 typByVal);
2565 }
2566 }
2567 }
2568
2569 /*
2570 * Not replaceable, so just copy the Param (no need to
2571 * recurse)
2572 */
2573 return (Node *) copyObject(param);
2574 }
2575 case T_WindowFunc:
2576 {
2577 WindowFunc *expr = (WindowFunc *) node;
2578 Oid funcid = expr->winfnoid;
2579 List *args;
2580 Expr *aggfilter;
2581 HeapTuple func_tuple;
2582 WindowFunc *newexpr;
2583
2584 /*
2585 * We can't really simplify a WindowFunc node, but we mustn't
2586 * just fall through to the default processing, because we
2587 * have to apply expand_function_arguments to its argument
2588 * list. That takes care of inserting default arguments and
2589 * expanding named-argument notation.
2590 */
2591 func_tuple = SearchSysCache1(PROCOID, ObjectIdGetDatum(funcid));
2592 if (!HeapTupleIsValid(func_tuple))
2593 elog(ERROR, "cache lookup failed for function %u", funcid);
2594
2595 args = expand_function_arguments(expr->args, expr->wintype,
2596 func_tuple);
2597
2598 ReleaseSysCache(func_tuple);
2599
2600 /* Now, recursively simplify the args (which are a List) */
2601 args = (List *)
2602 expression_tree_mutator((Node *) args,
2603 eval_const_expressions_mutator,
2604 (void *) context);
2605 /* ... and the filter expression, which isn't */
2606 aggfilter = (Expr *)
2607 eval_const_expressions_mutator((Node *) expr->aggfilter,
2608 context);
2609
2610 /* And build the replacement WindowFunc node */
2611 newexpr = makeNode(WindowFunc);
2612 newexpr->winfnoid = expr->winfnoid;
2613 newexpr->wintype = expr->wintype;
2614 newexpr->wincollid = expr->wincollid;
2615 newexpr->inputcollid = expr->inputcollid;
2616 newexpr->args = args;
2617 newexpr->aggfilter = aggfilter;
2618 newexpr->winref = expr->winref;
2619 newexpr->winstar = expr->winstar;
2620 newexpr->winagg = expr->winagg;
2621 newexpr->location = expr->location;
2622
2623 return (Node *) newexpr;
2624 }
2625 case T_FuncExpr:
2626 {
2627 FuncExpr *expr = (FuncExpr *) node;
2628 List *args = expr->args;
2629 Expr *simple;
2630 FuncExpr *newexpr;
2631
2632 /*
2633 * Code for op/func reduction is pretty bulky, so split it out
2634 * as a separate function. Note: exprTypmod normally returns
2635 * -1 for a FuncExpr, but not when the node is recognizably a
2636 * length coercion; we want to preserve the typmod in the
2637 * eventual Const if so.
2638 */
2639 simple = simplify_function(expr->funcid,
2640 expr->funcresulttype,
2641 exprTypmod(node),
2642 expr->funccollid,
2643 expr->inputcollid,
2644 &args,
2645 expr->funcvariadic,
2646 true,
2647 true,
2648 context);
2649 if (simple) /* successfully simplified it */
2650 return (Node *) simple;
2651
2652 /*
2653 * The expression cannot be simplified any further, so build
2654 * and return a replacement FuncExpr node using the
2655 * possibly-simplified arguments. Note that we have also
2656 * converted the argument list to positional notation.
2657 */
2658 newexpr = makeNode(FuncExpr);
2659 newexpr->funcid = expr->funcid;
2660 newexpr->funcresulttype = expr->funcresulttype;
2661 newexpr->funcretset = expr->funcretset;
2662 newexpr->funcvariadic = expr->funcvariadic;
2663 newexpr->funcformat = expr->funcformat;
2664 newexpr->funccollid = expr->funccollid;
2665 newexpr->inputcollid = expr->inputcollid;
2666 newexpr->args = args;
2667 newexpr->location = expr->location;
2668 return (Node *) newexpr;
2669 }
2670 case T_OpExpr:
2671 {
2672 OpExpr *expr = (OpExpr *) node;
2673 List *args = expr->args;
2674 Expr *simple;
2675 OpExpr *newexpr;
2676
2677 /*
2678 * Need to get OID of underlying function. Okay to scribble
2679 * on input to this extent.
2680 */
2681 set_opfuncid(expr);
2682
2683 /*
2684 * Code for op/func reduction is pretty bulky, so split it out
2685 * as a separate function.
2686 */
2687 simple = simplify_function(expr->opfuncid,
2688 expr->opresulttype, -1,
2689 expr->opcollid,
2690 expr->inputcollid,
2691 &args,
2692 false,
2693 true,
2694 true,
2695 context);
2696 if (simple) /* successfully simplified it */
2697 return (Node *) simple;
2698
2699 /*
2700 * If the operator is boolean equality or inequality, we know
2701 * how to simplify cases involving one constant and one
2702 * non-constant argument.
2703 */
2704 if (expr->opno == BooleanEqualOperator ||
2705 expr->opno == BooleanNotEqualOperator)
2706 {
2707 simple = (Expr *) simplify_boolean_equality(expr->opno,
2708 args);
2709 if (simple) /* successfully simplified it */
2710 return (Node *) simple;
2711 }
2712
2713 /*
2714 * The expression cannot be simplified any further, so build
2715 * and return a replacement OpExpr node using the
2716 * possibly-simplified arguments.
2717 */
2718 newexpr = makeNode(OpExpr);
2719 newexpr->opno = expr->opno;
2720 newexpr->opfuncid = expr->opfuncid;
2721 newexpr->opresulttype = expr->opresulttype;
2722 newexpr->opretset = expr->opretset;
2723 newexpr->opcollid = expr->opcollid;
2724 newexpr->inputcollid = expr->inputcollid;
2725 newexpr->args = args;
2726 newexpr->location = expr->location;
2727 return (Node *) newexpr;
2728 }
2729 case T_DistinctExpr:
2730 {
2731 DistinctExpr *expr = (DistinctExpr *) node;
2732 List *args;
2733 ListCell *arg;
2734 bool has_null_input = false;
2735 bool all_null_input = true;
2736 bool has_nonconst_input = false;
2737 Expr *simple;
2738 DistinctExpr *newexpr;
2739
2740 /*
2741 * Reduce constants in the DistinctExpr's arguments. We know
2742 * args is either NIL or a List node, so we can call
2743 * expression_tree_mutator directly rather than recursing to
2744 * self.
2745 */
2746 args = (List *) expression_tree_mutator((Node *) expr->args,
2747 eval_const_expressions_mutator,
2748 (void *) context);
2749
2750 /*
2751 * We must do our own check for NULLs because DistinctExpr has
2752 * different results for NULL input than the underlying
2753 * operator does.
2754 */
2755 foreach(arg, args)
2756 {
2757 if (IsA(lfirst(arg), Const))
2758 {
2759 has_null_input |= ((Const *) lfirst(arg))->constisnull;
2760 all_null_input &= ((Const *) lfirst(arg))->constisnull;
2761 }
2762 else
2763 has_nonconst_input = true;
2764 }
2765
2766 /* all constants? then can optimize this out */
2767 if (!has_nonconst_input)
2768 {
2769 /* all nulls? then not distinct */
2770 if (all_null_input)
2771 return makeBoolConst(false, false);
2772
2773 /* one null? then distinct */
2774 if (has_null_input)
2775 return makeBoolConst(true, false);
2776
2777 /* otherwise try to evaluate the '=' operator */
2778 /* (NOT okay to try to inline it, though!) */
2779
2780 /*
2781 * Need to get OID of underlying function. Okay to
2782 * scribble on input to this extent.
2783 */
2784 set_opfuncid((OpExpr *) expr); /* rely on struct
2785 * equivalence */
2786
2787 /*
2788 * Code for op/func reduction is pretty bulky, so split it
2789 * out as a separate function.
2790 */
2791 simple = simplify_function(expr->opfuncid,
2792 expr->opresulttype, -1,
2793 expr->opcollid,
2794 expr->inputcollid,
2795 &args,
2796 false,
2797 false,
2798 false,
2799 context);
2800 if (simple) /* successfully simplified it */
2801 {
2802 /*
2803 * Since the underlying operator is "=", must negate
2804 * its result
2805 */
2806 Const *csimple = (Const *) simple;
2807
2808 Assert(IsA(csimple, Const));
2809 csimple->constvalue =
2810 BoolGetDatum(!DatumGetBool(csimple->constvalue));
2811 return (Node *) csimple;
2812 }
2813 }
2814
2815 /*
2816 * The expression cannot be simplified any further, so build
2817 * and return a replacement DistinctExpr node using the
2818 * possibly-simplified arguments.
2819 */
2820 newexpr = makeNode(DistinctExpr);
2821 newexpr->opno = expr->opno;
2822 newexpr->opfuncid = expr->opfuncid;
2823 newexpr->opresulttype = expr->opresulttype;
2824 newexpr->opretset = expr->opretset;
2825 newexpr->opcollid = expr->opcollid;
2826 newexpr->inputcollid = expr->inputcollid;
2827 newexpr->args = args;
2828 newexpr->location = expr->location;
2829 return (Node *) newexpr;
2830 }
2831 case T_BoolExpr:
2832 {
2833 BoolExpr *expr = (BoolExpr *) node;
2834
2835 switch (expr->boolop)
2836 {
2837 case OR_EXPR:
2838 {
2839 List *newargs;
2840 bool haveNull = false;
2841 bool forceTrue = false;
2842
2843 newargs = simplify_or_arguments(expr->args,
2844 context,
2845 &haveNull,
2846 &forceTrue);
2847 if (forceTrue)
2848 return makeBoolConst(true, false);
2849 if (haveNull)
2850 newargs = lappend(newargs,
2851 makeBoolConst(false, true));
2852 /* If all the inputs are FALSE, result is FALSE */
2853 if (newargs == NIL)
2854 return makeBoolConst(false, false);
2855
2856 /*
2857 * If only one nonconst-or-NULL input, it's the
2858 * result
2859 */
2860 if (list_length(newargs) == 1)
2861 return (Node *) linitial(newargs);
2862 /* Else we still need an OR node */
2863 return (Node *) make_orclause(newargs);
2864 }
2865 case AND_EXPR:
2866 {
2867 List *newargs;
2868 bool haveNull = false;
2869 bool forceFalse = false;
2870
2871 newargs = simplify_and_arguments(expr->args,
2872 context,
2873 &haveNull,
2874 &forceFalse);
2875 if (forceFalse)
2876 return makeBoolConst(false, false);
2877 if (haveNull)
2878 newargs = lappend(newargs,
2879 makeBoolConst(false, true));
2880 /* If all the inputs are TRUE, result is TRUE */
2881 if (newargs == NIL)
2882 return makeBoolConst(true, false);
2883
2884 /*
2885 * If only one nonconst-or-NULL input, it's the
2886 * result
2887 */
2888 if (list_length(newargs) == 1)
2889 return (Node *) linitial(newargs);
2890 /* Else we still need an AND node */
2891 return (Node *) make_andclause(newargs);
2892 }
2893 case NOT_EXPR:
2894 {
2895 Node *arg;
2896
2897 Assert(list_length(expr->args) == 1);
2898 arg = eval_const_expressions_mutator(linitial(expr->args),
2899 context);
2900
2901 /*
2902 * Use negate_clause() to see if we can simplify
2903 * away the NOT.
2904 */
2905 return negate_clause(arg);
2906 }
2907 default:
2908 elog(ERROR, "unrecognized boolop: %d",
2909 (int) expr->boolop);
2910 break;
2911 }
2912 break;
2913 }
2914 case T_SubPlan:
2915 case T_AlternativeSubPlan:
2916
2917 /*
2918 * Return a SubPlan unchanged --- too late to do anything with it.
2919 *
2920 * XXX should we ereport() here instead? Probably this routine
2921 * should never be invoked after SubPlan creation.
2922 */
2923 return node;
2924 case T_RelabelType:
2925 {
2926 /*
2927 * If we can simplify the input to a constant, then we don't
2928 * need the RelabelType node anymore: just change the type
2929 * field of the Const node. Otherwise, must copy the
2930 * RelabelType node.
2931 */
2932 RelabelType *relabel = (RelabelType *) node;
2933 Node *arg;
2934
2935 arg = eval_const_expressions_mutator((Node *) relabel->arg,
2936 context);
2937
2938 /*
2939 * If we find stacked RelabelTypes (eg, from foo :: int ::
2940 * oid) we can discard all but the top one.
2941 */
2942 while (arg && IsA(arg, RelabelType))
2943 arg = (Node *) ((RelabelType *) arg)->arg;
2944
2945 if (arg && IsA(arg, Const))
2946 {
2947 Const *con = (Const *) arg;
2948
2949 con->consttype = relabel->resulttype;
2950 con->consttypmod = relabel->resulttypmod;
2951 con->constcollid = relabel->resultcollid;
2952 return (Node *) con;
2953 }
2954 else
2955 {
2956 RelabelType *newrelabel = makeNode(RelabelType);
2957
2958 newrelabel->arg = (Expr *) arg;
2959 newrelabel->resulttype = relabel->resulttype;
2960 newrelabel->resulttypmod = relabel->resulttypmod;
2961 newrelabel->resultcollid = relabel->resultcollid;
2962 newrelabel->relabelformat = relabel->relabelformat;
2963 newrelabel->location = relabel->location;
2964 return (Node *) newrelabel;
2965 }
2966 }
2967 case T_CoerceViaIO:
2968 {
2969 CoerceViaIO *expr = (CoerceViaIO *) node;
2970 List *args;
2971 Oid outfunc;
2972 bool outtypisvarlena;
2973 Oid infunc;
2974 Oid intypioparam;
2975 Expr *simple;
2976 CoerceViaIO *newexpr;
2977
2978 /* Make a List so we can use simplify_function */
2979 args = list_make1(expr->arg);
2980
2981 /*
2982 * CoerceViaIO represents calling the source type's output
2983 * function then the result type's input function. So, try to
2984 * simplify it as though it were a stack of two such function
2985 * calls. First we need to know what the functions are.
2986 *
2987 * Note that the coercion functions are assumed not to care
2988 * about input collation, so we just pass InvalidOid for that.
2989 */
2990 getTypeOutputInfo(exprType((Node *) expr->arg),
2991 &outfunc, &outtypisvarlena);
2992 getTypeInputInfo(expr->resulttype,
2993 &infunc, &intypioparam);
2994
2995 simple = simplify_function(outfunc,
2996 CSTRINGOID, -1,
2997 InvalidOid,
2998 InvalidOid,
2999 &args,
3000 false,
3001 true,
3002 true,
3003 context);
3004 if (simple) /* successfully simplified output fn */
3005 {
3006 /*
3007 * Input functions may want 1 to 3 arguments. We always
3008 * supply all three, trusting that nothing downstream will
3009 * complain.
3010 */
3011 args = list_make3(simple,
3012 makeConst(OIDOID,
3013 -1,
3014 InvalidOid,
3015 sizeof(Oid),
3016 ObjectIdGetDatum(intypioparam),
3017 false,
3018 true),
3019 makeConst(INT4OID,
3020 -1,
3021 InvalidOid,
3022 sizeof(int32),
3023 Int32GetDatum(-1),
3024 false,
3025 true));
3026
3027 simple = simplify_function(infunc,
3028 expr->resulttype, -1,
3029 expr->resultcollid,
3030 InvalidOid,
3031 &args,
3032 false,
3033 false,
3034 true,
3035 context);
3036 if (simple) /* successfully simplified input fn */
3037 return (Node *) simple;
3038 }
3039
3040 /*
3041 * The expression cannot be simplified any further, so build
3042 * and return a replacement CoerceViaIO node using the
3043 * possibly-simplified argument.
3044 */
3045 newexpr = makeNode(CoerceViaIO);
3046 newexpr->arg = (Expr *) linitial(args);
3047 newexpr->resulttype = expr->resulttype;
3048 newexpr->resultcollid = expr->resultcollid;
3049 newexpr->coerceformat = expr->coerceformat;
3050 newexpr->location = expr->location;
3051 return (Node *) newexpr;
3052 }
3053 case T_ArrayCoerceExpr:
3054 {
3055 ArrayCoerceExpr *expr = (ArrayCoerceExpr *) node;
3056 Expr *arg;
3057 ArrayCoerceExpr *newexpr;
3058
3059 /*
3060 * Reduce constants in the ArrayCoerceExpr's argument, then
3061 * build a new ArrayCoerceExpr.
3062 */
3063 arg = (Expr *) eval_const_expressions_mutator((Node *) expr->arg,
3064 context);
3065
3066 newexpr = makeNode(ArrayCoerceExpr);
3067 newexpr->arg = arg;
3068 newexpr->elemfuncid = expr->elemfuncid;
3069 newexpr->resulttype = expr->resulttype;
3070 newexpr->resulttypmod = expr->resulttypmod;
3071 newexpr->resultcollid = expr->resultcollid;
3072 newexpr->isExplicit = expr->isExplicit;
3073 newexpr->coerceformat = expr->coerceformat;
3074 newexpr->location = expr->location;
3075
3076 /*
3077 * If constant argument and it's a binary-coercible or
3078 * immutable conversion, we can simplify it to a constant.
3079 */
3080 if (arg && IsA(arg, Const) &&
3081 (!OidIsValid(newexpr->elemfuncid) ||
3082 func_volatile(newexpr->elemfuncid) == PROVOLATILE_IMMUTABLE))
3083 return (Node *) evaluate_expr((Expr *) newexpr,
3084 newexpr->resulttype,
3085 newexpr->resulttypmod,
3086 newexpr->resultcollid);
3087
3088 /* Else we must return the partially-simplified node */
3089 return (Node *) newexpr;
3090 }
3091 case T_CollateExpr:
3092 {
3093 /*
3094 * If we can simplify the input to a constant, then we don't
3095 * need the CollateExpr node at all: just change the
3096 * constcollid field of the Const node. Otherwise, replace
3097 * the CollateExpr with a RelabelType. (We do that so as to
3098 * improve uniformity of expression representation and thus
3099 * simplify comparison of expressions.)
3100 */
3101 CollateExpr *collate = (CollateExpr *) node;
3102 Node *arg;
3103
3104 arg = eval_const_expressions_mutator((Node *) collate->arg,
3105 context);
3106
3107 if (arg && IsA(arg, Const))
3108 {
3109 Const *con = (Const *) arg;
3110
3111 con->constcollid = collate->collOid;
3112 return (Node *) con;
3113 }
3114 else if (collate->collOid == exprCollation(arg))
3115 {
3116 /* Don't need a RelabelType either... */
3117 return arg;
3118 }
3119 else
3120 {
3121 RelabelType *relabel = makeNode(RelabelType);
3122
3123 relabel->resulttype = exprType(arg);
3124 relabel->resulttypmod = exprTypmod(arg);
3125 relabel->resultcollid = collate->collOid;
3126 relabel->relabelformat = COERCE_IMPLICIT_CAST;
3127 relabel->location = collate->location;
3128
3129 /* Don't create stacked RelabelTypes */
3130 while (arg && IsA(arg, RelabelType))
3131 arg = (Node *) ((RelabelType *) arg)->arg;
3132 relabel->arg = (Expr *) arg;
3133
3134 return (Node *) relabel;
3135 }
3136 }
3137 case T_CaseExpr:
3138 {
3139 /*----------
3140 * CASE expressions can be simplified if there are constant
3141 * condition clauses:
3142 * FALSE (or NULL): drop the alternative
3143 * TRUE: drop all remaining alternatives
3144 * If the first non-FALSE alternative is a constant TRUE,
3145 * we can simplify the entire CASE to that alternative's
3146 * expression. If there are no non-FALSE alternatives,
3147 * we simplify the entire CASE to the default result (ELSE).
3148 *
3149 * If we have a simple-form CASE with constant test
3150 * expression, we substitute the constant value for contained
3151 * CaseTestExpr placeholder nodes, so that we have the
3152 * opportunity to reduce constant test conditions. For
3153 * example this allows
3154 * CASE 0 WHEN 0 THEN 1 ELSE 1/0 END
3155 * to reduce to 1 rather than drawing a divide-by-0 error.
3156 * Note that when the test expression is constant, we don't
3157 * have to include it in the resulting CASE; for example
3158 * CASE 0 WHEN x THEN y ELSE z END
3159 * is transformed by the parser to
3160 * CASE 0 WHEN CaseTestExpr = x THEN y ELSE z END
3161 * which we can simplify to
3162 * CASE WHEN 0 = x THEN y ELSE z END
3163 * It is not necessary for the executor to evaluate the "arg"
3164 * expression when executing the CASE, since any contained
3165 * CaseTestExprs that might have referred to it will have been
3166 * replaced by the constant.
3167 *----------
3168 */
3169 CaseExpr *caseexpr = (CaseExpr *) node;
3170 CaseExpr *newcase;
3171 Node *save_case_val;
3172 Node *newarg;
3173 List *newargs;
3174 bool const_true_cond;
3175 Node *defresult = NULL;
3176 ListCell *arg;
3177
3178 /* Simplify the test expression, if any */
3179 newarg = eval_const_expressions_mutator((Node *) caseexpr->arg,
3180 context);
3181
3182 /* Set up for contained CaseTestExpr nodes */
3183 save_case_val = context->case_val;
3184 if (newarg && IsA(newarg, Const))
3185 {
3186 context->case_val = newarg;
3187 newarg = NULL; /* not needed anymore, see above */
3188 }
3189 else
3190 context->case_val = NULL;
3191
3192 /* Simplify the WHEN clauses */
3193 newargs = NIL;
3194 const_true_cond = false;
3195 foreach(arg, caseexpr->args)
3196 {
3197 CaseWhen *oldcasewhen = (CaseWhen *) lfirst(arg);
3198 Node *casecond;
3199 Node *caseresult;
3200
3201 Assert(IsA(oldcasewhen, CaseWhen));
3202
3203 /* Simplify this alternative's test condition */
3204 casecond = eval_const_expressions_mutator((Node *) oldcasewhen->expr,
3205 context);
3206
3207 /*
3208 * If the test condition is constant FALSE (or NULL), then
3209 * drop this WHEN clause completely, without processing
3210 * the result.
3211 */
3212 if (casecond && IsA(casecond, Const))
3213 {
3214 Const *const_input = (Const *) casecond;
3215
3216 if (const_input->constisnull ||
3217 !DatumGetBool(const_input->constvalue))
3218 continue; /* drop alternative with FALSE cond */
3219 /* Else it's constant TRUE */
3220 const_true_cond = true;
3221 }
3222
3223 /* Simplify this alternative's result value */
3224 caseresult = eval_const_expressions_mutator((Node *) oldcasewhen->result,
3225 context);
3226
3227 /* If non-constant test condition, emit a new WHEN node */
3228 if (!const_true_cond)
3229 {
3230 CaseWhen *newcasewhen = makeNode(CaseWhen);
3231
3232 newcasewhen->expr = (Expr *) casecond;
3233 newcasewhen->result = (Expr *) caseresult;
3234 newcasewhen->location = oldcasewhen->location;
3235 newargs = lappend(newargs, newcasewhen);
3236 continue;
3237 }
3238
3239 /*
3240 * Found a TRUE condition, so none of the remaining
3241 * alternatives can be reached. We treat the result as
3242 * the default result.
3243 */
3244 defresult = caseresult;
3245 break;
3246 }
3247
3248 /* Simplify the default result, unless we replaced it above */
3249 if (!const_true_cond)
3250 defresult = eval_const_expressions_mutator((Node *) caseexpr->defresult,
3251 context);
3252
3253 context->case_val = save_case_val;
3254
3255 /*
3256 * If no non-FALSE alternatives, CASE reduces to the default
3257 * result
3258 */
3259 if (newargs == NIL)
3260 return defresult;
3261 /* Otherwise we need a new CASE node */
3262 newcase = makeNode(CaseExpr);
3263 newcase->casetype = caseexpr->casetype;
3264 newcase->casecollid = caseexpr->casecollid;
3265 newcase->arg = (Expr *) newarg;
3266 newcase->args = newargs;
3267 newcase->defresult = (Expr *) defresult;
3268 newcase->location = caseexpr->location;
3269 return (Node *) newcase;
3270 }
3271 case T_CaseTestExpr:
3272 {
3273 /*
3274 * If we know a constant test value for the current CASE
3275 * construct, substitute it for the placeholder. Else just
3276 * return the placeholder as-is.
3277 */
3278 if (context->case_val)
3279 return copyObject(context->case_val);
3280 else
3281 return copyObject(node);
3282 }
3283 case T_ArrayExpr:
3284 {
3285 ArrayExpr *arrayexpr = (ArrayExpr *) node;
3286 ArrayExpr *newarray;
3287 bool all_const = true;
3288 List *newelems;
3289 ListCell *element;
3290
3291 newelems = NIL;
3292 foreach(element, arrayexpr->elements)
3293 {
3294 Node *e;
3295
3296 e = eval_const_expressions_mutator((Node *) lfirst(element),
3297 context);
3298 if (!IsA(e, Const))
3299 all_const = false;
3300 newelems = lappend(newelems, e);
3301 }
3302
3303 newarray = makeNode(ArrayExpr);
3304 newarray->array_typeid = arrayexpr->array_typeid;
3305 newarray->array_collid = arrayexpr->array_collid;
3306 newarray->element_typeid = arrayexpr->element_typeid;
3307 newarray->elements = newelems;
3308 newarray->multidims = arrayexpr->multidims;
3309 newarray->location = arrayexpr->location;
3310
3311 if (all_const)
3312 return (Node *) evaluate_expr((Expr *) newarray,
3313 newarray->array_typeid,
3314 exprTypmod(node),
3315 newarray->array_collid);
3316
3317 return (Node *) newarray;
3318 }
3319 case T_CoalesceExpr:
3320 {
3321 CoalesceExpr *coalesceexpr = (CoalesceExpr *) node;
3322 CoalesceExpr *newcoalesce;
3323 List *newargs;
3324 ListCell *arg;
3325
3326 newargs = NIL;
3327 foreach(arg, coalesceexpr->args)
3328 {
3329 Node *e;
3330
3331 e = eval_const_expressions_mutator((Node *) lfirst(arg),
3332 context);
3333
3334 /*
3335 * We can remove null constants from the list. For a
3336 * non-null constant, if it has not been preceded by any
3337 * other non-null-constant expressions then it is the
3338 * result. Otherwise, it's the next argument, but we can
3339 * drop following arguments since they will never be
3340 * reached.
3341 */
3342 if (IsA(e, Const))
3343 {
3344 if (((Const *) e)->constisnull)
3345 continue; /* drop null constant */
3346 if (newargs == NIL)
3347 return e; /* first expr */
3348 newargs = lappend(newargs, e);
3349 break;
3350 }
3351 newargs = lappend(newargs, e);
3352 }
3353
3354 /*
3355 * If all the arguments were constant null, the result is just
3356 * null
3357 */
3358 if (newargs == NIL)
3359 return (Node *) makeNullConst(coalesceexpr->coalescetype,
3360 -1,
3361 coalesceexpr->coalescecollid);
3362
3363 newcoalesce = makeNode(CoalesceExpr);
3364 newcoalesce->coalescetype = coalesceexpr->coalescetype;
3365 newcoalesce->coalescecollid = coalesceexpr->coalescecollid;
3366 newcoalesce->args = newargs;
3367 newcoalesce->location = coalesceexpr->location;
3368 return (Node *) newcoalesce;
3369 }
3370 case T_FieldSelect:
3371 {
3372 /*
3373 * We can optimize field selection from a whole-row Var into a
3374 * simple Var. (This case won't be generated directly by the
3375 * parser, because ParseComplexProjection short-circuits it.
3376 * But it can arise while simplifying functions.) Also, we
3377 * can optimize field selection from a RowExpr construct.
3378 *
3379 * However, replacing a whole-row Var in this way has a
3380 * pitfall: if we've already built the rel targetlist for the
3381 * source relation, then the whole-row Var is scheduled to be
3382 * produced by the relation scan, but the simple Var probably
3383 * isn't, which will lead to a failure in setrefs.c. This is
3384 * not a problem when handling simple single-level queries, in
3385 * which expression simplification always happens first. It
3386 * is a risk for lateral references from subqueries, though.
3387 * To avoid such failures, don't optimize uplevel references.
3388 *
3389 * We must also check that the declared type of the field is
3390 * still the same as when the FieldSelect was created --- this
3391 * can change if someone did ALTER COLUMN TYPE on the rowtype.
3392 */
3393 FieldSelect *fselect = (FieldSelect *) node;
3394 FieldSelect *newfselect;
3395 Node *arg;
3396
3397 arg = eval_const_expressions_mutator((Node *) fselect->arg,
3398 context);
3399 if (arg && IsA(arg, Var) &&
3400 ((Var *) arg)->varattno == InvalidAttrNumber &&
3401 ((Var *) arg)->varlevelsup == 0)
3402 {
3403 if (rowtype_field_matches(((Var *) arg)->vartype,
3404 fselect->fieldnum,
3405 fselect->resulttype,
3406 fselect->resulttypmod,
3407 fselect->resultcollid))
3408 return (Node *) makeVar(((Var *) arg)->varno,
3409 fselect->fieldnum,
3410 fselect->resulttype,
3411 fselect->resulttypmod,
3412 fselect->resultcollid,
3413 ((Var *) arg)->varlevelsup);
3414 }
3415 if (arg && IsA(arg, RowExpr))
3416 {
3417 RowExpr *rowexpr = (RowExpr *) arg;
3418
3419 if (fselect->fieldnum > 0 &&
3420 fselect->fieldnum <= list_length(rowexpr->args))
3421 {
3422 Node *fld = (Node *) list_nth(rowexpr->args,
3423 fselect->fieldnum - 1);
3424
3425 if (rowtype_field_matches(rowexpr->row_typeid,
3426 fselect->fieldnum,
3427 fselect->resulttype,
3428 fselect->resulttypmod,
3429 fselect->resultcollid) &&
3430 fselect->resulttype == exprType(fld) &&
3431 fselect->resulttypmod == exprTypmod(fld) &&
3432 fselect->resultcollid == exprCollation(fld))
3433 return fld;
3434 }
3435 }
3436 newfselect = makeNode(FieldSelect);
3437 newfselect->arg = (Expr *) arg;
3438 newfselect->fieldnum = fselect->fieldnum;
3439 newfselect->resulttype = fselect->resulttype;
3440 newfselect->resulttypmod = fselect->resulttypmod;
3441 newfselect->resultcollid = fselect->resultcollid;
3442 return (Node *) newfselect;
3443 }
3444 case T_NullTest:
3445 {
3446 NullTest *ntest = (NullTest *) node;
3447 NullTest *newntest;
3448 Node *arg;
3449
3450 arg = eval_const_expressions_mutator((Node *) ntest->arg,
3451 context);
3452 if (ntest->argisrow && arg && IsA(arg, RowExpr))
3453 {
3454 /*
3455 * We break ROW(...) IS [NOT] NULL into separate tests on
3456 * its component fields. This form is usually more
3457 * efficient to evaluate, as well as being more amenable
3458 * to optimization.
3459 */
3460 RowExpr *rarg = (RowExpr *) arg;
3461 List *newargs = NIL;
3462 ListCell *l;
3463
3464 foreach(l, rarg->args)
3465 {
3466 Node *relem = (Node *) lfirst(l);
3467
3468 /*
3469 * A constant field refutes the whole NullTest if it's
3470 * of the wrong nullness; else we can discard it.
3471 */
3472 if (relem && IsA(relem, Const))
3473 {
3474 Const *carg = (Const *) relem;
3475
3476 if (carg->constisnull ?
3477 (ntest->nulltesttype == IS_NOT_NULL) :
3478 (ntest->nulltesttype == IS_NULL))
3479 return makeBoolConst(false, false);
3480 continue;
3481 }
3482
3483 /*
3484 * Else, make a scalar (argisrow == false) NullTest
3485 * for this field. Scalar semantics are required
3486 * because IS [NOT] NULL doesn't recurse; see comments
3487 * in ExecEvalNullTest().
3488 */
3489 newntest = makeNode(NullTest);
3490 newntest->arg = (Expr *) relem;
3491 newntest->nulltesttype = ntest->nulltesttype;
3492 newntest->argisrow = false;
3493 newntest->location = ntest->location;
3494 newargs = lappend(newargs, newntest);
3495 }
3496 /* If all the inputs were constants, result is TRUE */
3497 if (newargs == NIL)
3498 return makeBoolConst(true, false);
3499 /* If only one nonconst input, it's the result */
3500 if (list_length(newargs) == 1)
3501 return (Node *) linitial(newargs);
3502 /* Else we need an AND node */
3503 return (Node *) make_andclause(newargs);
3504 }
3505 if (!ntest->argisrow && arg && IsA(arg, Const))
3506 {
3507 Const *carg = (Const *) arg;
3508 bool result;
3509
3510 switch (ntest->nulltesttype)
3511 {
3512 case IS_NULL:
3513 result = carg->constisnull;
3514 break;
3515 case IS_NOT_NULL:
3516 result = !carg->constisnull;
3517 break;
3518 default:
3519 elog(ERROR, "unrecognized nulltesttype: %d",
3520 (int) ntest->nulltesttype);
3521 result = false; /* keep compiler quiet */
3522 break;
3523 }
3524
3525 return makeBoolConst(result, false);
3526 }
3527
3528 newntest = makeNode(NullTest);
3529 newntest->arg = (Expr *) arg;
3530 newntest->nulltesttype = ntest->nulltesttype;
3531 newntest->argisrow = ntest->argisrow;
3532 newntest->location = ntest->location;
3533 return (Node *) newntest;
3534 }
3535 case T_BooleanTest:
3536 {
3537 BooleanTest *btest = (BooleanTest *) node;
3538 BooleanTest *newbtest;
3539 Node *arg;
3540
3541 arg = eval_const_expressions_mutator((Node *) btest->arg,
3542 context);
3543 if (arg && IsA(arg, Const))
3544 {
3545 Const *carg = (Const *) arg;
3546 bool result;
3547
3548 switch (btest->booltesttype)
3549 {
3550 case IS_TRUE:
3551 result = (!carg->constisnull &&
3552 DatumGetBool(carg->constvalue));
3553 break;
3554 case IS_NOT_TRUE:
3555 result = (carg->constisnull ||
3556 !DatumGetBool(carg->constvalue));
3557 break;
3558 case IS_FALSE:
3559 result = (!carg->constisnull &&
3560 !DatumGetBool(carg->constvalue));
3561 break;
3562 case IS_NOT_FALSE:
3563 result = (carg->constisnull ||
3564 DatumGetBool(carg->constvalue));
3565 break;
3566 case IS_UNKNOWN:
3567 result = carg->constisnull;
3568 break;
3569 case IS_NOT_UNKNOWN:
3570 result = !carg->constisnull;
3571 break;
3572 default:
3573 elog(ERROR, "unrecognized booltesttype: %d",
3574 (int) btest->booltesttype);
3575 result = false; /* keep compiler quiet */
3576 break;
3577 }
3578
3579 return makeBoolConst(result, false);
3580 }
3581
3582 newbtest = makeNode(BooleanTest);
3583 newbtest->arg = (Expr *) arg;
3584 newbtest->booltesttype = btest->booltesttype;
3585 newbtest->location = btest->location;
3586 return (Node *) newbtest;
3587 }
3588 case T_PlaceHolderVar:
3589
3590 /*
3591 * In estimation mode, just strip the PlaceHolderVar node
3592 * altogether; this amounts to estimating that the contained value
3593 * won't be forced to null by an outer join. In regular mode we
3594 * just use the default behavior (ie, simplify the expression but
3595 * leave the PlaceHolderVar node intact).
3596 */
3597 if (context->estimate)
3598 {
3599 PlaceHolderVar *phv = (PlaceHolderVar *) node;
3600
3601 return eval_const_expressions_mutator((Node *) phv->phexpr,
3602 context);
3603 }
3604 break;
3605 default:
3606 break;
3607 }
3608
3609 /*
3610 * For any node type not handled above, we recurse using
3611 * expression_tree_mutator, which will copy the node unchanged but try to
3612 * simplify its arguments (if any) using this routine. For example: we
3613 * cannot eliminate an ArrayRef node, but we might be able to simplify
3614 * constant expressions in its subscripts.
3615 */
3616 return expression_tree_mutator(node, eval_const_expressions_mutator,
3617 (void *) context);
3618 }
3619
3620 /*
3621 * Subroutine for eval_const_expressions: process arguments of an OR clause
3622 *
3623 * This includes flattening of nested ORs as well as recursion to
3624 * eval_const_expressions to simplify the OR arguments.
3625 *
3626 * After simplification, OR arguments are handled as follows:
3627 * non constant: keep
3628 * FALSE: drop (does not affect result)
3629 * TRUE: force result to TRUE
3630 * NULL: keep only one
3631 * We must keep one NULL input because ExecEvalOr returns NULL when no input
3632 * is TRUE and at least one is NULL. We don't actually include the NULL
3633 * here, that's supposed to be done by the caller.
3634 *
3635 * The output arguments *haveNull and *forceTrue must be initialized FALSE
3636 * by the caller. They will be set TRUE if a null constant or true constant,
3637 * respectively, is detected anywhere in the argument list.
3638 */
3639 static List *
simplify_or_arguments(List * args,eval_const_expressions_context * context,bool * haveNull,bool * forceTrue)3640 simplify_or_arguments(List *args,
3641 eval_const_expressions_context *context,
3642 bool *haveNull, bool *forceTrue)
3643 {
3644 List *newargs = NIL;
3645 List *unprocessed_args;
3646
3647 /*
3648 * We want to ensure that any OR immediately beneath another OR gets
3649 * flattened into a single OR-list, so as to simplify later reasoning.
3650 *
3651 * To avoid stack overflow from recursion of eval_const_expressions, we
3652 * resort to some tenseness here: we keep a list of not-yet-processed
3653 * inputs, and handle flattening of nested ORs by prepending to the to-do
3654 * list instead of recursing. Now that the parser generates N-argument
3655 * ORs from simple lists, this complexity is probably less necessary than
3656 * it once was, but we might as well keep the logic.
3657 */
3658 unprocessed_args = list_copy(args);
3659 while (unprocessed_args)
3660 {
3661 Node *arg = (Node *) linitial(unprocessed_args);
3662
3663 unprocessed_args = list_delete_first(unprocessed_args);
3664
3665 /* flatten nested ORs as per above comment */
3666 if (or_clause(arg))
3667 {
3668 List *subargs = list_copy(((BoolExpr *) arg)->args);
3669
3670 /* overly tense code to avoid leaking unused list header */
3671 if (!unprocessed_args)
3672 unprocessed_args = subargs;
3673 else
3674 {
3675 List *oldhdr = unprocessed_args;
3676
3677 unprocessed_args = list_concat(subargs, unprocessed_args);
3678 pfree(oldhdr);
3679 }
3680 continue;
3681 }
3682
3683 /* If it's not an OR, simplify it */
3684 arg = eval_const_expressions_mutator(arg, context);
3685
3686 /*
3687 * It is unlikely but not impossible for simplification of a non-OR
3688 * clause to produce an OR. Recheck, but don't be too tense about it
3689 * since it's not a mainstream case. In particular we don't worry
3690 * about const-simplifying the input twice.
3691 */
3692 if (or_clause(arg))
3693 {
3694 List *subargs = list_copy(((BoolExpr *) arg)->args);
3695
3696 unprocessed_args = list_concat(subargs, unprocessed_args);
3697 continue;
3698 }
3699
3700 /*
3701 * OK, we have a const-simplified non-OR argument. Process it per
3702 * comments above.
3703 */
3704 if (IsA(arg, Const))
3705 {
3706 Const *const_input = (Const *) arg;
3707
3708 if (const_input->constisnull)
3709 *haveNull = true;
3710 else if (DatumGetBool(const_input->constvalue))
3711 {
3712 *forceTrue = true;
3713
3714 /*
3715 * Once we detect a TRUE result we can just exit the loop
3716 * immediately. However, if we ever add a notion of
3717 * non-removable functions, we'd need to keep scanning.
3718 */
3719 return NIL;
3720 }
3721 /* otherwise, we can drop the constant-false input */
3722 continue;
3723 }
3724
3725 /* else emit the simplified arg into the result list */
3726 newargs = lappend(newargs, arg);
3727 }
3728
3729 return newargs;
3730 }
3731
3732 /*
3733 * Subroutine for eval_const_expressions: process arguments of an AND clause
3734 *
3735 * This includes flattening of nested ANDs as well as recursion to
3736 * eval_const_expressions to simplify the AND arguments.
3737 *
3738 * After simplification, AND arguments are handled as follows:
3739 * non constant: keep
3740 * TRUE: drop (does not affect result)
3741 * FALSE: force result to FALSE
3742 * NULL: keep only one
3743 * We must keep one NULL input because ExecEvalAnd returns NULL when no input
3744 * is FALSE and at least one is NULL. We don't actually include the NULL
3745 * here, that's supposed to be done by the caller.
3746 *
3747 * The output arguments *haveNull and *forceFalse must be initialized FALSE
3748 * by the caller. They will be set TRUE if a null constant or false constant,
3749 * respectively, is detected anywhere in the argument list.
3750 */
3751 static List *
simplify_and_arguments(List * args,eval_const_expressions_context * context,bool * haveNull,bool * forceFalse)3752 simplify_and_arguments(List *args,
3753 eval_const_expressions_context *context,
3754 bool *haveNull, bool *forceFalse)
3755 {
3756 List *newargs = NIL;
3757 List *unprocessed_args;
3758
3759 /* See comments in simplify_or_arguments */
3760 unprocessed_args = list_copy(args);
3761 while (unprocessed_args)
3762 {
3763 Node *arg = (Node *) linitial(unprocessed_args);
3764
3765 unprocessed_args = list_delete_first(unprocessed_args);
3766
3767 /* flatten nested ANDs as per above comment */
3768 if (and_clause(arg))
3769 {
3770 List *subargs = list_copy(((BoolExpr *) arg)->args);
3771
3772 /* overly tense code to avoid leaking unused list header */
3773 if (!unprocessed_args)
3774 unprocessed_args = subargs;
3775 else
3776 {
3777 List *oldhdr = unprocessed_args;
3778
3779 unprocessed_args = list_concat(subargs, unprocessed_args);
3780 pfree(oldhdr);
3781 }
3782 continue;
3783 }
3784
3785 /* If it's not an AND, simplify it */
3786 arg = eval_const_expressions_mutator(arg, context);
3787
3788 /*
3789 * It is unlikely but not impossible for simplification of a non-AND
3790 * clause to produce an AND. Recheck, but don't be too tense about it
3791 * since it's not a mainstream case. In particular we don't worry
3792 * about const-simplifying the input twice.
3793 */
3794 if (and_clause(arg))
3795 {
3796 List *subargs = list_copy(((BoolExpr *) arg)->args);
3797
3798 unprocessed_args = list_concat(subargs, unprocessed_args);
3799 continue;
3800 }
3801
3802 /*
3803 * OK, we have a const-simplified non-AND argument. Process it per
3804 * comments above.
3805 */
3806 if (IsA(arg, Const))
3807 {
3808 Const *const_input = (Const *) arg;
3809
3810 if (const_input->constisnull)
3811 *haveNull = true;
3812 else if (!DatumGetBool(const_input->constvalue))
3813 {
3814 *forceFalse = true;
3815
3816 /*
3817 * Once we detect a FALSE result we can just exit the loop
3818 * immediately. However, if we ever add a notion of
3819 * non-removable functions, we'd need to keep scanning.
3820 */
3821 return NIL;
3822 }
3823 /* otherwise, we can drop the constant-true input */
3824 continue;
3825 }
3826
3827 /* else emit the simplified arg into the result list */
3828 newargs = lappend(newargs, arg);
3829 }
3830
3831 return newargs;
3832 }
3833
3834 /*
3835 * Subroutine for eval_const_expressions: try to simplify boolean equality
3836 * or inequality condition
3837 *
3838 * Inputs are the operator OID and the simplified arguments to the operator.
3839 * Returns a simplified expression if successful, or NULL if cannot
3840 * simplify the expression.
3841 *
3842 * The idea here is to reduce "x = true" to "x" and "x = false" to "NOT x",
3843 * or similarly "x <> true" to "NOT x" and "x <> false" to "x".
3844 * This is only marginally useful in itself, but doing it in constant folding
3845 * ensures that we will recognize these forms as being equivalent in, for
3846 * example, partial index matching.
3847 *
3848 * We come here only if simplify_function has failed; therefore we cannot
3849 * see two constant inputs, nor a constant-NULL input.
3850 */
3851 static Node *
simplify_boolean_equality(Oid opno,List * args)3852 simplify_boolean_equality(Oid opno, List *args)
3853 {
3854 Node *leftop;
3855 Node *rightop;
3856
3857 Assert(list_length(args) == 2);
3858 leftop = linitial(args);
3859 rightop = lsecond(args);
3860 if (leftop && IsA(leftop, Const))
3861 {
3862 Assert(!((Const *) leftop)->constisnull);
3863 if (opno == BooleanEqualOperator)
3864 {
3865 if (DatumGetBool(((Const *) leftop)->constvalue))
3866 return rightop; /* true = foo */
3867 else
3868 return negate_clause(rightop); /* false = foo */
3869 }
3870 else
3871 {
3872 if (DatumGetBool(((Const *) leftop)->constvalue))
3873 return negate_clause(rightop); /* true <> foo */
3874 else
3875 return rightop; /* false <> foo */
3876 }
3877 }
3878 if (rightop && IsA(rightop, Const))
3879 {
3880 Assert(!((Const *) rightop)->constisnull);
3881 if (opno == BooleanEqualOperator)
3882 {
3883 if (DatumGetBool(((Const *) rightop)->constvalue))
3884 return leftop; /* foo = true */
3885 else
3886 return negate_clause(leftop); /* foo = false */
3887 }
3888 else
3889 {
3890 if (DatumGetBool(((Const *) rightop)->constvalue))
3891 return negate_clause(leftop); /* foo <> true */
3892 else
3893 return leftop; /* foo <> false */
3894 }
3895 }
3896 return NULL;
3897 }
3898
3899 /*
3900 * Subroutine for eval_const_expressions: try to simplify a function call
3901 * (which might originally have been an operator; we don't care)
3902 *
3903 * Inputs are the function OID, actual result type OID (which is needed for
3904 * polymorphic functions), result typmod, result collation, the input
3905 * collation to use for the function, the original argument list (not
3906 * const-simplified yet, unless process_args is false), and some flags;
3907 * also the context data for eval_const_expressions.
3908 *
3909 * Returns a simplified expression if successful, or NULL if cannot
3910 * simplify the function call.
3911 *
3912 * This function is also responsible for converting named-notation argument
3913 * lists into positional notation and/or adding any needed default argument
3914 * expressions; which is a bit grotty, but it avoids extra fetches of the
3915 * function's pg_proc tuple. For this reason, the args list is
3916 * pass-by-reference. Conversion and const-simplification of the args list
3917 * will be done even if simplification of the function call itself is not
3918 * possible.
3919 */
3920 static Expr *
simplify_function(Oid funcid,Oid result_type,int32 result_typmod,Oid result_collid,Oid input_collid,List ** args_p,bool funcvariadic,bool process_args,bool allow_non_const,eval_const_expressions_context * context)3921 simplify_function(Oid funcid, Oid result_type, int32 result_typmod,
3922 Oid result_collid, Oid input_collid, List **args_p,
3923 bool funcvariadic, bool process_args, bool allow_non_const,
3924 eval_const_expressions_context *context)
3925 {
3926 List *args = *args_p;
3927 HeapTuple func_tuple;
3928 Form_pg_proc func_form;
3929 Expr *newexpr;
3930
3931 /*
3932 * We have three strategies for simplification: execute the function to
3933 * deliver a constant result, use a transform function to generate a
3934 * substitute node tree, or expand in-line the body of the function
3935 * definition (which only works for simple SQL-language functions, but
3936 * that is a common case). Each case needs access to the function's
3937 * pg_proc tuple, so fetch it just once.
3938 *
3939 * Note: the allow_non_const flag suppresses both the second and third
3940 * strategies; so if !allow_non_const, simplify_function can only return a
3941 * Const or NULL. Argument-list rewriting happens anyway, though.
3942 */
3943 func_tuple = SearchSysCache1(PROCOID, ObjectIdGetDatum(funcid));
3944 if (!HeapTupleIsValid(func_tuple))
3945 elog(ERROR, "cache lookup failed for function %u", funcid);
3946 func_form = (Form_pg_proc) GETSTRUCT(func_tuple);
3947
3948 /*
3949 * Process the function arguments, unless the caller did it already.
3950 *
3951 * Here we must deal with named or defaulted arguments, and then
3952 * recursively apply eval_const_expressions to the whole argument list.
3953 */
3954 if (process_args)
3955 {
3956 args = expand_function_arguments(args, result_type, func_tuple);
3957 args = (List *) expression_tree_mutator((Node *) args,
3958 eval_const_expressions_mutator,
3959 (void *) context);
3960 /* Argument processing done, give it back to the caller */
3961 *args_p = args;
3962 }
3963
3964 /* Now attempt simplification of the function call proper. */
3965
3966 newexpr = evaluate_function(funcid, result_type, result_typmod,
3967 result_collid, input_collid,
3968 args, funcvariadic,
3969 func_tuple, context);
3970
3971 if (!newexpr && allow_non_const && OidIsValid(func_form->protransform))
3972 {
3973 /*
3974 * Build a dummy FuncExpr node containing the simplified arg list. We
3975 * use this approach to present a uniform interface to the transform
3976 * function regardless of how the function is actually being invoked.
3977 */
3978 FuncExpr fexpr;
3979
3980 fexpr.xpr.type = T_FuncExpr;
3981 fexpr.funcid = funcid;
3982 fexpr.funcresulttype = result_type;
3983 fexpr.funcretset = func_form->proretset;
3984 fexpr.funcvariadic = funcvariadic;
3985 fexpr.funcformat = COERCE_EXPLICIT_CALL;
3986 fexpr.funccollid = result_collid;
3987 fexpr.inputcollid = input_collid;
3988 fexpr.args = args;
3989 fexpr.location = -1;
3990
3991 newexpr = (Expr *)
3992 DatumGetPointer(OidFunctionCall1(func_form->protransform,
3993 PointerGetDatum(&fexpr)));
3994 }
3995
3996 if (!newexpr && allow_non_const)
3997 newexpr = inline_function(funcid, result_type, result_collid,
3998 input_collid, args, funcvariadic,
3999 func_tuple, context);
4000
4001 ReleaseSysCache(func_tuple);
4002
4003 return newexpr;
4004 }
4005
4006 /*
4007 * expand_function_arguments: convert named-notation args to positional args
4008 * and/or insert default args, as needed
4009 *
4010 * If we need to change anything, the input argument list is copied, not
4011 * modified.
4012 *
4013 * Note: this gets applied to operator argument lists too, even though the
4014 * cases it handles should never occur there. This should be OK since it
4015 * will fall through very quickly if there's nothing to do.
4016 */
4017 static List *
expand_function_arguments(List * args,Oid result_type,HeapTuple func_tuple)4018 expand_function_arguments(List *args, Oid result_type, HeapTuple func_tuple)
4019 {
4020 Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4021 bool has_named_args = false;
4022 ListCell *lc;
4023
4024 /* Do we have any named arguments? */
4025 foreach(lc, args)
4026 {
4027 Node *arg = (Node *) lfirst(lc);
4028
4029 if (IsA(arg, NamedArgExpr))
4030 {
4031 has_named_args = true;
4032 break;
4033 }
4034 }
4035
4036 /* If so, we must apply reorder_function_arguments */
4037 if (has_named_args)
4038 {
4039 args = reorder_function_arguments(args, func_tuple);
4040 /* Recheck argument types and add casts if needed */
4041 recheck_cast_function_args(args, result_type, func_tuple);
4042 }
4043 else if (list_length(args) < funcform->pronargs)
4044 {
4045 /* No named args, but we seem to be short some defaults */
4046 args = add_function_defaults(args, func_tuple);
4047 /* Recheck argument types and add casts if needed */
4048 recheck_cast_function_args(args, result_type, func_tuple);
4049 }
4050
4051 return args;
4052 }
4053
4054 /*
4055 * reorder_function_arguments: convert named-notation args to positional args
4056 *
4057 * This function also inserts default argument values as needed, since it's
4058 * impossible to form a truly valid positional call without that.
4059 */
4060 static List *
reorder_function_arguments(List * args,HeapTuple func_tuple)4061 reorder_function_arguments(List *args, HeapTuple func_tuple)
4062 {
4063 Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4064 int pronargs = funcform->pronargs;
4065 int nargsprovided = list_length(args);
4066 Node *argarray[FUNC_MAX_ARGS];
4067 ListCell *lc;
4068 int i;
4069
4070 Assert(nargsprovided <= pronargs);
4071 if (pronargs < 0 || pronargs > FUNC_MAX_ARGS)
4072 elog(ERROR, "too many function arguments");
4073 memset(argarray, 0, pronargs * sizeof(Node *));
4074
4075 /* Deconstruct the argument list into an array indexed by argnumber */
4076 i = 0;
4077 foreach(lc, args)
4078 {
4079 Node *arg = (Node *) lfirst(lc);
4080
4081 if (!IsA(arg, NamedArgExpr))
4082 {
4083 /* positional argument, assumed to precede all named args */
4084 Assert(argarray[i] == NULL);
4085 argarray[i++] = arg;
4086 }
4087 else
4088 {
4089 NamedArgExpr *na = (NamedArgExpr *) arg;
4090
4091 Assert(argarray[na->argnumber] == NULL);
4092 argarray[na->argnumber] = (Node *) na->arg;
4093 }
4094 }
4095
4096 /*
4097 * Fetch default expressions, if needed, and insert into array at proper
4098 * locations (they aren't necessarily consecutive or all used)
4099 */
4100 if (nargsprovided < pronargs)
4101 {
4102 List *defaults = fetch_function_defaults(func_tuple);
4103
4104 i = pronargs - funcform->pronargdefaults;
4105 foreach(lc, defaults)
4106 {
4107 if (argarray[i] == NULL)
4108 argarray[i] = (Node *) lfirst(lc);
4109 i++;
4110 }
4111 }
4112
4113 /* Now reconstruct the args list in proper order */
4114 args = NIL;
4115 for (i = 0; i < pronargs; i++)
4116 {
4117 Assert(argarray[i] != NULL);
4118 args = lappend(args, argarray[i]);
4119 }
4120
4121 return args;
4122 }
4123
4124 /*
4125 * add_function_defaults: add missing function arguments from its defaults
4126 *
4127 * This is used only when the argument list was positional to begin with,
4128 * and so we know we just need to add defaults at the end.
4129 */
4130 static List *
add_function_defaults(List * args,HeapTuple func_tuple)4131 add_function_defaults(List *args, HeapTuple func_tuple)
4132 {
4133 Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4134 int nargsprovided = list_length(args);
4135 List *defaults;
4136 int ndelete;
4137
4138 /* Get all the default expressions from the pg_proc tuple */
4139 defaults = fetch_function_defaults(func_tuple);
4140
4141 /* Delete any unused defaults from the list */
4142 ndelete = nargsprovided + list_length(defaults) - funcform->pronargs;
4143 if (ndelete < 0)
4144 elog(ERROR, "not enough default arguments");
4145 while (ndelete-- > 0)
4146 defaults = list_delete_first(defaults);
4147
4148 /* And form the combined argument list, not modifying the input list */
4149 return list_concat(list_copy(args), defaults);
4150 }
4151
4152 /*
4153 * fetch_function_defaults: get function's default arguments as expression list
4154 */
4155 static List *
fetch_function_defaults(HeapTuple func_tuple)4156 fetch_function_defaults(HeapTuple func_tuple)
4157 {
4158 List *defaults;
4159 Datum proargdefaults;
4160 bool isnull;
4161 char *str;
4162
4163 /* The error cases here shouldn't happen, but check anyway */
4164 proargdefaults = SysCacheGetAttr(PROCOID, func_tuple,
4165 Anum_pg_proc_proargdefaults,
4166 &isnull);
4167 if (isnull)
4168 elog(ERROR, "not enough default arguments");
4169 str = TextDatumGetCString(proargdefaults);
4170 defaults = (List *) stringToNode(str);
4171 Assert(IsA(defaults, List));
4172 pfree(str);
4173 return defaults;
4174 }
4175
4176 /*
4177 * recheck_cast_function_args: recheck function args and typecast as needed
4178 * after adding defaults.
4179 *
4180 * It is possible for some of the defaulted arguments to be polymorphic;
4181 * therefore we can't assume that the default expressions have the correct
4182 * data types already. We have to re-resolve polymorphics and do coercion
4183 * just like the parser did.
4184 *
4185 * This should be a no-op if there are no polymorphic arguments,
4186 * but we do it anyway to be sure.
4187 *
4188 * Note: if any casts are needed, the args list is modified in-place;
4189 * caller should have already copied the list structure.
4190 */
4191 static void
recheck_cast_function_args(List * args,Oid result_type,HeapTuple func_tuple)4192 recheck_cast_function_args(List *args, Oid result_type, HeapTuple func_tuple)
4193 {
4194 Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4195 int nargs;
4196 Oid actual_arg_types[FUNC_MAX_ARGS];
4197 Oid declared_arg_types[FUNC_MAX_ARGS];
4198 Oid rettype;
4199 ListCell *lc;
4200
4201 if (list_length(args) > FUNC_MAX_ARGS)
4202 elog(ERROR, "too many function arguments");
4203 nargs = 0;
4204 foreach(lc, args)
4205 {
4206 actual_arg_types[nargs++] = exprType((Node *) lfirst(lc));
4207 }
4208 Assert(nargs == funcform->pronargs);
4209 memcpy(declared_arg_types, funcform->proargtypes.values,
4210 funcform->pronargs * sizeof(Oid));
4211 rettype = enforce_generic_type_consistency(actual_arg_types,
4212 declared_arg_types,
4213 nargs,
4214 funcform->prorettype,
4215 false);
4216 /* let's just check we got the same answer as the parser did ... */
4217 if (rettype != result_type)
4218 elog(ERROR, "function's resolved result type changed during planning");
4219
4220 /* perform any necessary typecasting of arguments */
4221 make_fn_arguments(NULL, args, actual_arg_types, declared_arg_types);
4222 }
4223
4224 /*
4225 * evaluate_function: try to pre-evaluate a function call
4226 *
4227 * We can do this if the function is strict and has any constant-null inputs
4228 * (just return a null constant), or if the function is immutable and has all
4229 * constant inputs (call it and return the result as a Const node). In
4230 * estimation mode we are willing to pre-evaluate stable functions too.
4231 *
4232 * Returns a simplified expression if successful, or NULL if cannot
4233 * simplify the function.
4234 */
4235 static Expr *
evaluate_function(Oid funcid,Oid result_type,int32 result_typmod,Oid result_collid,Oid input_collid,List * args,bool funcvariadic,HeapTuple func_tuple,eval_const_expressions_context * context)4236 evaluate_function(Oid funcid, Oid result_type, int32 result_typmod,
4237 Oid result_collid, Oid input_collid, List *args,
4238 bool funcvariadic,
4239 HeapTuple func_tuple,
4240 eval_const_expressions_context *context)
4241 {
4242 Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4243 bool has_nonconst_input = false;
4244 bool has_null_input = false;
4245 ListCell *arg;
4246 FuncExpr *newexpr;
4247
4248 /*
4249 * Can't simplify if it returns a set.
4250 */
4251 if (funcform->proretset)
4252 return NULL;
4253
4254 /*
4255 * Can't simplify if it returns RECORD. The immediate problem is that it
4256 * will be needing an expected tupdesc which we can't supply here.
4257 *
4258 * In the case where it has OUT parameters, it could get by without an
4259 * expected tupdesc, but we still have issues: get_expr_result_type()
4260 * doesn't know how to extract type info from a RECORD constant, and in
4261 * the case of a NULL function result there doesn't seem to be any clean
4262 * way to fix that. In view of the likelihood of there being still other
4263 * gotchas, seems best to leave the function call unreduced.
4264 */
4265 if (funcform->prorettype == RECORDOID)
4266 return NULL;
4267
4268 /*
4269 * Check for constant inputs and especially constant-NULL inputs.
4270 */
4271 foreach(arg, args)
4272 {
4273 if (IsA(lfirst(arg), Const))
4274 has_null_input |= ((Const *) lfirst(arg))->constisnull;
4275 else
4276 has_nonconst_input = true;
4277 }
4278
4279 /*
4280 * If the function is strict and has a constant-NULL input, it will never
4281 * be called at all, so we can replace the call by a NULL constant, even
4282 * if there are other inputs that aren't constant, and even if the
4283 * function is not otherwise immutable.
4284 */
4285 if (funcform->proisstrict && has_null_input)
4286 return (Expr *) makeNullConst(result_type, result_typmod,
4287 result_collid);
4288
4289 /*
4290 * Otherwise, can simplify only if all inputs are constants. (For a
4291 * non-strict function, constant NULL inputs are treated the same as
4292 * constant non-NULL inputs.)
4293 */
4294 if (has_nonconst_input)
4295 return NULL;
4296
4297 /*
4298 * Ordinarily we are only allowed to simplify immutable functions. But for
4299 * purposes of estimation, we consider it okay to simplify functions that
4300 * are merely stable; the risk that the result might change from planning
4301 * time to execution time is worth taking in preference to not being able
4302 * to estimate the value at all.
4303 */
4304 if (funcform->provolatile == PROVOLATILE_IMMUTABLE)
4305 /* okay */ ;
4306 else if (context->estimate && funcform->provolatile == PROVOLATILE_STABLE)
4307 /* okay */ ;
4308 else
4309 return NULL;
4310
4311 /*
4312 * OK, looks like we can simplify this operator/function.
4313 *
4314 * Build a new FuncExpr node containing the already-simplified arguments.
4315 */
4316 newexpr = makeNode(FuncExpr);
4317 newexpr->funcid = funcid;
4318 newexpr->funcresulttype = result_type;
4319 newexpr->funcretset = false;
4320 newexpr->funcvariadic = funcvariadic;
4321 newexpr->funcformat = COERCE_EXPLICIT_CALL; /* doesn't matter */
4322 newexpr->funccollid = result_collid; /* doesn't matter */
4323 newexpr->inputcollid = input_collid;
4324 newexpr->args = args;
4325 newexpr->location = -1;
4326
4327 return evaluate_expr((Expr *) newexpr, result_type, result_typmod,
4328 result_collid);
4329 }
4330
4331 /*
4332 * inline_function: try to expand a function call inline
4333 *
4334 * If the function is a sufficiently simple SQL-language function
4335 * (just "SELECT expression"), then we can inline it and avoid the rather
4336 * high per-call overhead of SQL functions. Furthermore, this can expose
4337 * opportunities for constant-folding within the function expression.
4338 *
4339 * We have to beware of some special cases however. A directly or
4340 * indirectly recursive function would cause us to recurse forever,
4341 * so we keep track of which functions we are already expanding and
4342 * do not re-expand them. Also, if a parameter is used more than once
4343 * in the SQL-function body, we require it not to contain any volatile
4344 * functions (volatiles might deliver inconsistent answers) nor to be
4345 * unreasonably expensive to evaluate. The expensiveness check not only
4346 * prevents us from doing multiple evaluations of an expensive parameter
4347 * at runtime, but is a safety value to limit growth of an expression due
4348 * to repeated inlining.
4349 *
4350 * We must also beware of changing the volatility or strictness status of
4351 * functions by inlining them.
4352 *
4353 * Also, at the moment we can't inline functions returning RECORD. This
4354 * doesn't work in the general case because it discards information such
4355 * as OUT-parameter declarations.
4356 *
4357 * Also, context-dependent expression nodes in the argument list are trouble.
4358 *
4359 * Returns a simplified expression if successful, or NULL if cannot
4360 * simplify the function.
4361 */
4362 static Expr *
inline_function(Oid funcid,Oid result_type,Oid result_collid,Oid input_collid,List * args,bool funcvariadic,HeapTuple func_tuple,eval_const_expressions_context * context)4363 inline_function(Oid funcid, Oid result_type, Oid result_collid,
4364 Oid input_collid, List *args,
4365 bool funcvariadic,
4366 HeapTuple func_tuple,
4367 eval_const_expressions_context *context)
4368 {
4369 Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4370 char *src;
4371 Datum tmp;
4372 bool isNull;
4373 bool modifyTargetList;
4374 MemoryContext oldcxt;
4375 MemoryContext mycxt;
4376 inline_error_callback_arg callback_arg;
4377 ErrorContextCallback sqlerrcontext;
4378 FuncExpr *fexpr;
4379 SQLFunctionParseInfoPtr pinfo;
4380 ParseState *pstate;
4381 List *raw_parsetree_list;
4382 Query *querytree;
4383 Node *newexpr;
4384 int *usecounts;
4385 ListCell *arg;
4386 int i;
4387
4388 /*
4389 * Forget it if the function is not SQL-language or has other showstopper
4390 * properties. (The nargs check is just paranoia.)
4391 */
4392 if (funcform->prolang != SQLlanguageId ||
4393 funcform->prosecdef ||
4394 funcform->proretset ||
4395 funcform->prorettype == RECORDOID ||
4396 !heap_attisnull(func_tuple, Anum_pg_proc_proconfig) ||
4397 funcform->pronargs != list_length(args))
4398 return NULL;
4399
4400 /* Check for recursive function, and give up trying to expand if so */
4401 if (list_member_oid(context->active_fns, funcid))
4402 return NULL;
4403
4404 /* Check permission to call function (fail later, if not) */
4405 if (pg_proc_aclcheck(funcid, GetUserId(), ACL_EXECUTE) != ACLCHECK_OK)
4406 return NULL;
4407
4408 /* Check whether a plugin wants to hook function entry/exit */
4409 if (FmgrHookIsNeeded(funcid))
4410 return NULL;
4411
4412 /*
4413 * Make a temporary memory context, so that we don't leak all the stuff
4414 * that parsing might create.
4415 */
4416 mycxt = AllocSetContextCreate(CurrentMemoryContext,
4417 "inline_function",
4418 ALLOCSET_DEFAULT_SIZES);
4419 oldcxt = MemoryContextSwitchTo(mycxt);
4420
4421 /* Fetch the function body */
4422 tmp = SysCacheGetAttr(PROCOID,
4423 func_tuple,
4424 Anum_pg_proc_prosrc,
4425 &isNull);
4426 if (isNull)
4427 elog(ERROR, "null prosrc for function %u", funcid);
4428 src = TextDatumGetCString(tmp);
4429
4430 /*
4431 * Setup error traceback support for ereport(). This is so that we can
4432 * finger the function that bad information came from.
4433 */
4434 callback_arg.proname = NameStr(funcform->proname);
4435 callback_arg.prosrc = src;
4436
4437 sqlerrcontext.callback = sql_inline_error_callback;
4438 sqlerrcontext.arg = (void *) &callback_arg;
4439 sqlerrcontext.previous = error_context_stack;
4440 error_context_stack = &sqlerrcontext;
4441
4442 /*
4443 * Set up to handle parameters while parsing the function body. We need a
4444 * dummy FuncExpr node containing the already-simplified arguments to pass
4445 * to prepare_sql_fn_parse_info. (It is really only needed if there are
4446 * some polymorphic arguments, but for simplicity we always build it.)
4447 */
4448 fexpr = makeNode(FuncExpr);
4449 fexpr->funcid = funcid;
4450 fexpr->funcresulttype = result_type;
4451 fexpr->funcretset = false;
4452 fexpr->funcvariadic = funcvariadic;
4453 fexpr->funcformat = COERCE_EXPLICIT_CALL; /* doesn't matter */
4454 fexpr->funccollid = result_collid; /* doesn't matter */
4455 fexpr->inputcollid = input_collid;
4456 fexpr->args = args;
4457 fexpr->location = -1;
4458
4459 pinfo = prepare_sql_fn_parse_info(func_tuple,
4460 (Node *) fexpr,
4461 input_collid);
4462
4463 /*
4464 * We just do parsing and parse analysis, not rewriting, because rewriting
4465 * will not affect table-free-SELECT-only queries, which is all that we
4466 * care about. Also, we can punt as soon as we detect more than one
4467 * command in the function body.
4468 */
4469 raw_parsetree_list = pg_parse_query(src);
4470 if (list_length(raw_parsetree_list) != 1)
4471 goto fail;
4472
4473 pstate = make_parsestate(NULL);
4474 pstate->p_sourcetext = src;
4475 sql_fn_parser_setup(pstate, pinfo);
4476
4477 querytree = transformTopLevelStmt(pstate, linitial(raw_parsetree_list));
4478
4479 free_parsestate(pstate);
4480
4481 /*
4482 * The single command must be a simple "SELECT expression".
4483 */
4484 if (!IsA(querytree, Query) ||
4485 querytree->commandType != CMD_SELECT ||
4486 querytree->utilityStmt ||
4487 querytree->hasAggs ||
4488 querytree->hasWindowFuncs ||
4489 querytree->hasSubLinks ||
4490 querytree->cteList ||
4491 querytree->rtable ||
4492 querytree->jointree->fromlist ||
4493 querytree->jointree->quals ||
4494 querytree->groupClause ||
4495 querytree->groupingSets ||
4496 querytree->havingQual ||
4497 querytree->windowClause ||
4498 querytree->distinctClause ||
4499 querytree->sortClause ||
4500 querytree->limitOffset ||
4501 querytree->limitCount ||
4502 querytree->setOperations ||
4503 list_length(querytree->targetList) != 1)
4504 goto fail;
4505
4506 /*
4507 * Make sure the function (still) returns what it's declared to. This
4508 * will raise an error if wrong, but that's okay since the function would
4509 * fail at runtime anyway. Note that check_sql_fn_retval will also insert
4510 * a RelabelType if needed to make the tlist expression match the declared
4511 * type of the function.
4512 *
4513 * Note: we do not try this until we have verified that no rewriting was
4514 * needed; that's probably not important, but let's be careful.
4515 */
4516 if (check_sql_fn_retval(funcid, result_type, list_make1(querytree),
4517 &modifyTargetList, NULL))
4518 goto fail; /* reject whole-tuple-result cases */
4519
4520 /* Now we can grab the tlist expression */
4521 newexpr = (Node *) ((TargetEntry *) linitial(querytree->targetList))->expr;
4522
4523 /* Assert that check_sql_fn_retval did the right thing */
4524 Assert(exprType(newexpr) == result_type);
4525 /* It couldn't have made any dangerous tlist changes, either */
4526 Assert(!modifyTargetList);
4527
4528 /*
4529 * Additional validity checks on the expression. It mustn't return a set,
4530 * and it mustn't be more volatile than the surrounding function (this is
4531 * to avoid breaking hacks that involve pretending a function is immutable
4532 * when it really ain't). If the surrounding function is declared strict,
4533 * then the expression must contain only strict constructs and must use
4534 * all of the function parameters (this is overkill, but an exact analysis
4535 * is hard).
4536 */
4537 if (expression_returns_set(newexpr))
4538 goto fail;
4539
4540 if (funcform->provolatile == PROVOLATILE_IMMUTABLE &&
4541 contain_mutable_functions(newexpr))
4542 goto fail;
4543 else if (funcform->provolatile == PROVOLATILE_STABLE &&
4544 contain_volatile_functions(newexpr))
4545 goto fail;
4546
4547 if (funcform->proisstrict &&
4548 contain_nonstrict_functions(newexpr))
4549 goto fail;
4550
4551 /*
4552 * If any parameter expression contains a context-dependent node, we can't
4553 * inline, for fear of putting such a node into the wrong context.
4554 */
4555 if (contain_context_dependent_node((Node *) args))
4556 goto fail;
4557
4558 /*
4559 * We may be able to do it; there are still checks on parameter usage to
4560 * make, but those are most easily done in combination with the actual
4561 * substitution of the inputs. So start building expression with inputs
4562 * substituted.
4563 */
4564 usecounts = (int *) palloc0(funcform->pronargs * sizeof(int));
4565 newexpr = substitute_actual_parameters(newexpr, funcform->pronargs,
4566 args, usecounts);
4567
4568 /* Now check for parameter usage */
4569 i = 0;
4570 foreach(arg, args)
4571 {
4572 Node *param = lfirst(arg);
4573
4574 if (usecounts[i] == 0)
4575 {
4576 /* Param not used at all: uncool if func is strict */
4577 if (funcform->proisstrict)
4578 goto fail;
4579 }
4580 else if (usecounts[i] != 1)
4581 {
4582 /* Param used multiple times: uncool if expensive or volatile */
4583 QualCost eval_cost;
4584
4585 /*
4586 * We define "expensive" as "contains any subplan or more than 10
4587 * operators". Note that the subplan search has to be done
4588 * explicitly, since cost_qual_eval() will barf on unplanned
4589 * subselects.
4590 */
4591 if (contain_subplans(param))
4592 goto fail;
4593 cost_qual_eval(&eval_cost, list_make1(param), NULL);
4594 if (eval_cost.startup + eval_cost.per_tuple >
4595 10 * cpu_operator_cost)
4596 goto fail;
4597
4598 /*
4599 * Check volatility last since this is more expensive than the
4600 * above tests
4601 */
4602 if (contain_volatile_functions(param))
4603 goto fail;
4604 }
4605 i++;
4606 }
4607
4608 /*
4609 * Whew --- we can make the substitution. Copy the modified expression
4610 * out of the temporary memory context, and clean up.
4611 */
4612 MemoryContextSwitchTo(oldcxt);
4613
4614 newexpr = copyObject(newexpr);
4615
4616 MemoryContextDelete(mycxt);
4617
4618 /*
4619 * If the result is of a collatable type, force the result to expose the
4620 * correct collation. In most cases this does not matter, but it's
4621 * possible that the function result is used directly as a sort key or in
4622 * other places where we expect exprCollation() to tell the truth.
4623 */
4624 if (OidIsValid(result_collid))
4625 {
4626 Oid exprcoll = exprCollation(newexpr);
4627
4628 if (OidIsValid(exprcoll) && exprcoll != result_collid)
4629 {
4630 CollateExpr *newnode = makeNode(CollateExpr);
4631
4632 newnode->arg = (Expr *) newexpr;
4633 newnode->collOid = result_collid;
4634 newnode->location = -1;
4635
4636 newexpr = (Node *) newnode;
4637 }
4638 }
4639
4640 /*
4641 * Since there is now no trace of the function in the plan tree, we must
4642 * explicitly record the plan's dependency on the function.
4643 */
4644 if (context->root)
4645 record_plan_function_dependency(context->root, funcid);
4646
4647 /*
4648 * Recursively try to simplify the modified expression. Here we must add
4649 * the current function to the context list of active functions.
4650 */
4651 context->active_fns = lcons_oid(funcid, context->active_fns);
4652 newexpr = eval_const_expressions_mutator(newexpr, context);
4653 context->active_fns = list_delete_first(context->active_fns);
4654
4655 error_context_stack = sqlerrcontext.previous;
4656
4657 return (Expr *) newexpr;
4658
4659 /* Here if func is not inlinable: release temp memory and return NULL */
4660 fail:
4661 MemoryContextSwitchTo(oldcxt);
4662 MemoryContextDelete(mycxt);
4663 error_context_stack = sqlerrcontext.previous;
4664
4665 return NULL;
4666 }
4667
4668 /*
4669 * Replace Param nodes by appropriate actual parameters
4670 */
4671 static Node *
substitute_actual_parameters(Node * expr,int nargs,List * args,int * usecounts)4672 substitute_actual_parameters(Node *expr, int nargs, List *args,
4673 int *usecounts)
4674 {
4675 substitute_actual_parameters_context context;
4676
4677 context.nargs = nargs;
4678 context.args = args;
4679 context.usecounts = usecounts;
4680
4681 return substitute_actual_parameters_mutator(expr, &context);
4682 }
4683
4684 static Node *
substitute_actual_parameters_mutator(Node * node,substitute_actual_parameters_context * context)4685 substitute_actual_parameters_mutator(Node *node,
4686 substitute_actual_parameters_context *context)
4687 {
4688 if (node == NULL)
4689 return NULL;
4690 if (IsA(node, Param))
4691 {
4692 Param *param = (Param *) node;
4693
4694 if (param->paramkind != PARAM_EXTERN)
4695 elog(ERROR, "unexpected paramkind: %d", (int) param->paramkind);
4696 if (param->paramid <= 0 || param->paramid > context->nargs)
4697 elog(ERROR, "invalid paramid: %d", param->paramid);
4698
4699 /* Count usage of parameter */
4700 context->usecounts[param->paramid - 1]++;
4701
4702 /* Select the appropriate actual arg and replace the Param with it */
4703 /* We don't need to copy at this time (it'll get done later) */
4704 return list_nth(context->args, param->paramid - 1);
4705 }
4706 return expression_tree_mutator(node, substitute_actual_parameters_mutator,
4707 (void *) context);
4708 }
4709
4710 /*
4711 * error context callback to let us supply a call-stack traceback
4712 */
4713 static void
sql_inline_error_callback(void * arg)4714 sql_inline_error_callback(void *arg)
4715 {
4716 inline_error_callback_arg *callback_arg = (inline_error_callback_arg *) arg;
4717 int syntaxerrposition;
4718
4719 /* If it's a syntax error, convert to internal syntax error report */
4720 syntaxerrposition = geterrposition();
4721 if (syntaxerrposition > 0)
4722 {
4723 errposition(0);
4724 internalerrposition(syntaxerrposition);
4725 internalerrquery(callback_arg->prosrc);
4726 }
4727
4728 errcontext("SQL function \"%s\" during inlining", callback_arg->proname);
4729 }
4730
4731 /*
4732 * evaluate_expr: pre-evaluate a constant expression
4733 *
4734 * We use the executor's routine ExecEvalExpr() to avoid duplication of
4735 * code and ensure we get the same result as the executor would get.
4736 */
4737 static Expr *
evaluate_expr(Expr * expr,Oid result_type,int32 result_typmod,Oid result_collation)4738 evaluate_expr(Expr *expr, Oid result_type, int32 result_typmod,
4739 Oid result_collation)
4740 {
4741 EState *estate;
4742 ExprState *exprstate;
4743 MemoryContext oldcontext;
4744 Datum const_val;
4745 bool const_is_null;
4746 int16 resultTypLen;
4747 bool resultTypByVal;
4748
4749 /*
4750 * To use the executor, we need an EState.
4751 */
4752 estate = CreateExecutorState();
4753
4754 /* We can use the estate's working context to avoid memory leaks. */
4755 oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
4756
4757 /* Make sure any opfuncids are filled in. */
4758 fix_opfuncids((Node *) expr);
4759
4760 /*
4761 * Prepare expr for execution. (Note: we can't use ExecPrepareExpr
4762 * because it'd result in recursively invoking eval_const_expressions.)
4763 */
4764 exprstate = ExecInitExpr(expr, NULL);
4765
4766 /*
4767 * And evaluate it.
4768 *
4769 * It is OK to use a default econtext because none of the ExecEvalExpr()
4770 * code used in this situation will use econtext. That might seem
4771 * fortuitous, but it's not so unreasonable --- a constant expression does
4772 * not depend on context, by definition, n'est ce pas?
4773 */
4774 const_val = ExecEvalExprSwitchContext(exprstate,
4775 GetPerTupleExprContext(estate),
4776 &const_is_null, NULL);
4777
4778 /* Get info needed about result datatype */
4779 get_typlenbyval(result_type, &resultTypLen, &resultTypByVal);
4780
4781 /* Get back to outer memory context */
4782 MemoryContextSwitchTo(oldcontext);
4783
4784 /*
4785 * Must copy result out of sub-context used by expression eval.
4786 *
4787 * Also, if it's varlena, forcibly detoast it. This protects us against
4788 * storing TOAST pointers into plans that might outlive the referenced
4789 * data. (makeConst would handle detoasting anyway, but it's worth a few
4790 * extra lines here so that we can do the copy and detoast in one step.)
4791 */
4792 if (!const_is_null)
4793 {
4794 if (resultTypLen == -1)
4795 const_val = PointerGetDatum(PG_DETOAST_DATUM_COPY(const_val));
4796 else
4797 const_val = datumCopy(const_val, resultTypByVal, resultTypLen);
4798 }
4799
4800 /* Release all the junk we just created */
4801 FreeExecutorState(estate);
4802
4803 /*
4804 * Make the constant result node.
4805 */
4806 return (Expr *) makeConst(result_type, result_typmod, result_collation,
4807 resultTypLen,
4808 const_val, const_is_null,
4809 resultTypByVal);
4810 }
4811
4812
4813 /*
4814 * inline_set_returning_function
4815 * Attempt to "inline" a set-returning function in the FROM clause.
4816 *
4817 * "rte" is an RTE_FUNCTION rangetable entry. If it represents a call of a
4818 * set-returning SQL function that can safely be inlined, expand the function
4819 * and return the substitute Query structure. Otherwise, return NULL.
4820 *
4821 * This has a good deal of similarity to inline_function(), but that's
4822 * for the non-set-returning case, and there are enough differences to
4823 * justify separate functions.
4824 */
4825 Query *
inline_set_returning_function(PlannerInfo * root,RangeTblEntry * rte)4826 inline_set_returning_function(PlannerInfo *root, RangeTblEntry *rte)
4827 {
4828 RangeTblFunction *rtfunc;
4829 FuncExpr *fexpr;
4830 Oid func_oid;
4831 HeapTuple func_tuple;
4832 Form_pg_proc funcform;
4833 char *src;
4834 Datum tmp;
4835 bool isNull;
4836 bool modifyTargetList;
4837 MemoryContext oldcxt;
4838 MemoryContext mycxt;
4839 List *saveInvalItems;
4840 inline_error_callback_arg callback_arg;
4841 ErrorContextCallback sqlerrcontext;
4842 SQLFunctionParseInfoPtr pinfo;
4843 List *raw_parsetree_list;
4844 List *querytree_list;
4845 Query *querytree;
4846
4847 Assert(rte->rtekind == RTE_FUNCTION);
4848
4849 /*
4850 * It doesn't make a lot of sense for a SQL SRF to refer to itself in its
4851 * own FROM clause, since that must cause infinite recursion at runtime.
4852 * It will cause this code to recurse too, so check for stack overflow.
4853 * (There's no need to do more.)
4854 */
4855 check_stack_depth();
4856
4857 /* Fail if the RTE has ORDINALITY - we don't implement that here. */
4858 if (rte->funcordinality)
4859 return NULL;
4860
4861 /* Fail if RTE isn't a single, simple FuncExpr */
4862 if (list_length(rte->functions) != 1)
4863 return NULL;
4864 rtfunc = (RangeTblFunction *) linitial(rte->functions);
4865
4866 if (!IsA(rtfunc->funcexpr, FuncExpr))
4867 return NULL;
4868 fexpr = (FuncExpr *) rtfunc->funcexpr;
4869
4870 func_oid = fexpr->funcid;
4871
4872 /*
4873 * The function must be declared to return a set, else inlining would
4874 * change the results if the contained SELECT didn't return exactly one
4875 * row.
4876 */
4877 if (!fexpr->funcretset)
4878 return NULL;
4879
4880 /*
4881 * Refuse to inline if the arguments contain any volatile functions or
4882 * sub-selects. Volatile functions are rejected because inlining may
4883 * result in the arguments being evaluated multiple times, risking a
4884 * change in behavior. Sub-selects are rejected partly for implementation
4885 * reasons (pushing them down another level might change their behavior)
4886 * and partly because they're likely to be expensive and so multiple
4887 * evaluation would be bad.
4888 */
4889 if (contain_volatile_functions((Node *) fexpr->args) ||
4890 contain_subplans((Node *) fexpr->args))
4891 return NULL;
4892
4893 /* Check permission to call function (fail later, if not) */
4894 if (pg_proc_aclcheck(func_oid, GetUserId(), ACL_EXECUTE) != ACLCHECK_OK)
4895 return NULL;
4896
4897 /* Check whether a plugin wants to hook function entry/exit */
4898 if (FmgrHookIsNeeded(func_oid))
4899 return NULL;
4900
4901 /*
4902 * OK, let's take a look at the function's pg_proc entry.
4903 */
4904 func_tuple = SearchSysCache1(PROCOID, ObjectIdGetDatum(func_oid));
4905 if (!HeapTupleIsValid(func_tuple))
4906 elog(ERROR, "cache lookup failed for function %u", func_oid);
4907 funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4908
4909 /*
4910 * Forget it if the function is not SQL-language or has other showstopper
4911 * properties. In particular it mustn't be declared STRICT, since we
4912 * couldn't enforce that. It also mustn't be VOLATILE, because that is
4913 * supposed to cause it to be executed with its own snapshot, rather than
4914 * sharing the snapshot of the calling query. (Rechecking proretset is
4915 * just paranoia.)
4916 */
4917 if (funcform->prolang != SQLlanguageId ||
4918 funcform->proisstrict ||
4919 funcform->provolatile == PROVOLATILE_VOLATILE ||
4920 funcform->prosecdef ||
4921 !funcform->proretset ||
4922 !heap_attisnull(func_tuple, Anum_pg_proc_proconfig))
4923 {
4924 ReleaseSysCache(func_tuple);
4925 return NULL;
4926 }
4927
4928 /*
4929 * Make a temporary memory context, so that we don't leak all the stuff
4930 * that parsing might create.
4931 */
4932 mycxt = AllocSetContextCreate(CurrentMemoryContext,
4933 "inline_set_returning_function",
4934 ALLOCSET_DEFAULT_SIZES);
4935 oldcxt = MemoryContextSwitchTo(mycxt);
4936
4937 /*
4938 * When we call eval_const_expressions below, it might try to add items to
4939 * root->glob->invalItems. Since it is running in the temp context, those
4940 * items will be in that context, and will need to be copied out if we're
4941 * successful. Temporarily reset the list so that we can keep those items
4942 * separate from the pre-existing list contents.
4943 */
4944 saveInvalItems = root->glob->invalItems;
4945 root->glob->invalItems = NIL;
4946
4947 /* Fetch the function body */
4948 tmp = SysCacheGetAttr(PROCOID,
4949 func_tuple,
4950 Anum_pg_proc_prosrc,
4951 &isNull);
4952 if (isNull)
4953 elog(ERROR, "null prosrc for function %u", func_oid);
4954 src = TextDatumGetCString(tmp);
4955
4956 /*
4957 * Setup error traceback support for ereport(). This is so that we can
4958 * finger the function that bad information came from.
4959 */
4960 callback_arg.proname = NameStr(funcform->proname);
4961 callback_arg.prosrc = src;
4962
4963 sqlerrcontext.callback = sql_inline_error_callback;
4964 sqlerrcontext.arg = (void *) &callback_arg;
4965 sqlerrcontext.previous = error_context_stack;
4966 error_context_stack = &sqlerrcontext;
4967
4968 /*
4969 * Run eval_const_expressions on the function call. This is necessary to
4970 * ensure that named-argument notation is converted to positional notation
4971 * and any default arguments are inserted. It's a bit of overkill for the
4972 * arguments, since they'll get processed again later, but no harm will be
4973 * done.
4974 */
4975 fexpr = (FuncExpr *) eval_const_expressions(root, (Node *) fexpr);
4976
4977 /* It should still be a call of the same function, but let's check */
4978 if (!IsA(fexpr, FuncExpr) ||
4979 fexpr->funcid != func_oid)
4980 goto fail;
4981
4982 /* Arg list length should now match the function */
4983 if (list_length(fexpr->args) != funcform->pronargs)
4984 goto fail;
4985
4986 /*
4987 * Set up to handle parameters while parsing the function body. We can
4988 * use the FuncExpr just created as the input for
4989 * prepare_sql_fn_parse_info.
4990 */
4991 pinfo = prepare_sql_fn_parse_info(func_tuple,
4992 (Node *) fexpr,
4993 fexpr->inputcollid);
4994
4995 /*
4996 * Parse, analyze, and rewrite (unlike inline_function(), we can't skip
4997 * rewriting here). We can fail as soon as we find more than one query,
4998 * though.
4999 */
5000 raw_parsetree_list = pg_parse_query(src);
5001 if (list_length(raw_parsetree_list) != 1)
5002 goto fail;
5003
5004 querytree_list = pg_analyze_and_rewrite_params(linitial(raw_parsetree_list),
5005 src,
5006 (ParserSetupHook) sql_fn_parser_setup,
5007 pinfo);
5008 if (list_length(querytree_list) != 1)
5009 goto fail;
5010 querytree = linitial(querytree_list);
5011
5012 /*
5013 * The single command must be a plain SELECT.
5014 */
5015 if (!IsA(querytree, Query) ||
5016 querytree->commandType != CMD_SELECT ||
5017 querytree->utilityStmt)
5018 goto fail;
5019
5020 /*
5021 * Make sure the function (still) returns what it's declared to. This
5022 * will raise an error if wrong, but that's okay since the function would
5023 * fail at runtime anyway. Note that check_sql_fn_retval will also insert
5024 * RelabelType(s) and/or NULL columns if needed to make the tlist
5025 * expression(s) match the declared type of the function.
5026 *
5027 * If the function returns a composite type, don't inline unless the check
5028 * shows it's returning a whole tuple result; otherwise what it's
5029 * returning is a single composite column which is not what we need.
5030 */
5031 if (!check_sql_fn_retval(func_oid, fexpr->funcresulttype,
5032 querytree_list,
5033 &modifyTargetList, NULL) &&
5034 (get_typtype(fexpr->funcresulttype) == TYPTYPE_COMPOSITE ||
5035 fexpr->funcresulttype == RECORDOID))
5036 goto fail; /* reject not-whole-tuple-result cases */
5037
5038 /*
5039 * If we had to modify the tlist to make it match, and the statement is
5040 * one in which changing the tlist contents could change semantics, we
5041 * have to punt and not inline.
5042 */
5043 if (modifyTargetList)
5044 goto fail;
5045
5046 /*
5047 * If it returns RECORD, we have to check against the column type list
5048 * provided in the RTE; check_sql_fn_retval can't do that. (If no match,
5049 * we just fail to inline, rather than complaining; see notes for
5050 * tlist_matches_coltypelist.) We don't have to do this for functions
5051 * with declared OUT parameters, even though their funcresulttype is
5052 * RECORDOID, so check get_func_result_type too.
5053 */
5054 if (fexpr->funcresulttype == RECORDOID &&
5055 get_func_result_type(func_oid, NULL, NULL) == TYPEFUNC_RECORD &&
5056 !tlist_matches_coltypelist(querytree->targetList,
5057 rtfunc->funccoltypes))
5058 goto fail;
5059
5060 /*
5061 * Looks good --- substitute parameters into the query.
5062 */
5063 querytree = substitute_actual_srf_parameters(querytree,
5064 funcform->pronargs,
5065 fexpr->args);
5066
5067 /*
5068 * Copy the modified query out of the temporary memory context, and clean
5069 * up.
5070 */
5071 MemoryContextSwitchTo(oldcxt);
5072
5073 querytree = copyObject(querytree);
5074
5075 /* copy up any new invalItems, too */
5076 root->glob->invalItems = list_concat(saveInvalItems,
5077 copyObject(root->glob->invalItems));
5078
5079 MemoryContextDelete(mycxt);
5080 error_context_stack = sqlerrcontext.previous;
5081 ReleaseSysCache(func_tuple);
5082
5083 /*
5084 * We don't have to fix collations here because the upper query is already
5085 * parsed, ie, the collations in the RTE are what count.
5086 */
5087
5088 /*
5089 * Since there is now no trace of the function in the plan tree, we must
5090 * explicitly record the plan's dependency on the function.
5091 */
5092 record_plan_function_dependency(root, func_oid);
5093
5094 return querytree;
5095
5096 /* Here if func is not inlinable: release temp memory and return NULL */
5097 fail:
5098 MemoryContextSwitchTo(oldcxt);
5099 root->glob->invalItems = saveInvalItems;
5100 MemoryContextDelete(mycxt);
5101 error_context_stack = sqlerrcontext.previous;
5102 ReleaseSysCache(func_tuple);
5103
5104 return NULL;
5105 }
5106
5107 /*
5108 * Replace Param nodes by appropriate actual parameters
5109 *
5110 * This is just enough different from substitute_actual_parameters()
5111 * that it needs its own code.
5112 */
5113 static Query *
substitute_actual_srf_parameters(Query * expr,int nargs,List * args)5114 substitute_actual_srf_parameters(Query *expr, int nargs, List *args)
5115 {
5116 substitute_actual_srf_parameters_context context;
5117
5118 context.nargs = nargs;
5119 context.args = args;
5120 context.sublevels_up = 1;
5121
5122 return query_tree_mutator(expr,
5123 substitute_actual_srf_parameters_mutator,
5124 &context,
5125 0);
5126 }
5127
5128 static Node *
substitute_actual_srf_parameters_mutator(Node * node,substitute_actual_srf_parameters_context * context)5129 substitute_actual_srf_parameters_mutator(Node *node,
5130 substitute_actual_srf_parameters_context *context)
5131 {
5132 Node *result;
5133
5134 if (node == NULL)
5135 return NULL;
5136 if (IsA(node, Query))
5137 {
5138 context->sublevels_up++;
5139 result = (Node *) query_tree_mutator((Query *) node,
5140 substitute_actual_srf_parameters_mutator,
5141 (void *) context,
5142 0);
5143 context->sublevels_up--;
5144 return result;
5145 }
5146 if (IsA(node, Param))
5147 {
5148 Param *param = (Param *) node;
5149
5150 if (param->paramkind == PARAM_EXTERN)
5151 {
5152 if (param->paramid <= 0 || param->paramid > context->nargs)
5153 elog(ERROR, "invalid paramid: %d", param->paramid);
5154
5155 /*
5156 * Since the parameter is being inserted into a subquery, we must
5157 * adjust levels.
5158 */
5159 result = copyObject(list_nth(context->args, param->paramid - 1));
5160 IncrementVarSublevelsUp(result, context->sublevels_up, 0);
5161 return result;
5162 }
5163 }
5164 return expression_tree_mutator(node,
5165 substitute_actual_srf_parameters_mutator,
5166 (void *) context);
5167 }
5168
5169 /*
5170 * Check whether a SELECT targetlist emits the specified column types,
5171 * to see if it's safe to inline a function returning record.
5172 *
5173 * We insist on exact match here. The executor allows binary-coercible
5174 * cases too, but we don't have a way to preserve the correct column types
5175 * in the correct places if we inline the function in such a case.
5176 *
5177 * Note that we only check type OIDs not typmods; this agrees with what the
5178 * executor would do at runtime, and attributing a specific typmod to a
5179 * function result is largely wishful thinking anyway.
5180 */
5181 static bool
tlist_matches_coltypelist(List * tlist,List * coltypelist)5182 tlist_matches_coltypelist(List *tlist, List *coltypelist)
5183 {
5184 ListCell *tlistitem;
5185 ListCell *clistitem;
5186
5187 clistitem = list_head(coltypelist);
5188 foreach(tlistitem, tlist)
5189 {
5190 TargetEntry *tle = (TargetEntry *) lfirst(tlistitem);
5191 Oid coltype;
5192
5193 if (tle->resjunk)
5194 continue; /* ignore junk columns */
5195
5196 if (clistitem == NULL)
5197 return false; /* too many tlist items */
5198
5199 coltype = lfirst_oid(clistitem);
5200 clistitem = lnext(clistitem);
5201
5202 if (exprType((Node *) tle->expr) != coltype)
5203 return false; /* column type mismatch */
5204 }
5205
5206 if (clistitem != NULL)
5207 return false; /* too few tlist items */
5208
5209 return true;
5210 }
5211