PostgreSQL Source Code  git master
clauses.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * clauses.c
4  * routines to manipulate qualification clauses
5  *
6  * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
7  * Portions Copyright (c) 1994, Regents of the University of California
8  *
9  *
10  * IDENTIFICATION
11  * src/backend/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_language.h"
24 #include "catalog/pg_operator.h"
25 #include "catalog/pg_proc.h"
26 #include "catalog/pg_type.h"
27 #include "executor/executor.h"
28 #include "executor/functions.h"
29 #include "funcapi.h"
30 #include "miscadmin.h"
31 #include "nodes/makefuncs.h"
32 #include "nodes/multibitmapset.h"
33 #include "nodes/nodeFuncs.h"
34 #include "nodes/subscripting.h"
35 #include "nodes/supportnodes.h"
36 #include "optimizer/clauses.h"
37 #include "optimizer/cost.h"
38 #include "optimizer/optimizer.h"
39 #include "optimizer/plancat.h"
40 #include "optimizer/planmain.h"
41 #include "parser/analyze.h"
42 #include "parser/parse_coerce.h"
43 #include "parser/parse_func.h"
44 #include "rewrite/rewriteHandler.h"
45 #include "rewrite/rewriteManip.h"
46 #include "tcop/tcopprot.h"
47 #include "utils/acl.h"
48 #include "utils/builtins.h"
49 #include "utils/datum.h"
50 #include "utils/fmgroids.h"
51 #include "utils/json.h"
52 #include "utils/jsonb.h"
53 #include "utils/jsonpath.h"
54 #include "utils/lsyscache.h"
55 #include "utils/memutils.h"
56 #include "utils/syscache.h"
57 #include "utils/typcache.h"
58 
59 typedef struct
60 {
65  bool estimate;
67 
68 typedef struct
69 {
70  int nargs;
72  int *usecounts;
74 
75 typedef struct
76 {
77  int nargs;
81 
82 typedef struct
83 {
84  char *proname;
85  char *prosrc;
87 
88 typedef struct
89 {
90  char max_hazard; /* worst proparallel hazard found so far */
91  char max_interesting; /* worst proparallel hazard of interest */
92  List *safe_param_ids; /* PARAM_EXEC Param IDs to treat as safe */
94 
95 static bool contain_agg_clause_walker(Node *node, void *context);
96 static bool find_window_functions_walker(Node *node, WindowFuncLists *lists);
97 static bool contain_subplans_walker(Node *node, void *context);
98 static bool contain_mutable_functions_walker(Node *node, void *context);
99 static bool contain_volatile_functions_walker(Node *node, void *context);
101 static bool max_parallel_hazard_walker(Node *node,
103 static bool contain_nonstrict_functions_walker(Node *node, void *context);
104 static bool contain_exec_param_walker(Node *node, List *param_ids);
105 static bool contain_context_dependent_node(Node *clause);
106 static bool contain_context_dependent_node_walker(Node *node, int *flags);
107 static bool contain_leaked_vars_walker(Node *node, void *context);
108 static Relids find_nonnullable_rels_walker(Node *node, bool top_level);
109 static List *find_nonnullable_vars_walker(Node *node, bool top_level);
110 static bool is_strict_saop(ScalarArrayOpExpr *expr, bool falseOK);
111 static bool convert_saop_to_hashed_saop_walker(Node *node, void *context);
114 static bool contain_non_const_walker(Node *node, void *context);
115 static bool ece_function_is_safe(Oid funcid,
119  bool *haveNull, bool *forceTrue);
122  bool *haveNull, bool *forceFalse);
123 static Node *simplify_boolean_equality(Oid opno, List *args);
124 static Expr *simplify_function(Oid funcid,
125  Oid result_type, int32 result_typmod,
126  Oid result_collid, Oid input_collid, List **args_p,
127  bool funcvariadic, bool process_args, bool allow_non_const,
130  HeapTuple func_tuple);
132  HeapTuple func_tuple);
133 static List *fetch_function_defaults(HeapTuple func_tuple);
134 static void recheck_cast_function_args(List *args, Oid result_type,
135  Oid *proargtypes, int pronargs,
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,
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,
147 static Node *substitute_actual_parameters(Node *expr, int nargs, List *args,
148  int *usecounts);
151 static void sql_inline_error_callback(void *arg);
153  int nargs, List *args);
156 static bool pull_paramids_walker(Node *node, Bitmapset **context);
157 
158 
159 /*****************************************************************************
160  * Aggregate-function clause manipulation
161  *****************************************************************************/
162 
163 /*
164  * contain_agg_clause
165  * Recursively search for Aggref/GroupingFunc nodes within a clause.
166  *
167  * Returns true if any aggregate found.
168  *
169  * This does not descend into subqueries, and so should be used only after
170  * reduction of sublinks to subplans, or in contexts where it's known there
171  * are no subqueries. There mustn't be outer-aggregate references either.
172  *
173  * (If you want something like this but able to deal with subqueries,
174  * see rewriteManip.c's contain_aggs_of_level().)
175  */
176 bool
178 {
179  return contain_agg_clause_walker(clause, NULL);
180 }
181 
182 static bool
184 {
185  if (node == NULL)
186  return false;
187  if (IsA(node, Aggref))
188  {
189  Assert(((Aggref *) node)->agglevelsup == 0);
190  return true; /* abort the tree traversal and return true */
191  }
192  if (IsA(node, GroupingFunc))
193  {
194  Assert(((GroupingFunc *) node)->agglevelsup == 0);
195  return true; /* abort the tree traversal and return true */
196  }
197  Assert(!IsA(node, SubLink));
199 }
200 
201 /*****************************************************************************
202  * Window-function clause manipulation
203  *****************************************************************************/
204 
205 /*
206  * contain_window_function
207  * Recursively search for WindowFunc nodes within a clause.
208  *
209  * Since window functions don't have level fields, but are hard-wired to
210  * be associated with the current query level, this is just the same as
211  * rewriteManip.c's function.
212  */
213 bool
215 {
216  return contain_windowfuncs(clause);
217 }
218 
219 /*
220  * find_window_functions
221  * Locate all the WindowFunc nodes in an expression tree, and organize
222  * them by winref ID number.
223  *
224  * Caller must provide an upper bound on the winref IDs expected in the tree.
225  */
227 find_window_functions(Node *clause, Index maxWinRef)
228 {
229  WindowFuncLists *lists = palloc(sizeof(WindowFuncLists));
230 
231  lists->numWindowFuncs = 0;
232  lists->maxWinRef = maxWinRef;
233  lists->windowFuncs = (List **) palloc0((maxWinRef + 1) * sizeof(List *));
234  (void) find_window_functions_walker(clause, lists);
235  return lists;
236 }
237 
238 static bool
240 {
241  if (node == NULL)
242  return false;
243  if (IsA(node, WindowFunc))
244  {
245  WindowFunc *wfunc = (WindowFunc *) node;
246 
247  /* winref is unsigned, so one-sided test is OK */
248  if (wfunc->winref > lists->maxWinRef)
249  elog(ERROR, "WindowFunc contains out-of-range winref %u",
250  wfunc->winref);
251  /* eliminate duplicates, so that we avoid repeated computation */
252  if (!list_member(lists->windowFuncs[wfunc->winref], wfunc))
253  {
254  lists->windowFuncs[wfunc->winref] =
255  lappend(lists->windowFuncs[wfunc->winref], wfunc);
256  lists->numWindowFuncs++;
257  }
258 
259  /*
260  * We assume that the parser checked that there are no window
261  * functions in the arguments or filter clause. Hence, we need not
262  * recurse into them. (If either the parser or the planner screws up
263  * on this point, the executor will still catch it; see ExecInitExpr.)
264  */
265  return false;
266  }
267  Assert(!IsA(node, SubLink));
269  (void *) lists);
270 }
271 
272 
273 /*****************************************************************************
274  * Support for expressions returning sets
275  *****************************************************************************/
276 
277 /*
278  * expression_returns_set_rows
279  * Estimate the number of rows returned by a set-returning expression.
280  * The result is 1 if it's not a set-returning expression.
281  *
282  * We should only examine the top-level function or operator; it used to be
283  * appropriate to recurse, but not anymore. (Even if there are more SRFs in
284  * the function's inputs, their multipliers are accounted for separately.)
285  *
286  * Note: keep this in sync with expression_returns_set() in nodes/nodeFuncs.c.
287  */
288 double
290 {
291  if (clause == NULL)
292  return 1.0;
293  if (IsA(clause, FuncExpr))
294  {
295  FuncExpr *expr = (FuncExpr *) clause;
296 
297  if (expr->funcretset)
298  return clamp_row_est(get_function_rows(root, expr->funcid, clause));
299  }
300  if (IsA(clause, OpExpr))
301  {
302  OpExpr *expr = (OpExpr *) clause;
303 
304  if (expr->opretset)
305  {
306  set_opfuncid(expr);
307  return clamp_row_est(get_function_rows(root, expr->opfuncid, clause));
308  }
309  }
310  return 1.0;
311 }
312 
313 
314 /*****************************************************************************
315  * Subplan clause manipulation
316  *****************************************************************************/
317 
318 /*
319  * contain_subplans
320  * Recursively search for subplan nodes within a clause.
321  *
322  * If we see a SubLink node, we will return true. This is only possible if
323  * the expression tree hasn't yet been transformed by subselect.c. We do not
324  * know whether the node will produce a true subplan or just an initplan,
325  * but we make the conservative assumption that it will be a subplan.
326  *
327  * Returns true if any subplan found.
328  */
329 bool
331 {
332  return contain_subplans_walker(clause, NULL);
333 }
334 
335 static bool
337 {
338  if (node == NULL)
339  return false;
340  if (IsA(node, SubPlan) ||
341  IsA(node, AlternativeSubPlan) ||
342  IsA(node, SubLink))
343  return true; /* abort the tree traversal and return true */
345 }
346 
347 
348 /*****************************************************************************
349  * Check clauses for mutable functions
350  *****************************************************************************/
351 
352 /*
353  * contain_mutable_functions
354  * Recursively search for mutable functions within a clause.
355  *
356  * Returns true if any mutable function (or operator implemented by a
357  * mutable function) is found. This test is needed so that we don't
358  * mistakenly think that something like "WHERE random() < 0.5" can be treated
359  * as a constant qualification.
360  *
361  * This will give the right answer only for clauses that have been put
362  * through expression preprocessing. Callers outside the planner typically
363  * should use contain_mutable_functions_after_planning() instead, for the
364  * reasons given there.
365  *
366  * We will recursively look into Query nodes (i.e., SubLink sub-selects)
367  * but not into SubPlans. See comments for contain_volatile_functions().
368  */
369 bool
371 {
372  return contain_mutable_functions_walker(clause, NULL);
373 }
374 
375 static bool
377 {
378  return (func_volatile(func_id) != PROVOLATILE_IMMUTABLE);
379 }
380 
381 static bool
383 {
384  if (node == NULL)
385  return false;
386  /* Check for mutable functions in node itself */
388  context))
389  return true;
390 
391  if (IsA(node, JsonConstructorExpr))
392  {
393  const JsonConstructorExpr *ctor = (JsonConstructorExpr *) node;
394  ListCell *lc;
395  bool is_jsonb;
396 
397  is_jsonb = ctor->returning->format->format_type == JS_FORMAT_JSONB;
398 
399  /*
400  * Check argument_type => json[b] conversions specifically. We still
401  * recurse to check 'args' below, but here we want to specifically
402  * check whether or not the emitted clause would fail to be immutable
403  * because of TimeZone, for example.
404  */
405  foreach(lc, ctor->args)
406  {
407  Oid typid = exprType(lfirst(lc));
408 
409  if (is_jsonb ?
410  !to_jsonb_is_immutable(typid) :
411  !to_json_is_immutable(typid))
412  return true;
413  }
414 
415  /* Check all subnodes */
416  }
417 
418  if (IsA(node, JsonExpr))
419  {
420  JsonExpr *jexpr = castNode(JsonExpr, node);
421  Const *cnst;
422 
423  if (!IsA(jexpr->path_spec, Const))
424  return true;
425 
426  cnst = castNode(Const, jexpr->path_spec);
427 
428  Assert(cnst->consttype == JSONPATHOID);
429  if (cnst->constisnull)
430  return false;
431 
432  if (jspIsMutable(DatumGetJsonPathP(cnst->constvalue),
433  jexpr->passing_names, jexpr->passing_values))
434  return true;
435  }
436 
437  if (IsA(node, SQLValueFunction))
438  {
439  /* all variants of SQLValueFunction are stable */
440  return true;
441  }
442 
443  if (IsA(node, NextValueExpr))
444  {
445  /* NextValueExpr is volatile */
446  return true;
447  }
448 
449  /*
450  * It should be safe to treat MinMaxExpr as immutable, because it will
451  * depend on a non-cross-type btree comparison function, and those should
452  * always be immutable. Treating XmlExpr as immutable is more dubious,
453  * and treating CoerceToDomain as immutable is outright dangerous. But we
454  * have done so historically, and changing this would probably cause more
455  * problems than it would fix. In practice, if you have a non-immutable
456  * domain constraint you are in for pain anyhow.
457  */
458 
459  /* Recurse to check arguments */
460  if (IsA(node, Query))
461  {
462  /* Recurse into subselects */
463  return query_tree_walker((Query *) node,
465  context, 0);
466  }
468  context);
469 }
470 
471 /*
472  * contain_mutable_functions_after_planning
473  * Test whether given expression contains mutable functions.
474  *
475  * This is a wrapper for contain_mutable_functions() that is safe to use from
476  * outside the planner. The difference is that it first runs the expression
477  * through expression_planner(). There are two key reasons why we need that:
478  *
479  * First, function default arguments will get inserted, which may affect
480  * volatility (consider "default now()").
481  *
482  * Second, inline-able functions will get inlined, which may allow us to
483  * conclude that the function is really less volatile than it's marked.
484  * As an example, polymorphic functions must be marked with the most volatile
485  * behavior that they have for any input type, but once we inline the
486  * function we may be able to conclude that it's not so volatile for the
487  * particular input type we're dealing with.
488  */
489 bool
491 {
492  /* We assume here that expression_planner() won't scribble on its input */
493  expr = expression_planner(expr);
494 
495  /* Now we can search for non-immutable functions */
496  return contain_mutable_functions((Node *) expr);
497 }
498 
499 
500 /*****************************************************************************
501  * Check clauses for volatile functions
502  *****************************************************************************/
503 
504 /*
505  * contain_volatile_functions
506  * Recursively search for volatile functions within a clause.
507  *
508  * Returns true if any volatile function (or operator implemented by a
509  * volatile function) is found. This test prevents, for example,
510  * invalid conversions of volatile expressions into indexscan quals.
511  *
512  * This will give the right answer only for clauses that have been put
513  * through expression preprocessing. Callers outside the planner typically
514  * should use contain_volatile_functions_after_planning() instead, for the
515  * reasons given there.
516  *
517  * We will recursively look into Query nodes (i.e., SubLink sub-selects)
518  * but not into SubPlans. This is a bit odd, but intentional. If we are
519  * looking at a SubLink, we are probably deciding whether a query tree
520  * transformation is safe, and a contained sub-select should affect that;
521  * for example, duplicating a sub-select containing a volatile function
522  * would be bad. However, once we've got to the stage of having SubPlans,
523  * subsequent planning need not consider volatility within those, since
524  * the executor won't change its evaluation rules for a SubPlan based on
525  * volatility.
526  *
527  * For some node types, for example, RestrictInfo and PathTarget, we cache
528  * whether we found any volatile functions or not and reuse that value in any
529  * future checks for that node. All of the logic for determining if the
530  * cached value should be set to VOLATILITY_NOVOLATILE or VOLATILITY_VOLATILE
531  * belongs in this function. Any code which makes changes to these nodes
532  * which could change the outcome this function must set the cached value back
533  * to VOLATILITY_UNKNOWN. That allows this function to redetermine the
534  * correct value during the next call, should we need to redetermine if the
535  * node contains any volatile functions again in the future.
536  */
537 bool
539 {
540  return contain_volatile_functions_walker(clause, NULL);
541 }
542 
543 static bool
545 {
546  return (func_volatile(func_id) == PROVOLATILE_VOLATILE);
547 }
548 
549 static bool
551 {
552  if (node == NULL)
553  return false;
554  /* Check for volatile functions in node itself */
556  context))
557  return true;
558 
559  if (IsA(node, NextValueExpr))
560  {
561  /* NextValueExpr is volatile */
562  return true;
563  }
564 
565  if (IsA(node, RestrictInfo))
566  {
567  RestrictInfo *rinfo = (RestrictInfo *) node;
568 
569  /*
570  * For RestrictInfo, check if we've checked the volatility of it
571  * before. If so, we can just use the cached value and not bother
572  * checking it again. Otherwise, check it and cache if whether we
573  * found any volatile functions.
574  */
575  if (rinfo->has_volatile == VOLATILITY_NOVOLATILE)
576  return false;
577  else if (rinfo->has_volatile == VOLATILITY_VOLATILE)
578  return true;
579  else
580  {
581  bool hasvolatile;
582 
583  hasvolatile = contain_volatile_functions_walker((Node *) rinfo->clause,
584  context);
585  if (hasvolatile)
586  rinfo->has_volatile = VOLATILITY_VOLATILE;
587  else
588  rinfo->has_volatile = VOLATILITY_NOVOLATILE;
589 
590  return hasvolatile;
591  }
592  }
593 
594  if (IsA(node, PathTarget))
595  {
596  PathTarget *target = (PathTarget *) node;
597 
598  /*
599  * We also do caching for PathTarget the same as we do above for
600  * RestrictInfos.
601  */
603  return false;
604  else if (target->has_volatile_expr == VOLATILITY_VOLATILE)
605  return true;
606  else
607  {
608  bool hasvolatile;
609 
610  hasvolatile = contain_volatile_functions_walker((Node *) target->exprs,
611  context);
612 
613  if (hasvolatile)
615  else
617 
618  return hasvolatile;
619  }
620  }
621 
622  /*
623  * See notes in contain_mutable_functions_walker about why we treat
624  * MinMaxExpr, XmlExpr, and CoerceToDomain as immutable, while
625  * SQLValueFunction is stable. Hence, none of them are of interest here.
626  */
627 
628  /* Recurse to check arguments */
629  if (IsA(node, Query))
630  {
631  /* Recurse into subselects */
632  return query_tree_walker((Query *) node,
634  context, 0);
635  }
637  context);
638 }
639 
640 /*
641  * contain_volatile_functions_after_planning
642  * Test whether given expression contains volatile functions.
643  *
644  * This is a wrapper for contain_volatile_functions() that is safe to use from
645  * outside the planner. The difference is that it first runs the expression
646  * through expression_planner(). There are two key reasons why we need that:
647  *
648  * First, function default arguments will get inserted, which may affect
649  * volatility (consider "default random()").
650  *
651  * Second, inline-able functions will get inlined, which may allow us to
652  * conclude that the function is really less volatile than it's marked.
653  * As an example, polymorphic functions must be marked with the most volatile
654  * behavior that they have for any input type, but once we inline the
655  * function we may be able to conclude that it's not so volatile for the
656  * particular input type we're dealing with.
657  */
658 bool
660 {
661  /* We assume here that expression_planner() won't scribble on its input */
662  expr = expression_planner(expr);
663 
664  /* Now we can search for volatile functions */
665  return contain_volatile_functions((Node *) expr);
666 }
667 
668 /*
669  * Special purpose version of contain_volatile_functions() for use in COPY:
670  * ignore nextval(), but treat all other functions normally.
671  */
672 bool
674 {
676 }
677 
678 static bool
680 {
681  return (func_id != F_NEXTVAL &&
682  func_volatile(func_id) == PROVOLATILE_VOLATILE);
683 }
684 
685 static bool
687 {
688  if (node == NULL)
689  return false;
690  /* Check for volatile functions in node itself */
691  if (check_functions_in_node(node,
693  context))
694  return true;
695 
696  /*
697  * See notes in contain_mutable_functions_walker about why we treat
698  * MinMaxExpr, XmlExpr, and CoerceToDomain as immutable, while
699  * SQLValueFunction is stable. Hence, none of them are of interest here.
700  * Also, since we're intentionally ignoring nextval(), presumably we
701  * should ignore NextValueExpr.
702  */
703 
704  /* Recurse to check arguments */
705  if (IsA(node, Query))
706  {
707  /* Recurse into subselects */
708  return query_tree_walker((Query *) node,
710  context, 0);
711  }
712  return expression_tree_walker(node,
714  context);
715 }
716 
717 
718 /*****************************************************************************
719  * Check queries for parallel unsafe and/or restricted constructs
720  *****************************************************************************/
721 
722 /*
723  * max_parallel_hazard
724  * Find the worst parallel-hazard level in the given query
725  *
726  * Returns the worst function hazard property (the earliest in this list:
727  * PROPARALLEL_UNSAFE, PROPARALLEL_RESTRICTED, PROPARALLEL_SAFE) that can
728  * be found in the given parsetree. We use this to find out whether the query
729  * can be parallelized at all. The caller will also save the result in
730  * PlannerGlobal so as to short-circuit checks of portions of the querytree
731  * later, in the common case where everything is SAFE.
732  */
733 char
735 {
737 
738  context.max_hazard = PROPARALLEL_SAFE;
739  context.max_interesting = PROPARALLEL_UNSAFE;
740  context.safe_param_ids = NIL;
742  return context.max_hazard;
743 }
744 
745 /*
746  * is_parallel_safe
747  * Detect whether the given expr contains only parallel-safe functions
748  *
749  * root->glob->maxParallelHazard must previously have been set to the
750  * result of max_parallel_hazard() on the whole query.
751  */
752 bool
754 {
756  PlannerInfo *proot;
757  ListCell *l;
758 
759  /*
760  * Even if the original querytree contained nothing unsafe, we need to
761  * search the expression if we have generated any PARAM_EXEC Params while
762  * planning, because those are parallel-restricted and there might be one
763  * in this expression. But otherwise we don't need to look.
764  */
765  if (root->glob->maxParallelHazard == PROPARALLEL_SAFE &&
766  root->glob->paramExecTypes == NIL)
767  return true;
768  /* Else use max_parallel_hazard's search logic, but stop on RESTRICTED */
769  context.max_hazard = PROPARALLEL_SAFE;
770  context.max_interesting = PROPARALLEL_RESTRICTED;
771  context.safe_param_ids = NIL;
772 
773  /*
774  * The params that refer to the same or parent query level are considered
775  * parallel-safe. The idea is that we compute such params at Gather or
776  * Gather Merge node and pass their value to workers.
777  */
778  for (proot = root; proot != NULL; proot = proot->parent_root)
779  {
780  foreach(l, proot->init_plans)
781  {
782  SubPlan *initsubplan = (SubPlan *) lfirst(l);
783 
784  context.safe_param_ids = list_concat(context.safe_param_ids,
785  initsubplan->setParam);
786  }
787  }
788 
789  return !max_parallel_hazard_walker(node, &context);
790 }
791 
792 /* core logic for all parallel-hazard checks */
793 static bool
795 {
796  switch (proparallel)
797  {
798  case PROPARALLEL_SAFE:
799  /* nothing to see here, move along */
800  break;
801  case PROPARALLEL_RESTRICTED:
802  /* increase max_hazard to RESTRICTED */
803  Assert(context->max_hazard != PROPARALLEL_UNSAFE);
804  context->max_hazard = proparallel;
805  /* done if we are not expecting any unsafe functions */
806  if (context->max_interesting == proparallel)
807  return true;
808  break;
809  case PROPARALLEL_UNSAFE:
810  context->max_hazard = proparallel;
811  /* we're always done at the first unsafe construct */
812  return true;
813  default:
814  elog(ERROR, "unrecognized proparallel value \"%c\"", proparallel);
815  break;
816  }
817  return false;
818 }
819 
820 /* check_functions_in_node callback */
821 static bool
823 {
824  return max_parallel_hazard_test(func_parallel(func_id),
826 }
827 
828 static bool
830 {
831  if (node == NULL)
832  return false;
833 
834  /* Check for hazardous functions in node itself */
836  context))
837  return true;
838 
839  /*
840  * It should be OK to treat MinMaxExpr as parallel-safe, since btree
841  * opclass support functions are generally parallel-safe. XmlExpr is a
842  * bit more dubious but we can probably get away with it. We err on the
843  * side of caution by treating CoerceToDomain as parallel-restricted.
844  * (Note: in principle that's wrong because a domain constraint could
845  * contain a parallel-unsafe function; but useful constraints probably
846  * never would have such, and assuming they do would cripple use of
847  * parallel query in the presence of domain types.) SQLValueFunction
848  * should be safe in all cases. NextValueExpr is parallel-unsafe.
849  */
850  if (IsA(node, CoerceToDomain))
851  {
852  if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context))
853  return true;
854  }
855 
856  else if (IsA(node, NextValueExpr))
857  {
858  if (max_parallel_hazard_test(PROPARALLEL_UNSAFE, context))
859  return true;
860  }
861 
862  /*
863  * Treat window functions as parallel-restricted because we aren't sure
864  * whether the input row ordering is fully deterministic, and the output
865  * of window functions might vary across workers if not. (In some cases,
866  * like where the window frame orders by a primary key, we could relax
867  * this restriction. But it doesn't currently seem worth expending extra
868  * effort to do so.)
869  */
870  else if (IsA(node, WindowFunc))
871  {
872  if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context))
873  return true;
874  }
875 
876  /*
877  * As a notational convenience for callers, look through RestrictInfo.
878  */
879  else if (IsA(node, RestrictInfo))
880  {
881  RestrictInfo *rinfo = (RestrictInfo *) node;
882 
883  return max_parallel_hazard_walker((Node *) rinfo->clause, context);
884  }
885 
886  /*
887  * Really we should not see SubLink during a max_interesting == restricted
888  * scan, but if we do, return true.
889  */
890  else if (IsA(node, SubLink))
891  {
892  if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context))
893  return true;
894  }
895 
896  /*
897  * Only parallel-safe SubPlans can be sent to workers. Within the
898  * testexpr of the SubPlan, Params representing the output columns of the
899  * subplan can be treated as parallel-safe, so temporarily add their IDs
900  * to the safe_param_ids list while examining the testexpr.
901  */
902  else if (IsA(node, SubPlan))
903  {
904  SubPlan *subplan = (SubPlan *) node;
905  List *save_safe_param_ids;
906 
907  if (!subplan->parallel_safe &&
908  max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context))
909  return true;
910  save_safe_param_ids = context->safe_param_ids;
911  context->safe_param_ids = list_concat_copy(context->safe_param_ids,
912  subplan->paramIds);
914  return true; /* no need to restore safe_param_ids */
915  list_free(context->safe_param_ids);
916  context->safe_param_ids = save_safe_param_ids;
917  /* we must also check args, but no special Param treatment there */
918  if (max_parallel_hazard_walker((Node *) subplan->args, context))
919  return true;
920  /* don't want to recurse normally, so we're done */
921  return false;
922  }
923 
924  /*
925  * We can't pass Params to workers at the moment either, so they are also
926  * parallel-restricted, unless they are PARAM_EXTERN Params or are
927  * PARAM_EXEC Params listed in safe_param_ids, meaning they could be
928  * either generated within workers or can be computed by the leader and
929  * then their value can be passed to workers.
930  */
931  else if (IsA(node, Param))
932  {
933  Param *param = (Param *) node;
934 
935  if (param->paramkind == PARAM_EXTERN)
936  return false;
937 
938  if (param->paramkind != PARAM_EXEC ||
939  !list_member_int(context->safe_param_ids, param->paramid))
940  {
941  if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context))
942  return true;
943  }
944  return false; /* nothing to recurse to */
945  }
946 
947  /*
948  * When we're first invoked on a completely unplanned tree, we must
949  * recurse into subqueries so to as to locate parallel-unsafe constructs
950  * anywhere in the tree.
951  */
952  else if (IsA(node, Query))
953  {
954  Query *query = (Query *) node;
955 
956  /* SELECT FOR UPDATE/SHARE must be treated as unsafe */
957  if (query->rowMarks != NULL)
958  {
959  context->max_hazard = PROPARALLEL_UNSAFE;
960  return true;
961  }
962 
963  /* Recurse into subselects */
964  return query_tree_walker(query,
966  context, 0);
967  }
968 
969  /* Recurse to check arguments */
970  return expression_tree_walker(node,
972  context);
973 }
974 
975 
976 /*****************************************************************************
977  * Check clauses for nonstrict functions
978  *****************************************************************************/
979 
980 /*
981  * contain_nonstrict_functions
982  * Recursively search for nonstrict functions within a clause.
983  *
984  * Returns true if any nonstrict construct is found --- ie, anything that
985  * could produce non-NULL output with a NULL input.
986  *
987  * The idea here is that the caller has verified that the expression contains
988  * one or more Var or Param nodes (as appropriate for the caller's need), and
989  * now wishes to prove that the expression result will be NULL if any of these
990  * inputs is NULL. If we return false, then the proof succeeded.
991  */
992 bool
994 {
995  return contain_nonstrict_functions_walker(clause, NULL);
996 }
997 
998 static bool
1000 {
1001  return !func_strict(func_id);
1002 }
1003 
1004 static bool
1006 {
1007  if (node == NULL)
1008  return false;
1009  if (IsA(node, Aggref))
1010  {
1011  /* an aggregate could return non-null with null input */
1012  return true;
1013  }
1014  if (IsA(node, GroupingFunc))
1015  {
1016  /*
1017  * A GroupingFunc doesn't evaluate its arguments, and therefore must
1018  * be treated as nonstrict.
1019  */
1020  return true;
1021  }
1022  if (IsA(node, WindowFunc))
1023  {
1024  /* a window function could return non-null with null input */
1025  return true;
1026  }
1027  if (IsA(node, SubscriptingRef))
1028  {
1029  SubscriptingRef *sbsref = (SubscriptingRef *) node;
1030  const SubscriptRoutines *sbsroutines;
1031 
1032  /* Subscripting assignment is always presumed nonstrict */
1033  if (sbsref->refassgnexpr != NULL)
1034  return true;
1035  /* Otherwise we must look up the subscripting support methods */
1036  sbsroutines = getSubscriptingRoutines(sbsref->refcontainertype, NULL);
1037  if (!(sbsroutines && sbsroutines->fetch_strict))
1038  return true;
1039  /* else fall through to check args */
1040  }
1041  if (IsA(node, DistinctExpr))
1042  {
1043  /* IS DISTINCT FROM is inherently non-strict */
1044  return true;
1045  }
1046  if (IsA(node, NullIfExpr))
1047  {
1048  /* NULLIF is inherently non-strict */
1049  return true;
1050  }
1051  if (IsA(node, BoolExpr))
1052  {
1053  BoolExpr *expr = (BoolExpr *) node;
1054 
1055  switch (expr->boolop)
1056  {
1057  case AND_EXPR:
1058  case OR_EXPR:
1059  /* AND, OR are inherently non-strict */
1060  return true;
1061  default:
1062  break;
1063  }
1064  }
1065  if (IsA(node, SubLink))
1066  {
1067  /* In some cases a sublink might be strict, but in general not */
1068  return true;
1069  }
1070  if (IsA(node, SubPlan))
1071  return true;
1072  if (IsA(node, AlternativeSubPlan))
1073  return true;
1074  if (IsA(node, FieldStore))
1075  return true;
1076  if (IsA(node, CoerceViaIO))
1077  {
1078  /*
1079  * CoerceViaIO is strict regardless of whether the I/O functions are,
1080  * so just go look at its argument; asking check_functions_in_node is
1081  * useless expense and could deliver the wrong answer.
1082  */
1083  return contain_nonstrict_functions_walker((Node *) ((CoerceViaIO *) node)->arg,
1084  context);
1085  }
1086  if (IsA(node, ArrayCoerceExpr))
1087  {
1088  /*
1089  * ArrayCoerceExpr is strict at the array level, regardless of what
1090  * the per-element expression is; so we should ignore elemexpr and
1091  * recurse only into the arg.
1092  */
1094  context);
1095  }
1096  if (IsA(node, CaseExpr))
1097  return true;
1098  if (IsA(node, ArrayExpr))
1099  return true;
1100  if (IsA(node, RowExpr))
1101  return true;
1102  if (IsA(node, RowCompareExpr))
1103  return true;
1104  if (IsA(node, CoalesceExpr))
1105  return true;
1106  if (IsA(node, MinMaxExpr))
1107  return true;
1108  if (IsA(node, XmlExpr))
1109  return true;
1110  if (IsA(node, NullTest))
1111  return true;
1112  if (IsA(node, BooleanTest))
1113  return true;
1114 
1115  /* Check other function-containing nodes */
1117  context))
1118  return true;
1119 
1121  context);
1122 }
1123 
1124 /*****************************************************************************
1125  * Check clauses for Params
1126  *****************************************************************************/
1127 
1128 /*
1129  * contain_exec_param
1130  * Recursively search for PARAM_EXEC Params within a clause.
1131  *
1132  * Returns true if the clause contains any PARAM_EXEC Param with a paramid
1133  * appearing in the given list of Param IDs. Does not descend into
1134  * subqueries!
1135  */
1136 bool
1137 contain_exec_param(Node *clause, List *param_ids)
1138 {
1139  return contain_exec_param_walker(clause, param_ids);
1140 }
1141 
1142 static bool
1144 {
1145  if (node == NULL)
1146  return false;
1147  if (IsA(node, Param))
1148  {
1149  Param *p = (Param *) node;
1150 
1151  if (p->paramkind == PARAM_EXEC &&
1152  list_member_int(param_ids, p->paramid))
1153  return true;
1154  }
1155  return expression_tree_walker(node, contain_exec_param_walker, param_ids);
1156 }
1157 
1158 /*****************************************************************************
1159  * Check clauses for context-dependent nodes
1160  *****************************************************************************/
1161 
1162 /*
1163  * contain_context_dependent_node
1164  * Recursively search for context-dependent nodes within a clause.
1165  *
1166  * CaseTestExpr nodes must appear directly within the corresponding CaseExpr,
1167  * not nested within another one, or they'll see the wrong test value. If one
1168  * appears "bare" in the arguments of a SQL function, then we can't inline the
1169  * SQL function for fear of creating such a situation. The same applies for
1170  * CaseTestExpr used within the elemexpr of an ArrayCoerceExpr.
1171  *
1172  * CoerceToDomainValue would have the same issue if domain CHECK expressions
1173  * could get inlined into larger expressions, but presently that's impossible.
1174  * Still, it might be allowed in future, or other node types with similar
1175  * issues might get invented. So give this function a generic name, and set
1176  * up the recursion state to allow multiple flag bits.
1177  */
1178 static bool
1180 {
1181  int flags = 0;
1182 
1183  return contain_context_dependent_node_walker(clause, &flags);
1184 }
1185 
1186 #define CCDN_CASETESTEXPR_OK 0x0001 /* CaseTestExpr okay here? */
1187 
1188 static bool
1190 {
1191  if (node == NULL)
1192  return false;
1193  if (IsA(node, CaseTestExpr))
1194  return !(*flags & CCDN_CASETESTEXPR_OK);
1195  else if (IsA(node, CaseExpr))
1196  {
1197  CaseExpr *caseexpr = (CaseExpr *) node;
1198 
1199  /*
1200  * If this CASE doesn't have a test expression, then it doesn't create
1201  * a context in which CaseTestExprs should appear, so just fall
1202  * through and treat it as a generic expression node.
1203  */
1204  if (caseexpr->arg)
1205  {
1206  int save_flags = *flags;
1207  bool res;
1208 
1209  /*
1210  * Note: in principle, we could distinguish the various sub-parts
1211  * of a CASE construct and set the flag bit only for some of them,
1212  * since we are only expecting CaseTestExprs to appear in the
1213  * "expr" subtree of the CaseWhen nodes. But it doesn't really
1214  * seem worth any extra code. If there are any bare CaseTestExprs
1215  * elsewhere in the CASE, something's wrong already.
1216  */
1217  *flags |= CCDN_CASETESTEXPR_OK;
1218  res = expression_tree_walker(node,
1220  (void *) flags);
1221  *flags = save_flags;
1222  return res;
1223  }
1224  }
1225  else if (IsA(node, ArrayCoerceExpr))
1226  {
1227  ArrayCoerceExpr *ac = (ArrayCoerceExpr *) node;
1228  int save_flags;
1229  bool res;
1230 
1231  /* Check the array expression */
1232  if (contain_context_dependent_node_walker((Node *) ac->arg, flags))
1233  return true;
1234 
1235  /* Check the elemexpr, which is allowed to contain CaseTestExpr */
1236  save_flags = *flags;
1237  *flags |= CCDN_CASETESTEXPR_OK;
1239  flags);
1240  *flags = save_flags;
1241  return res;
1242  }
1244  (void *) flags);
1245 }
1246 
1247 /*****************************************************************************
1248  * Check clauses for Vars passed to non-leakproof functions
1249  *****************************************************************************/
1250 
1251 /*
1252  * contain_leaked_vars
1253  * Recursively scan a clause to discover whether it contains any Var
1254  * nodes (of the current query level) that are passed as arguments to
1255  * leaky functions.
1256  *
1257  * Returns true if the clause contains any non-leakproof functions that are
1258  * passed Var nodes of the current query level, and which might therefore leak
1259  * data. Such clauses must be applied after any lower-level security barrier
1260  * clauses.
1261  */
1262 bool
1264 {
1265  return contain_leaked_vars_walker(clause, NULL);
1266 }
1267 
1268 static bool
1270 {
1271  return !get_func_leakproof(func_id);
1272 }
1273 
1274 static bool
1276 {
1277  if (node == NULL)
1278  return false;
1279 
1280  switch (nodeTag(node))
1281  {
1282  case T_Var:
1283  case T_Const:
1284  case T_Param:
1285  case T_ArrayExpr:
1286  case T_FieldSelect:
1287  case T_FieldStore:
1288  case T_NamedArgExpr:
1289  case T_BoolExpr:
1290  case T_RelabelType:
1291  case T_CollateExpr:
1292  case T_CaseExpr:
1293  case T_CaseTestExpr:
1294  case T_RowExpr:
1295  case T_SQLValueFunction:
1296  case T_NullTest:
1297  case T_BooleanTest:
1298  case T_NextValueExpr:
1299  case T_List:
1300 
1301  /*
1302  * We know these node types don't contain function calls; but
1303  * something further down in the node tree might.
1304  */
1305  break;
1306 
1307  case T_FuncExpr:
1308  case T_OpExpr:
1309  case T_DistinctExpr:
1310  case T_NullIfExpr:
1311  case T_ScalarArrayOpExpr:
1312  case T_CoerceViaIO:
1313  case T_ArrayCoerceExpr:
1314 
1315  /*
1316  * If node contains a leaky function call, and there's any Var
1317  * underneath it, reject.
1318  */
1320  context) &&
1321  contain_var_clause(node))
1322  return true;
1323  break;
1324 
1325  case T_SubscriptingRef:
1326  {
1327  SubscriptingRef *sbsref = (SubscriptingRef *) node;
1328  const SubscriptRoutines *sbsroutines;
1329 
1330  /* Consult the subscripting support method info */
1331  sbsroutines = getSubscriptingRoutines(sbsref->refcontainertype,
1332  NULL);
1333  if (!sbsroutines ||
1334  !(sbsref->refassgnexpr != NULL ?
1335  sbsroutines->store_leakproof :
1336  sbsroutines->fetch_leakproof))
1337  {
1338  /* Node is leaky, so reject if it contains Vars */
1339  if (contain_var_clause(node))
1340  return true;
1341  }
1342  }
1343  break;
1344 
1345  case T_RowCompareExpr:
1346  {
1347  /*
1348  * It's worth special-casing this because a leaky comparison
1349  * function only compromises one pair of row elements, which
1350  * might not contain Vars while others do.
1351  */
1352  RowCompareExpr *rcexpr = (RowCompareExpr *) node;
1353  ListCell *opid;
1354  ListCell *larg;
1355  ListCell *rarg;
1356 
1357  forthree(opid, rcexpr->opnos,
1358  larg, rcexpr->largs,
1359  rarg, rcexpr->rargs)
1360  {
1361  Oid funcid = get_opcode(lfirst_oid(opid));
1362 
1363  if (!get_func_leakproof(funcid) &&
1364  (contain_var_clause((Node *) lfirst(larg)) ||
1365  contain_var_clause((Node *) lfirst(rarg))))
1366  return true;
1367  }
1368  }
1369  break;
1370 
1371  case T_MinMaxExpr:
1372  {
1373  /*
1374  * MinMaxExpr is leakproof if the comparison function it calls
1375  * is leakproof.
1376  */
1377  MinMaxExpr *minmaxexpr = (MinMaxExpr *) node;
1378  TypeCacheEntry *typentry;
1379  bool leakproof;
1380 
1381  /* Look up the btree comparison function for the datatype */
1382  typentry = lookup_type_cache(minmaxexpr->minmaxtype,
1384  if (OidIsValid(typentry->cmp_proc))
1385  leakproof = get_func_leakproof(typentry->cmp_proc);
1386  else
1387  {
1388  /*
1389  * The executor will throw an error, but here we just
1390  * treat the missing function as leaky.
1391  */
1392  leakproof = false;
1393  }
1394 
1395  if (!leakproof &&
1396  contain_var_clause((Node *) minmaxexpr->args))
1397  return true;
1398  }
1399  break;
1400 
1401  case T_CurrentOfExpr:
1402 
1403  /*
1404  * WHERE CURRENT OF doesn't contain leaky function calls.
1405  * Moreover, it is essential that this is considered non-leaky,
1406  * since the planner must always generate a TID scan when CURRENT
1407  * OF is present -- cf. cost_tidscan.
1408  */
1409  return false;
1410 
1411  default:
1412 
1413  /*
1414  * If we don't recognize the node tag, assume it might be leaky.
1415  * This prevents an unexpected security hole if someone adds a new
1416  * node type that can call a function.
1417  */
1418  return true;
1419  }
1421  context);
1422 }
1423 
1424 /*
1425  * find_nonnullable_rels
1426  * Determine which base rels are forced nonnullable by given clause.
1427  *
1428  * Returns the set of all Relids that are referenced in the clause in such
1429  * a way that the clause cannot possibly return TRUE if any of these Relids
1430  * is an all-NULL row. (It is OK to err on the side of conservatism; hence
1431  * the analysis here is simplistic.)
1432  *
1433  * The semantics here are subtly different from contain_nonstrict_functions:
1434  * that function is concerned with NULL results from arbitrary expressions,
1435  * but here we assume that the input is a Boolean expression, and wish to
1436  * see if NULL inputs will provably cause a FALSE-or-NULL result. We expect
1437  * the expression to have been AND/OR flattened and converted to implicit-AND
1438  * format.
1439  *
1440  * Note: this function is largely duplicative of find_nonnullable_vars().
1441  * The reason not to simplify this function into a thin wrapper around
1442  * find_nonnullable_vars() is that the tested conditions really are different:
1443  * a clause like "t1.v1 IS NOT NULL OR t1.v2 IS NOT NULL" does not prove
1444  * that either v1 or v2 can't be NULL, but it does prove that the t1 row
1445  * as a whole can't be all-NULL. Also, the behavior for PHVs is different.
1446  *
1447  * top_level is true while scanning top-level AND/OR structure; here, showing
1448  * the result is either FALSE or NULL is good enough. top_level is false when
1449  * we have descended below a NOT or a strict function: now we must be able to
1450  * prove that the subexpression goes to NULL.
1451  *
1452  * We don't use expression_tree_walker here because we don't want to descend
1453  * through very many kinds of nodes; only the ones we can be sure are strict.
1454  */
1455 Relids
1457 {
1458  return find_nonnullable_rels_walker(clause, true);
1459 }
1460 
1461 static Relids
1462 find_nonnullable_rels_walker(Node *node, bool top_level)
1463 {
1464  Relids result = NULL;
1465  ListCell *l;
1466 
1467  if (node == NULL)
1468  return NULL;
1469  if (IsA(node, Var))
1470  {
1471  Var *var = (Var *) node;
1472 
1473  if (var->varlevelsup == 0)
1474  result = bms_make_singleton(var->varno);
1475  }
1476  else if (IsA(node, List))
1477  {
1478  /*
1479  * At top level, we are examining an implicit-AND list: if any of the
1480  * arms produces FALSE-or-NULL then the result is FALSE-or-NULL. If
1481  * not at top level, we are examining the arguments of a strict
1482  * function: if any of them produce NULL then the result of the
1483  * function must be NULL. So in both cases, the set of nonnullable
1484  * rels is the union of those found in the arms, and we pass down the
1485  * top_level flag unmodified.
1486  */
1487  foreach(l, (List *) node)
1488  {
1489  result = bms_join(result,
1491  top_level));
1492  }
1493  }
1494  else if (IsA(node, FuncExpr))
1495  {
1496  FuncExpr *expr = (FuncExpr *) node;
1497 
1498  if (func_strict(expr->funcid))
1499  result = find_nonnullable_rels_walker((Node *) expr->args, false);
1500  }
1501  else if (IsA(node, OpExpr))
1502  {
1503  OpExpr *expr = (OpExpr *) node;
1504 
1505  set_opfuncid(expr);
1506  if (func_strict(expr->opfuncid))
1507  result = find_nonnullable_rels_walker((Node *) expr->args, false);
1508  }
1509  else if (IsA(node, ScalarArrayOpExpr))
1510  {
1511  ScalarArrayOpExpr *expr = (ScalarArrayOpExpr *) node;
1512 
1513  if (is_strict_saop(expr, true))
1514  result = find_nonnullable_rels_walker((Node *) expr->args, false);
1515  }
1516  else if (IsA(node, BoolExpr))
1517  {
1518  BoolExpr *expr = (BoolExpr *) node;
1519 
1520  switch (expr->boolop)
1521  {
1522  case AND_EXPR:
1523  /* At top level we can just recurse (to the List case) */
1524  if (top_level)
1525  {
1526  result = find_nonnullable_rels_walker((Node *) expr->args,
1527  top_level);
1528  break;
1529  }
1530 
1531  /*
1532  * Below top level, even if one arm produces NULL, the result
1533  * could be FALSE (hence not NULL). However, if *all* the
1534  * arms produce NULL then the result is NULL, so we can take
1535  * the intersection of the sets of nonnullable rels, just as
1536  * for OR. Fall through to share code.
1537  */
1538  /* FALL THRU */
1539  case OR_EXPR:
1540 
1541  /*
1542  * OR is strict if all of its arms are, so we can take the
1543  * intersection of the sets of nonnullable rels for each arm.
1544  * This works for both values of top_level.
1545  */
1546  foreach(l, expr->args)
1547  {
1548  Relids subresult;
1549 
1550  subresult = find_nonnullable_rels_walker(lfirst(l),
1551  top_level);
1552  if (result == NULL) /* first subresult? */
1553  result = subresult;
1554  else
1555  result = bms_int_members(result, subresult);
1556 
1557  /*
1558  * If the intersection is empty, we can stop looking. This
1559  * also justifies the test for first-subresult above.
1560  */
1561  if (bms_is_empty(result))
1562  break;
1563  }
1564  break;
1565  case NOT_EXPR:
1566  /* NOT will return null if its arg is null */
1567  result = find_nonnullable_rels_walker((Node *) expr->args,
1568  false);
1569  break;
1570  default:
1571  elog(ERROR, "unrecognized boolop: %d", (int) expr->boolop);
1572  break;
1573  }
1574  }
1575  else if (IsA(node, RelabelType))
1576  {
1577  RelabelType *expr = (RelabelType *) node;
1578 
1579  result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
1580  }
1581  else if (IsA(node, CoerceViaIO))
1582  {
1583  /* not clear this is useful, but it can't hurt */
1584  CoerceViaIO *expr = (CoerceViaIO *) node;
1585 
1586  result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
1587  }
1588  else if (IsA(node, ArrayCoerceExpr))
1589  {
1590  /* ArrayCoerceExpr is strict at the array level; ignore elemexpr */
1591  ArrayCoerceExpr *expr = (ArrayCoerceExpr *) node;
1592 
1593  result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
1594  }
1595  else if (IsA(node, ConvertRowtypeExpr))
1596  {
1597  /* not clear this is useful, but it can't hurt */
1598  ConvertRowtypeExpr *expr = (ConvertRowtypeExpr *) node;
1599 
1600  result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
1601  }
1602  else if (IsA(node, CollateExpr))
1603  {
1604  CollateExpr *expr = (CollateExpr *) node;
1605 
1606  result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
1607  }
1608  else if (IsA(node, NullTest))
1609  {
1610  /* IS NOT NULL can be considered strict, but only at top level */
1611  NullTest *expr = (NullTest *) node;
1612 
1613  if (top_level && expr->nulltesttype == IS_NOT_NULL && !expr->argisrow)
1614  result = find_nonnullable_rels_walker((Node *) expr->arg, false);
1615  }
1616  else if (IsA(node, BooleanTest))
1617  {
1618  /* Boolean tests that reject NULL are strict at top level */
1619  BooleanTest *expr = (BooleanTest *) node;
1620 
1621  if (top_level &&
1622  (expr->booltesttype == IS_TRUE ||
1623  expr->booltesttype == IS_FALSE ||
1624  expr->booltesttype == IS_NOT_UNKNOWN))
1625  result = find_nonnullable_rels_walker((Node *) expr->arg, false);
1626  }
1627  else if (IsA(node, SubPlan))
1628  {
1629  SubPlan *splan = (SubPlan *) node;
1630 
1631  /*
1632  * For some types of SubPlan, we can infer strictness from Vars in the
1633  * testexpr (the LHS of the original SubLink).
1634  *
1635  * For ANY_SUBLINK, if the subquery produces zero rows, the result is
1636  * always FALSE. If the subquery produces more than one row, the
1637  * per-row results of the testexpr are combined using OR semantics.
1638  * Hence ANY_SUBLINK can be strict only at top level, but there it's
1639  * as strict as the testexpr is.
1640  *
1641  * For ROWCOMPARE_SUBLINK, if the subquery produces zero rows, the
1642  * result is always NULL. Otherwise, the result is as strict as the
1643  * testexpr is. So we can check regardless of top_level.
1644  *
1645  * We can't prove anything for other sublink types (in particular,
1646  * note that ALL_SUBLINK will return TRUE if the subquery is empty).
1647  */
1648  if ((top_level && splan->subLinkType == ANY_SUBLINK) ||
1649  splan->subLinkType == ROWCOMPARE_SUBLINK)
1650  result = find_nonnullable_rels_walker(splan->testexpr, top_level);
1651  }
1652  else if (IsA(node, PlaceHolderVar))
1653  {
1654  PlaceHolderVar *phv = (PlaceHolderVar *) node;
1655 
1656  /*
1657  * If the contained expression forces any rels non-nullable, so does
1658  * the PHV.
1659  */
1660  result = find_nonnullable_rels_walker((Node *) phv->phexpr, top_level);
1661 
1662  /*
1663  * If the PHV's syntactic scope is exactly one rel, it will be forced
1664  * to be evaluated at that rel, and so it will behave like a Var of
1665  * that rel: if the rel's entire output goes to null, so will the PHV.
1666  * (If the syntactic scope is a join, we know that the PHV will go to
1667  * null if the whole join does; but that is AND semantics while we
1668  * need OR semantics for find_nonnullable_rels' result, so we can't do
1669  * anything with the knowledge.)
1670  */
1671  if (phv->phlevelsup == 0 &&
1672  bms_membership(phv->phrels) == BMS_SINGLETON)
1673  result = bms_add_members(result, phv->phrels);
1674  }
1675  return result;
1676 }
1677 
1678 /*
1679  * find_nonnullable_vars
1680  * Determine which Vars are forced nonnullable by given clause.
1681  *
1682  * Returns the set of all level-zero Vars that are referenced in the clause in
1683  * such a way that the clause cannot possibly return TRUE if any of these Vars
1684  * is NULL. (It is OK to err on the side of conservatism; hence the analysis
1685  * here is simplistic.)
1686  *
1687  * The semantics here are subtly different from contain_nonstrict_functions:
1688  * that function is concerned with NULL results from arbitrary expressions,
1689  * but here we assume that the input is a Boolean expression, and wish to
1690  * see if NULL inputs will provably cause a FALSE-or-NULL result. We expect
1691  * the expression to have been AND/OR flattened and converted to implicit-AND
1692  * format.
1693  *
1694  * Attnos of the identified Vars are returned in a multibitmapset (a List of
1695  * Bitmapsets). List indexes correspond to relids (varnos), while the per-rel
1696  * Bitmapsets hold varattnos offset by FirstLowInvalidHeapAttributeNumber.
1697  *
1698  * top_level is true while scanning top-level AND/OR structure; here, showing
1699  * the result is either FALSE or NULL is good enough. top_level is false when
1700  * we have descended below a NOT or a strict function: now we must be able to
1701  * prove that the subexpression goes to NULL.
1702  *
1703  * We don't use expression_tree_walker here because we don't want to descend
1704  * through very many kinds of nodes; only the ones we can be sure are strict.
1705  */
1706 List *
1708 {
1709  return find_nonnullable_vars_walker(clause, true);
1710 }
1711 
1712 static List *
1713 find_nonnullable_vars_walker(Node *node, bool top_level)
1714 {
1715  List *result = NIL;
1716  ListCell *l;
1717 
1718  if (node == NULL)
1719  return NIL;
1720  if (IsA(node, Var))
1721  {
1722  Var *var = (Var *) node;
1723 
1724  if (var->varlevelsup == 0)
1725  result = mbms_add_member(result,
1726  var->varno,
1728  }
1729  else if (IsA(node, List))
1730  {
1731  /*
1732  * At top level, we are examining an implicit-AND list: if any of the
1733  * arms produces FALSE-or-NULL then the result is FALSE-or-NULL. If
1734  * not at top level, we are examining the arguments of a strict
1735  * function: if any of them produce NULL then the result of the
1736  * function must be NULL. So in both cases, the set of nonnullable
1737  * vars is the union of those found in the arms, and we pass down the
1738  * top_level flag unmodified.
1739  */
1740  foreach(l, (List *) node)
1741  {
1742  result = mbms_add_members(result,
1744  top_level));
1745  }
1746  }
1747  else if (IsA(node, FuncExpr))
1748  {
1749  FuncExpr *expr = (FuncExpr *) node;
1750 
1751  if (func_strict(expr->funcid))
1752  result = find_nonnullable_vars_walker((Node *) expr->args, false);
1753  }
1754  else if (IsA(node, OpExpr))
1755  {
1756  OpExpr *expr = (OpExpr *) node;
1757 
1758  set_opfuncid(expr);
1759  if (func_strict(expr->opfuncid))
1760  result = find_nonnullable_vars_walker((Node *) expr->args, false);
1761  }
1762  else if (IsA(node, ScalarArrayOpExpr))
1763  {
1764  ScalarArrayOpExpr *expr = (ScalarArrayOpExpr *) node;
1765 
1766  if (is_strict_saop(expr, true))
1767  result = find_nonnullable_vars_walker((Node *) expr->args, false);
1768  }
1769  else if (IsA(node, BoolExpr))
1770  {
1771  BoolExpr *expr = (BoolExpr *) node;
1772 
1773  switch (expr->boolop)
1774  {
1775  case AND_EXPR:
1776 
1777  /*
1778  * At top level we can just recurse (to the List case), since
1779  * the result should be the union of what we can prove in each
1780  * arm.
1781  */
1782  if (top_level)
1783  {
1784  result = find_nonnullable_vars_walker((Node *) expr->args,
1785  top_level);
1786  break;
1787  }
1788 
1789  /*
1790  * Below top level, even if one arm produces NULL, the result
1791  * could be FALSE (hence not NULL). However, if *all* the
1792  * arms produce NULL then the result is NULL, so we can take
1793  * the intersection of the sets of nonnullable vars, just as
1794  * for OR. Fall through to share code.
1795  */
1796  /* FALL THRU */
1797  case OR_EXPR:
1798 
1799  /*
1800  * OR is strict if all of its arms are, so we can take the
1801  * intersection of the sets of nonnullable vars for each arm.
1802  * This works for both values of top_level.
1803  */
1804  foreach(l, expr->args)
1805  {
1806  List *subresult;
1807 
1808  subresult = find_nonnullable_vars_walker(lfirst(l),
1809  top_level);
1810  if (result == NIL) /* first subresult? */
1811  result = subresult;
1812  else
1813  result = mbms_int_members(result, subresult);
1814 
1815  /*
1816  * If the intersection is empty, we can stop looking. This
1817  * also justifies the test for first-subresult above.
1818  */
1819  if (result == NIL)
1820  break;
1821  }
1822  break;
1823  case NOT_EXPR:
1824  /* NOT will return null if its arg is null */
1825  result = find_nonnullable_vars_walker((Node *) expr->args,
1826  false);
1827  break;
1828  default:
1829  elog(ERROR, "unrecognized boolop: %d", (int) expr->boolop);
1830  break;
1831  }
1832  }
1833  else if (IsA(node, RelabelType))
1834  {
1835  RelabelType *expr = (RelabelType *) node;
1836 
1837  result = find_nonnullable_vars_walker((Node *) expr->arg, top_level);
1838  }
1839  else if (IsA(node, CoerceViaIO))
1840  {
1841  /* not clear this is useful, but it can't hurt */
1842  CoerceViaIO *expr = (CoerceViaIO *) node;
1843 
1844  result = find_nonnullable_vars_walker((Node *) expr->arg, false);
1845  }
1846  else if (IsA(node, ArrayCoerceExpr))
1847  {
1848  /* ArrayCoerceExpr is strict at the array level; ignore elemexpr */
1849  ArrayCoerceExpr *expr = (ArrayCoerceExpr *) node;
1850 
1851  result = find_nonnullable_vars_walker((Node *) expr->arg, top_level);
1852  }
1853  else if (IsA(node, ConvertRowtypeExpr))
1854  {
1855  /* not clear this is useful, but it can't hurt */
1856  ConvertRowtypeExpr *expr = (ConvertRowtypeExpr *) node;
1857 
1858  result = find_nonnullable_vars_walker((Node *) expr->arg, top_level);
1859  }
1860  else if (IsA(node, CollateExpr))
1861  {
1862  CollateExpr *expr = (CollateExpr *) node;
1863 
1864  result = find_nonnullable_vars_walker((Node *) expr->arg, top_level);
1865  }
1866  else if (IsA(node, NullTest))
1867  {
1868  /* IS NOT NULL can be considered strict, but only at top level */
1869  NullTest *expr = (NullTest *) node;
1870 
1871  if (top_level && expr->nulltesttype == IS_NOT_NULL && !expr->argisrow)
1872  result = find_nonnullable_vars_walker((Node *) expr->arg, false);
1873  }
1874  else if (IsA(node, BooleanTest))
1875  {
1876  /* Boolean tests that reject NULL are strict at top level */
1877  BooleanTest *expr = (BooleanTest *) node;
1878 
1879  if (top_level &&
1880  (expr->booltesttype == IS_TRUE ||
1881  expr->booltesttype == IS_FALSE ||
1882  expr->booltesttype == IS_NOT_UNKNOWN))
1883  result = find_nonnullable_vars_walker((Node *) expr->arg, false);
1884  }
1885  else if (IsA(node, SubPlan))
1886  {
1887  SubPlan *splan = (SubPlan *) node;
1888 
1889  /* See analysis in find_nonnullable_rels_walker */
1890  if ((top_level && splan->subLinkType == ANY_SUBLINK) ||
1891  splan->subLinkType == ROWCOMPARE_SUBLINK)
1892  result = find_nonnullable_vars_walker(splan->testexpr, top_level);
1893  }
1894  else if (IsA(node, PlaceHolderVar))
1895  {
1896  PlaceHolderVar *phv = (PlaceHolderVar *) node;
1897 
1898  result = find_nonnullable_vars_walker((Node *) phv->phexpr, top_level);
1899  }
1900  return result;
1901 }
1902 
1903 /*
1904  * find_forced_null_vars
1905  * Determine which Vars must be NULL for the given clause to return TRUE.
1906  *
1907  * This is the complement of find_nonnullable_vars: find the level-zero Vars
1908  * that must be NULL for the clause to return TRUE. (It is OK to err on the
1909  * side of conservatism; hence the analysis here is simplistic. In fact,
1910  * we only detect simple "var IS NULL" tests at the top level.)
1911  *
1912  * As with find_nonnullable_vars, we return the varattnos of the identified
1913  * Vars in a multibitmapset.
1914  */
1915 List *
1917 {
1918  List *result = NIL;
1919  Var *var;
1920  ListCell *l;
1921 
1922  if (node == NULL)
1923  return NIL;
1924  /* Check single-clause cases using subroutine */
1925  var = find_forced_null_var(node);
1926  if (var)
1927  {
1928  result = mbms_add_member(result,
1929  var->varno,
1931  }
1932  /* Otherwise, handle AND-conditions */
1933  else if (IsA(node, List))
1934  {
1935  /*
1936  * At top level, we are examining an implicit-AND list: if any of the
1937  * arms produces FALSE-or-NULL then the result is FALSE-or-NULL.
1938  */
1939  foreach(l, (List *) node)
1940  {
1941  result = mbms_add_members(result,
1943  }
1944  }
1945  else if (IsA(node, BoolExpr))
1946  {
1947  BoolExpr *expr = (BoolExpr *) node;
1948 
1949  /*
1950  * We don't bother considering the OR case, because it's fairly
1951  * unlikely anyone would write "v1 IS NULL OR v1 IS NULL". Likewise,
1952  * the NOT case isn't worth expending code on.
1953  */
1954  if (expr->boolop == AND_EXPR)
1955  {
1956  /* At top level we can just recurse (to the List case) */
1957  result = find_forced_null_vars((Node *) expr->args);
1958  }
1959  }
1960  return result;
1961 }
1962 
1963 /*
1964  * find_forced_null_var
1965  * Return the Var forced null by the given clause, or NULL if it's
1966  * not an IS NULL-type clause. For success, the clause must enforce
1967  * *only* nullness of the particular Var, not any other conditions.
1968  *
1969  * This is just the single-clause case of find_forced_null_vars(), without
1970  * any allowance for AND conditions. It's used by initsplan.c on individual
1971  * qual clauses. The reason for not just applying find_forced_null_vars()
1972  * is that if an AND of an IS NULL clause with something else were to somehow
1973  * survive AND/OR flattening, initsplan.c might get fooled into discarding
1974  * the whole clause when only the IS NULL part of it had been proved redundant.
1975  */
1976 Var *
1978 {
1979  if (node == NULL)
1980  return NULL;
1981  if (IsA(node, NullTest))
1982  {
1983  /* check for var IS NULL */
1984  NullTest *expr = (NullTest *) node;
1985 
1986  if (expr->nulltesttype == IS_NULL && !expr->argisrow)
1987  {
1988  Var *var = (Var *) expr->arg;
1989 
1990  if (var && IsA(var, Var) &&
1991  var->varlevelsup == 0)
1992  return var;
1993  }
1994  }
1995  else if (IsA(node, BooleanTest))
1996  {
1997  /* var IS UNKNOWN is equivalent to var IS NULL */
1998  BooleanTest *expr = (BooleanTest *) node;
1999 
2000  if (expr->booltesttype == IS_UNKNOWN)
2001  {
2002  Var *var = (Var *) expr->arg;
2003 
2004  if (var && IsA(var, Var) &&
2005  var->varlevelsup == 0)
2006  return var;
2007  }
2008  }
2009  return NULL;
2010 }
2011 
2012 /*
2013  * Can we treat a ScalarArrayOpExpr as strict?
2014  *
2015  * If "falseOK" is true, then a "false" result can be considered strict,
2016  * else we need to guarantee an actual NULL result for NULL input.
2017  *
2018  * "foo op ALL array" is strict if the op is strict *and* we can prove
2019  * that the array input isn't an empty array. We can check that
2020  * for the cases of an array constant and an ARRAY[] construct.
2021  *
2022  * "foo op ANY array" is strict in the falseOK sense if the op is strict.
2023  * If not falseOK, the test is the same as for "foo op ALL array".
2024  */
2025 static bool
2026 is_strict_saop(ScalarArrayOpExpr *expr, bool falseOK)
2027 {
2028  Node *rightop;
2029 
2030  /* The contained operator must be strict. */
2031  set_sa_opfuncid(expr);
2032  if (!func_strict(expr->opfuncid))
2033  return false;
2034  /* If ANY and falseOK, that's all we need to check. */
2035  if (expr->useOr && falseOK)
2036  return true;
2037  /* Else, we have to see if the array is provably non-empty. */
2038  Assert(list_length(expr->args) == 2);
2039  rightop = (Node *) lsecond(expr->args);
2040  if (rightop && IsA(rightop, Const))
2041  {
2042  Datum arraydatum = ((Const *) rightop)->constvalue;
2043  bool arrayisnull = ((Const *) rightop)->constisnull;
2044  ArrayType *arrayval;
2045  int nitems;
2046 
2047  if (arrayisnull)
2048  return false;
2049  arrayval = DatumGetArrayTypeP(arraydatum);
2050  nitems = ArrayGetNItems(ARR_NDIM(arrayval), ARR_DIMS(arrayval));
2051  if (nitems > 0)
2052  return true;
2053  }
2054  else if (rightop && IsA(rightop, ArrayExpr))
2055  {
2056  ArrayExpr *arrayexpr = (ArrayExpr *) rightop;
2057 
2058  if (arrayexpr->elements != NIL && !arrayexpr->multidims)
2059  return true;
2060  }
2061  return false;
2062 }
2063 
2064 
2065 /*****************************************************************************
2066  * Check for "pseudo-constant" clauses
2067  *****************************************************************************/
2068 
2069 /*
2070  * is_pseudo_constant_clause
2071  * Detect whether an expression is "pseudo constant", ie, it contains no
2072  * variables of the current query level and no uses of volatile functions.
2073  * Such an expr is not necessarily a true constant: it can still contain
2074  * Params and outer-level Vars, not to mention functions whose results
2075  * may vary from one statement to the next. However, the expr's value
2076  * will be constant over any one scan of the current query, so it can be
2077  * used as, eg, an indexscan key. (Actually, the condition for indexscan
2078  * keys is weaker than this; see is_pseudo_constant_for_index().)
2079  *
2080  * CAUTION: this function omits to test for one very important class of
2081  * not-constant expressions, namely aggregates (Aggrefs). In current usage
2082  * this is only applied to WHERE clauses and so a check for Aggrefs would be
2083  * a waste of cycles; but be sure to also check contain_agg_clause() if you
2084  * want to know about pseudo-constness in other contexts. The same goes
2085  * for window functions (WindowFuncs).
2086  */
2087 bool
2089 {
2090  /*
2091  * We could implement this check in one recursive scan. But since the
2092  * check for volatile functions is both moderately expensive and unlikely
2093  * to fail, it seems better to look for Vars first and only check for
2094  * volatile functions if we find no Vars.
2095  */
2096  if (!contain_var_clause(clause) &&
2097  !contain_volatile_functions(clause))
2098  return true;
2099  return false;
2100 }
2101 
2102 /*
2103  * is_pseudo_constant_clause_relids
2104  * Same as above, except caller already has available the var membership
2105  * of the expression; this lets us avoid the contain_var_clause() scan.
2106  */
2107 bool
2109 {
2110  if (bms_is_empty(relids) &&
2111  !contain_volatile_functions(clause))
2112  return true;
2113  return false;
2114 }
2115 
2116 
2117 /*****************************************************************************
2118  * *
2119  * General clause-manipulating routines *
2120  * *
2121  *****************************************************************************/
2122 
2123 /*
2124  * NumRelids
2125  * (formerly clause_relids)
2126  *
2127  * Returns the number of different base relations referenced in 'clause'.
2128  */
2129 int
2131 {
2132  int result;
2133  Relids varnos = pull_varnos(root, clause);
2134 
2135  varnos = bms_del_members(varnos, root->outer_join_rels);
2136  result = bms_num_members(varnos);
2137  bms_free(varnos);
2138  return result;
2139 }
2140 
2141 /*
2142  * CommuteOpExpr: commute a binary operator clause
2143  *
2144  * XXX the clause is destructively modified!
2145  */
2146 void
2148 {
2149  Oid opoid;
2150  Node *temp;
2151 
2152  /* Sanity checks: caller is at fault if these fail */
2153  if (!is_opclause(clause) ||
2154  list_length(clause->args) != 2)
2155  elog(ERROR, "cannot commute non-binary-operator clause");
2156 
2157  opoid = get_commutator(clause->opno);
2158 
2159  if (!OidIsValid(opoid))
2160  elog(ERROR, "could not find commutator for operator %u",
2161  clause->opno);
2162 
2163  /*
2164  * modify the clause in-place!
2165  */
2166  clause->opno = opoid;
2167  clause->opfuncid = InvalidOid;
2168  /* opresulttype, opretset, opcollid, inputcollid need not change */
2169 
2170  temp = linitial(clause->args);
2171  linitial(clause->args) = lsecond(clause->args);
2172  lsecond(clause->args) = temp;
2173 }
2174 
2175 /*
2176  * Helper for eval_const_expressions: check that datatype of an attribute
2177  * is still what it was when the expression was parsed. This is needed to
2178  * guard against improper simplification after ALTER COLUMN TYPE. (XXX we
2179  * may well need to make similar checks elsewhere?)
2180  *
2181  * rowtypeid may come from a whole-row Var, and therefore it can be a domain
2182  * over composite, but for this purpose we only care about checking the type
2183  * of a contained field.
2184  */
2185 static bool
2186 rowtype_field_matches(Oid rowtypeid, int fieldnum,
2187  Oid expectedtype, int32 expectedtypmod,
2188  Oid expectedcollation)
2189 {
2190  TupleDesc tupdesc;
2191  Form_pg_attribute attr;
2192 
2193  /* No issue for RECORD, since there is no way to ALTER such a type */
2194  if (rowtypeid == RECORDOID)
2195  return true;
2196  tupdesc = lookup_rowtype_tupdesc_domain(rowtypeid, -1, false);
2197  if (fieldnum <= 0 || fieldnum > tupdesc->natts)
2198  {
2199  ReleaseTupleDesc(tupdesc);
2200  return false;
2201  }
2202  attr = TupleDescAttr(tupdesc, fieldnum - 1);
2203  if (attr->attisdropped ||
2204  attr->atttypid != expectedtype ||
2205  attr->atttypmod != expectedtypmod ||
2206  attr->attcollation != expectedcollation)
2207  {
2208  ReleaseTupleDesc(tupdesc);
2209  return false;
2210  }
2211  ReleaseTupleDesc(tupdesc);
2212  return true;
2213 }
2214 
2215 
2216 /*--------------------
2217  * eval_const_expressions
2218  *
2219  * Reduce any recognizably constant subexpressions of the given
2220  * expression tree, for example "2 + 2" => "4". More interestingly,
2221  * we can reduce certain boolean expressions even when they contain
2222  * non-constant subexpressions: "x OR true" => "true" no matter what
2223  * the subexpression x is. (XXX We assume that no such subexpression
2224  * will have important side-effects, which is not necessarily a good
2225  * assumption in the presence of user-defined functions; do we need a
2226  * pg_proc flag that prevents discarding the execution of a function?)
2227  *
2228  * We do understand that certain functions may deliver non-constant
2229  * results even with constant inputs, "nextval()" being the classic
2230  * example. Functions that are not marked "immutable" in pg_proc
2231  * will not be pre-evaluated here, although we will reduce their
2232  * arguments as far as possible.
2233  *
2234  * Whenever a function is eliminated from the expression by means of
2235  * constant-expression evaluation or inlining, we add the function to
2236  * root->glob->invalItems. This ensures the plan is known to depend on
2237  * such functions, even though they aren't referenced anymore.
2238  *
2239  * We assume that the tree has already been type-checked and contains
2240  * only operators and functions that are reasonable to try to execute.
2241  *
2242  * NOTE: "root" can be passed as NULL if the caller never wants to do any
2243  * Param substitutions nor receive info about inlined functions.
2244  *
2245  * NOTE: the planner assumes that this will always flatten nested AND and
2246  * OR clauses into N-argument form. See comments in prepqual.c.
2247  *
2248  * NOTE: another critical effect is that any function calls that require
2249  * default arguments will be expanded, and named-argument calls will be
2250  * converted to positional notation. The executor won't handle either.
2251  *--------------------
2252  */
2253 Node *
2255 {
2257 
2258  if (root)
2259  context.boundParams = root->glob->boundParams; /* bound Params */
2260  else
2261  context.boundParams = NULL;
2262  context.root = root; /* for inlined-function dependencies */
2263  context.active_fns = NIL; /* nothing being recursively simplified */
2264  context.case_val = NULL; /* no CASE being examined */
2265  context.estimate = false; /* safe transformations only */
2266  return eval_const_expressions_mutator(node, &context);
2267 }
2268 
2269 #define MIN_ARRAY_SIZE_FOR_HASHED_SAOP 9
2270 /*--------------------
2271  * convert_saop_to_hashed_saop
2272  *
2273  * Recursively search 'node' for ScalarArrayOpExprs and fill in the hash
2274  * function for any ScalarArrayOpExpr that looks like it would be useful to
2275  * evaluate using a hash table rather than a linear search.
2276  *
2277  * We'll use a hash table if all of the following conditions are met:
2278  * 1. The 2nd argument of the array contain only Consts.
2279  * 2. useOr is true or there is a valid negator operator for the
2280  * ScalarArrayOpExpr's opno.
2281  * 3. There's valid hash function for both left and righthand operands and
2282  * these hash functions are the same.
2283  * 4. If the array contains enough elements for us to consider it to be
2284  * worthwhile using a hash table rather than a linear search.
2285  */
2286 void
2288 {
2289  (void) convert_saop_to_hashed_saop_walker(node, NULL);
2290 }
2291 
2292 static bool
2294 {
2295  if (node == NULL)
2296  return false;
2297 
2298  if (IsA(node, ScalarArrayOpExpr))
2299  {
2300  ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) node;
2301  Expr *arrayarg = (Expr *) lsecond(saop->args);
2302  Oid lefthashfunc;
2303  Oid righthashfunc;
2304 
2305  if (arrayarg && IsA(arrayarg, Const) &&
2306  !((Const *) arrayarg)->constisnull)
2307  {
2308  if (saop->useOr)
2309  {
2310  if (get_op_hash_functions(saop->opno, &lefthashfunc, &righthashfunc) &&
2311  lefthashfunc == righthashfunc)
2312  {
2313  Datum arrdatum = ((Const *) arrayarg)->constvalue;
2314  ArrayType *arr = (ArrayType *) DatumGetPointer(arrdatum);
2315  int nitems;
2316 
2317  /*
2318  * Only fill in the hash functions if the array looks
2319  * large enough for it to be worth hashing instead of
2320  * doing a linear search.
2321  */
2322  nitems = ArrayGetNItems(ARR_NDIM(arr), ARR_DIMS(arr));
2323 
2325  {
2326  /* Looks good. Fill in the hash functions */
2327  saop->hashfuncid = lefthashfunc;
2328  }
2329  return true;
2330  }
2331  }
2332  else /* !saop->useOr */
2333  {
2334  Oid negator = get_negator(saop->opno);
2335 
2336  /*
2337  * Check if this is a NOT IN using an operator whose negator
2338  * is hashable. If so we can still build a hash table and
2339  * just ensure the lookup items are not in the hash table.
2340  */
2341  if (OidIsValid(negator) &&
2342  get_op_hash_functions(negator, &lefthashfunc, &righthashfunc) &&
2343  lefthashfunc == righthashfunc)
2344  {
2345  Datum arrdatum = ((Const *) arrayarg)->constvalue;
2346  ArrayType *arr = (ArrayType *) DatumGetPointer(arrdatum);
2347  int nitems;
2348 
2349  /*
2350  * Only fill in the hash functions if the array looks
2351  * large enough for it to be worth hashing instead of
2352  * doing a linear search.
2353  */
2354  nitems = ArrayGetNItems(ARR_NDIM(arr), ARR_DIMS(arr));
2355 
2357  {
2358  /* Looks good. Fill in the hash functions */
2359  saop->hashfuncid = lefthashfunc;
2360 
2361  /*
2362  * Also set the negfuncid. The executor will need
2363  * that to perform hashtable lookups.
2364  */
2365  saop->negfuncid = get_opcode(negator);
2366  }
2367  return true;
2368  }
2369  }
2370  }
2371  }
2372 
2374 }
2375 
2376 
2377 /*--------------------
2378  * estimate_expression_value
2379  *
2380  * This function attempts to estimate the value of an expression for
2381  * planning purposes. It is in essence a more aggressive version of
2382  * eval_const_expressions(): we will perform constant reductions that are
2383  * not necessarily 100% safe, but are reasonable for estimation purposes.
2384  *
2385  * Currently the extra steps that are taken in this mode are:
2386  * 1. Substitute values for Params, where a bound Param value has been made
2387  * available by the caller of planner(), even if the Param isn't marked
2388  * constant. This effectively means that we plan using the first supplied
2389  * value of the Param.
2390  * 2. Fold stable, as well as immutable, functions to constants.
2391  * 3. Reduce PlaceHolderVar nodes to their contained expressions.
2392  *--------------------
2393  */
2394 Node *
2396 {
2398 
2399  context.boundParams = root->glob->boundParams; /* bound Params */
2400  /* we do not need to mark the plan as depending on inlined functions */
2401  context.root = NULL;
2402  context.active_fns = NIL; /* nothing being recursively simplified */
2403  context.case_val = NULL; /* no CASE being examined */
2404  context.estimate = true; /* unsafe transformations OK */
2405  return eval_const_expressions_mutator(node, &context);
2406 }
2407 
2408 /*
2409  * The generic case in eval_const_expressions_mutator is to recurse using
2410  * expression_tree_mutator, which will copy the given node unchanged but
2411  * const-simplify its arguments (if any) as far as possible. If the node
2412  * itself does immutable processing, and each of its arguments were reduced
2413  * to a Const, we can then reduce it to a Const using evaluate_expr. (Some
2414  * node types need more complicated logic; for example, a CASE expression
2415  * might be reducible to a constant even if not all its subtrees are.)
2416  */
2417 #define ece_generic_processing(node) \
2418  expression_tree_mutator((Node *) (node), eval_const_expressions_mutator, \
2419  (void *) context)
2420 
2421 /*
2422  * Check whether all arguments of the given node were reduced to Consts.
2423  * By going directly to expression_tree_walker, contain_non_const_walker
2424  * is not applied to the node itself, only to its children.
2425  */
2426 #define ece_all_arguments_const(node) \
2427  (!expression_tree_walker((Node *) (node), contain_non_const_walker, NULL))
2428 
2429 /* Generic macro for applying evaluate_expr */
2430 #define ece_evaluate_expr(node) \
2431  ((Node *) evaluate_expr((Expr *) (node), \
2432  exprType((Node *) (node)), \
2433  exprTypmod((Node *) (node)), \
2434  exprCollation((Node *) (node))))
2435 
2436 /*
2437  * Recursive guts of eval_const_expressions/estimate_expression_value
2438  */
2439 static Node *
2442 {
2443 
2444  /* since this function recurses, it could be driven to stack overflow */
2446 
2447  if (node == NULL)
2448  return NULL;
2449  switch (nodeTag(node))
2450  {
2451  case T_Param:
2452  {
2453  Param *param = (Param *) node;
2454  ParamListInfo paramLI = context->boundParams;
2455 
2456  /* Look to see if we've been given a value for this Param */
2457  if (param->paramkind == PARAM_EXTERN &&
2458  paramLI != NULL &&
2459  param->paramid > 0 &&
2460  param->paramid <= paramLI->numParams)
2461  {
2462  ParamExternData *prm;
2463  ParamExternData prmdata;
2464 
2465  /*
2466  * Give hook a chance in case parameter is dynamic. Tell
2467  * it that this fetch is speculative, so it should avoid
2468  * erroring out if parameter is unavailable.
2469  */
2470  if (paramLI->paramFetch != NULL)
2471  prm = paramLI->paramFetch(paramLI, param->paramid,
2472  true, &prmdata);
2473  else
2474  prm = &paramLI->params[param->paramid - 1];
2475 
2476  /*
2477  * We don't just check OidIsValid, but insist that the
2478  * fetched type match the Param, just in case the hook did
2479  * something unexpected. No need to throw an error here
2480  * though; leave that for runtime.
2481  */
2482  if (OidIsValid(prm->ptype) &&
2483  prm->ptype == param->paramtype)
2484  {
2485  /* OK to substitute parameter value? */
2486  if (context->estimate ||
2487  (prm->pflags & PARAM_FLAG_CONST))
2488  {
2489  /*
2490  * Return a Const representing the param value.
2491  * Must copy pass-by-ref datatypes, since the
2492  * Param might be in a memory context
2493  * shorter-lived than our output plan should be.
2494  */
2495  int16 typLen;
2496  bool typByVal;
2497  Datum pval;
2498  Const *con;
2499 
2500  get_typlenbyval(param->paramtype,
2501  &typLen, &typByVal);
2502  if (prm->isnull || typByVal)
2503  pval = prm->value;
2504  else
2505  pval = datumCopy(prm->value, typByVal, typLen);
2506  con = makeConst(param->paramtype,
2507  param->paramtypmod,
2508  param->paramcollid,
2509  (int) typLen,
2510  pval,
2511  prm->isnull,
2512  typByVal);
2513  con->location = param->location;
2514  return (Node *) con;
2515  }
2516  }
2517  }
2518 
2519  /*
2520  * Not replaceable, so just copy the Param (no need to
2521  * recurse)
2522  */
2523  return (Node *) copyObject(param);
2524  }
2525  case T_WindowFunc:
2526  {
2527  WindowFunc *expr = (WindowFunc *) node;
2528  Oid funcid = expr->winfnoid;
2529  List *args;
2530  Expr *aggfilter;
2531  HeapTuple func_tuple;
2532  WindowFunc *newexpr;
2533 
2534  /*
2535  * We can't really simplify a WindowFunc node, but we mustn't
2536  * just fall through to the default processing, because we
2537  * have to apply expand_function_arguments to its argument
2538  * list. That takes care of inserting default arguments and
2539  * expanding named-argument notation.
2540  */
2541  func_tuple = SearchSysCache1(PROCOID, ObjectIdGetDatum(funcid));
2542  if (!HeapTupleIsValid(func_tuple))
2543  elog(ERROR, "cache lookup failed for function %u", funcid);
2544 
2546  false, expr->wintype,
2547  func_tuple);
2548 
2549  ReleaseSysCache(func_tuple);
2550 
2551  /* Now, recursively simplify the args (which are a List) */
2552  args = (List *)
2555  (void *) context);
2556  /* ... and the filter expression, which isn't */
2557  aggfilter = (Expr *)
2559  context);
2560 
2561  /* And build the replacement WindowFunc node */
2562  newexpr = makeNode(WindowFunc);
2563  newexpr->winfnoid = expr->winfnoid;
2564  newexpr->wintype = expr->wintype;
2565  newexpr->wincollid = expr->wincollid;
2566  newexpr->inputcollid = expr->inputcollid;
2567  newexpr->args = args;
2568  newexpr->aggfilter = aggfilter;
2569  newexpr->runCondition = expr->runCondition;
2570  newexpr->winref = expr->winref;
2571  newexpr->winstar = expr->winstar;
2572  newexpr->winagg = expr->winagg;
2573  newexpr->location = expr->location;
2574 
2575  return (Node *) newexpr;
2576  }
2577  case T_FuncExpr:
2578  {
2579  FuncExpr *expr = (FuncExpr *) node;
2580  List *args = expr->args;
2581  Expr *simple;
2582  FuncExpr *newexpr;
2583 
2584  /*
2585  * Code for op/func reduction is pretty bulky, so split it out
2586  * as a separate function. Note: exprTypmod normally returns
2587  * -1 for a FuncExpr, but not when the node is recognizably a
2588  * length coercion; we want to preserve the typmod in the
2589  * eventual Const if so.
2590  */
2591  simple = simplify_function(expr->funcid,
2592  expr->funcresulttype,
2593  exprTypmod(node),
2594  expr->funccollid,
2595  expr->inputcollid,
2596  &args,
2597  expr->funcvariadic,
2598  true,
2599  true,
2600  context);
2601  if (simple) /* successfully simplified it */
2602  return (Node *) simple;
2603 
2604  /*
2605  * The expression cannot be simplified any further, so build
2606  * and return a replacement FuncExpr node using the
2607  * possibly-simplified arguments. Note that we have also
2608  * converted the argument list to positional notation.
2609  */
2610  newexpr = makeNode(FuncExpr);
2611  newexpr->funcid = expr->funcid;
2612  newexpr->funcresulttype = expr->funcresulttype;
2613  newexpr->funcretset = expr->funcretset;
2614  newexpr->funcvariadic = expr->funcvariadic;
2615  newexpr->funcformat = expr->funcformat;
2616  newexpr->funccollid = expr->funccollid;
2617  newexpr->inputcollid = expr->inputcollid;
2618  newexpr->args = args;
2619  newexpr->location = expr->location;
2620  return (Node *) newexpr;
2621  }
2622  case T_OpExpr:
2623  {
2624  OpExpr *expr = (OpExpr *) node;
2625  List *args = expr->args;
2626  Expr *simple;
2627  OpExpr *newexpr;
2628 
2629  /*
2630  * Need to get OID of underlying function. Okay to scribble
2631  * on input to this extent.
2632  */
2633  set_opfuncid(expr);
2634 
2635  /*
2636  * Code for op/func reduction is pretty bulky, so split it out
2637  * as a separate function.
2638  */
2639  simple = simplify_function(expr->opfuncid,
2640  expr->opresulttype, -1,
2641  expr->opcollid,
2642  expr->inputcollid,
2643  &args,
2644  false,
2645  true,
2646  true,
2647  context);
2648  if (simple) /* successfully simplified it */
2649  return (Node *) simple;
2650 
2651  /*
2652  * If the operator is boolean equality or inequality, we know
2653  * how to simplify cases involving one constant and one
2654  * non-constant argument.
2655  */
2656  if (expr->opno == BooleanEqualOperator ||
2657  expr->opno == BooleanNotEqualOperator)
2658  {
2659  simple = (Expr *) simplify_boolean_equality(expr->opno,
2660  args);
2661  if (simple) /* successfully simplified it */
2662  return (Node *) simple;
2663  }
2664 
2665  /*
2666  * The expression cannot be simplified any further, so build
2667  * and return a replacement OpExpr node using the
2668  * possibly-simplified arguments.
2669  */
2670  newexpr = makeNode(OpExpr);
2671  newexpr->opno = expr->opno;
2672  newexpr->opfuncid = expr->opfuncid;
2673  newexpr->opresulttype = expr->opresulttype;
2674  newexpr->opretset = expr->opretset;
2675  newexpr->opcollid = expr->opcollid;
2676  newexpr->inputcollid = expr->inputcollid;
2677  newexpr->args = args;
2678  newexpr->location = expr->location;
2679  return (Node *) newexpr;
2680  }
2681  case T_DistinctExpr:
2682  {
2683  DistinctExpr *expr = (DistinctExpr *) node;
2684  List *args;
2685  ListCell *arg;
2686  bool has_null_input = false;
2687  bool all_null_input = true;
2688  bool has_nonconst_input = false;
2689  Expr *simple;
2690  DistinctExpr *newexpr;
2691 
2692  /*
2693  * Reduce constants in the DistinctExpr's arguments. We know
2694  * args is either NIL or a List node, so we can call
2695  * expression_tree_mutator directly rather than recursing to
2696  * self.
2697  */
2698  args = (List *) expression_tree_mutator((Node *) expr->args,
2700  (void *) context);
2701 
2702  /*
2703  * We must do our own check for NULLs because DistinctExpr has
2704  * different results for NULL input than the underlying
2705  * operator does.
2706  */
2707  foreach(arg, args)
2708  {
2709  if (IsA(lfirst(arg), Const))
2710  {
2711  has_null_input |= ((Const *) lfirst(arg))->constisnull;
2712  all_null_input &= ((Const *) lfirst(arg))->constisnull;
2713  }
2714  else
2715  has_nonconst_input = true;
2716  }
2717 
2718  /* all constants? then can optimize this out */
2719  if (!has_nonconst_input)
2720  {
2721  /* all nulls? then not distinct */
2722  if (all_null_input)
2723  return makeBoolConst(false, false);
2724 
2725  /* one null? then distinct */
2726  if (has_null_input)
2727  return makeBoolConst(true, false);
2728 
2729  /* otherwise try to evaluate the '=' operator */
2730  /* (NOT okay to try to inline it, though!) */
2731 
2732  /*
2733  * Need to get OID of underlying function. Okay to
2734  * scribble on input to this extent.
2735  */
2736  set_opfuncid((OpExpr *) expr); /* rely on struct
2737  * equivalence */
2738 
2739  /*
2740  * Code for op/func reduction is pretty bulky, so split it
2741  * out as a separate function.
2742  */
2743  simple = simplify_function(expr->opfuncid,
2744  expr->opresulttype, -1,
2745  expr->opcollid,
2746  expr->inputcollid,
2747  &args,
2748  false,
2749  false,
2750  false,
2751  context);
2752  if (simple) /* successfully simplified it */
2753  {
2754  /*
2755  * Since the underlying operator is "=", must negate
2756  * its result
2757  */
2758  Const *csimple = castNode(Const, simple);
2759 
2760  csimple->constvalue =
2761  BoolGetDatum(!DatumGetBool(csimple->constvalue));
2762  return (Node *) csimple;
2763  }
2764  }
2765 
2766  /*
2767  * The expression cannot be simplified any further, so build
2768  * and return a replacement DistinctExpr node using the
2769  * possibly-simplified arguments.
2770  */
2771  newexpr = makeNode(DistinctExpr);
2772  newexpr->opno = expr->opno;
2773  newexpr->opfuncid = expr->opfuncid;
2774  newexpr->opresulttype = expr->opresulttype;
2775  newexpr->opretset = expr->opretset;
2776  newexpr->opcollid = expr->opcollid;
2777  newexpr->inputcollid = expr->inputcollid;
2778  newexpr->args = args;
2779  newexpr->location = expr->location;
2780  return (Node *) newexpr;
2781  }
2782  case T_NullIfExpr:
2783  {
2784  NullIfExpr *expr;
2785  ListCell *arg;
2786  bool has_nonconst_input = false;
2787 
2788  /* Copy the node and const-simplify its arguments */
2789  expr = (NullIfExpr *) ece_generic_processing(node);
2790 
2791  /* If either argument is NULL they can't be equal */
2792  foreach(arg, expr->args)
2793  {
2794  if (!IsA(lfirst(arg), Const))
2795  has_nonconst_input = true;
2796  else if (((Const *) lfirst(arg))->constisnull)
2797  return (Node *) linitial(expr->args);
2798  }
2799 
2800  /*
2801  * Need to get OID of underlying function before checking if
2802  * the function is OK to evaluate.
2803  */
2804  set_opfuncid((OpExpr *) expr);
2805 
2806  if (!has_nonconst_input &&
2807  ece_function_is_safe(expr->opfuncid, context))
2808  return ece_evaluate_expr(expr);
2809 
2810  return (Node *) expr;
2811  }
2812  case T_ScalarArrayOpExpr:
2813  {
2814  ScalarArrayOpExpr *saop;
2815 
2816  /* Copy the node and const-simplify its arguments */
2817  saop = (ScalarArrayOpExpr *) ece_generic_processing(node);
2818 
2819  /* Make sure we know underlying function */
2820  set_sa_opfuncid(saop);
2821 
2822  /*
2823  * If all arguments are Consts, and it's a safe function, we
2824  * can fold to a constant
2825  */
2826  if (ece_all_arguments_const(saop) &&
2827  ece_function_is_safe(saop->opfuncid, context))
2828  return ece_evaluate_expr(saop);
2829  return (Node *) saop;
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);
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 
2915  case T_JsonValueExpr:
2916  {
2917  JsonValueExpr *jve = (JsonValueExpr *) node;
2918  Node *formatted;
2919 
2920  formatted = eval_const_expressions_mutator((Node *) jve->formatted_expr,
2921  context);
2922  if (formatted && IsA(formatted, Const))
2923  return formatted;
2924  break;
2925  }
2926 
2927  case T_SubPlan:
2928  case T_AlternativeSubPlan:
2929 
2930  /*
2931  * Return a SubPlan unchanged --- too late to do anything with it.
2932  *
2933  * XXX should we ereport() here instead? Probably this routine
2934  * should never be invoked after SubPlan creation.
2935  */
2936  return node;
2937  case T_RelabelType:
2938  {
2939  RelabelType *relabel = (RelabelType *) node;
2940  Node *arg;
2941 
2942  /* Simplify the input ... */
2943  arg = eval_const_expressions_mutator((Node *) relabel->arg,
2944  context);
2945  /* ... and attach a new RelabelType node, if needed */
2946  return applyRelabelType(arg,
2947  relabel->resulttype,
2948  relabel->resulttypmod,
2949  relabel->resultcollid,
2950  relabel->relabelformat,
2951  relabel->location,
2952  true);
2953  }
2954  case T_CoerceViaIO:
2955  {
2956  CoerceViaIO *expr = (CoerceViaIO *) node;
2957  List *args;
2958  Oid outfunc;
2959  bool outtypisvarlena;
2960  Oid infunc;
2961  Oid intypioparam;
2962  Expr *simple;
2963  CoerceViaIO *newexpr;
2964 
2965  /* Make a List so we can use simplify_function */
2966  args = list_make1(expr->arg);
2967 
2968  /*
2969  * CoerceViaIO represents calling the source type's output
2970  * function then the result type's input function. So, try to
2971  * simplify it as though it were a stack of two such function
2972  * calls. First we need to know what the functions are.
2973  *
2974  * Note that the coercion functions are assumed not to care
2975  * about input collation, so we just pass InvalidOid for that.
2976  */
2977  getTypeOutputInfo(exprType((Node *) expr->arg),
2978  &outfunc, &outtypisvarlena);
2980  &infunc, &intypioparam);
2981 
2982  simple = simplify_function(outfunc,
2983  CSTRINGOID, -1,
2984  InvalidOid,
2985  InvalidOid,
2986  &args,
2987  false,
2988  true,
2989  true,
2990  context);
2991  if (simple) /* successfully simplified output fn */
2992  {
2993  /*
2994  * Input functions may want 1 to 3 arguments. We always
2995  * supply all three, trusting that nothing downstream will
2996  * complain.
2997  */
2998  args = list_make3(simple,
2999  makeConst(OIDOID,
3000  -1,
3001  InvalidOid,
3002  sizeof(Oid),
3003  ObjectIdGetDatum(intypioparam),
3004  false,
3005  true),
3006  makeConst(INT4OID,
3007  -1,
3008  InvalidOid,
3009  sizeof(int32),
3010  Int32GetDatum(-1),
3011  false,
3012  true));
3013 
3014  simple = simplify_function(infunc,
3015  expr->resulttype, -1,
3016  expr->resultcollid,
3017  InvalidOid,
3018  &args,
3019  false,
3020  false,
3021  true,
3022  context);
3023  if (simple) /* successfully simplified input fn */
3024  return (Node *) simple;
3025  }
3026 
3027  /*
3028  * The expression cannot be simplified any further, so build
3029  * and return a replacement CoerceViaIO node using the
3030  * possibly-simplified argument.
3031  */
3032  newexpr = makeNode(CoerceViaIO);
3033  newexpr->arg = (Expr *) linitial(args);
3034  newexpr->resulttype = expr->resulttype;
3035  newexpr->resultcollid = expr->resultcollid;
3036  newexpr->coerceformat = expr->coerceformat;
3037  newexpr->location = expr->location;
3038  return (Node *) newexpr;
3039  }
3040  case T_ArrayCoerceExpr:
3041  {
3043  Node *save_case_val;
3044 
3045  /*
3046  * Copy the node and const-simplify its arguments. We can't
3047  * use ece_generic_processing() here because we need to mess
3048  * with case_val only while processing the elemexpr.
3049  */
3050  memcpy(ac, node, sizeof(ArrayCoerceExpr));
3051  ac->arg = (Expr *)
3053  context);
3054 
3055  /*
3056  * Set up for the CaseTestExpr node contained in the elemexpr.
3057  * We must prevent it from absorbing any outer CASE value.
3058  */
3059  save_case_val = context->case_val;
3060  context->case_val = NULL;
3061 
3062  ac->elemexpr = (Expr *)
3064  context);
3065 
3066  context->case_val = save_case_val;
3067 
3068  /*
3069  * If constant argument and the per-element expression is
3070  * immutable, we can simplify the whole thing to a constant.
3071  * Exception: although contain_mutable_functions considers
3072  * CoerceToDomain immutable for historical reasons, let's not
3073  * do so here; this ensures coercion to an array-over-domain
3074  * does not apply the domain's constraints until runtime.
3075  */
3076  if (ac->arg && IsA(ac->arg, Const) &&
3077  ac->elemexpr && !IsA(ac->elemexpr, CoerceToDomain) &&
3079  return ece_evaluate_expr(ac);
3080 
3081  return (Node *) ac;
3082  }
3083  case T_CollateExpr:
3084  {
3085  /*
3086  * We replace CollateExpr with RelabelType, so as to improve
3087  * uniformity of expression representation and thus simplify
3088  * comparison of expressions. Hence this looks very nearly
3089  * the same as the RelabelType case, and we can apply the same
3090  * optimizations to avoid unnecessary RelabelTypes.
3091  */
3092  CollateExpr *collate = (CollateExpr *) node;
3093  Node *arg;
3094 
3095  /* Simplify the input ... */
3096  arg = eval_const_expressions_mutator((Node *) collate->arg,
3097  context);
3098  /* ... and attach a new RelabelType node, if needed */
3099  return applyRelabelType(arg,
3100  exprType(arg),
3101  exprTypmod(arg),
3102  collate->collOid,
3104  collate->location,
3105  true);
3106  }
3107  case T_CaseExpr:
3108  {
3109  /*----------
3110  * CASE expressions can be simplified if there are constant
3111  * condition clauses:
3112  * FALSE (or NULL): drop the alternative
3113  * TRUE: drop all remaining alternatives
3114  * If the first non-FALSE alternative is a constant TRUE,
3115  * we can simplify the entire CASE to that alternative's
3116  * expression. If there are no non-FALSE alternatives,
3117  * we simplify the entire CASE to the default result (ELSE).
3118  *
3119  * If we have a simple-form CASE with constant test
3120  * expression, we substitute the constant value for contained
3121  * CaseTestExpr placeholder nodes, so that we have the
3122  * opportunity to reduce constant test conditions. For
3123  * example this allows
3124  * CASE 0 WHEN 0 THEN 1 ELSE 1/0 END
3125  * to reduce to 1 rather than drawing a divide-by-0 error.
3126  * Note that when the test expression is constant, we don't
3127  * have to include it in the resulting CASE; for example
3128  * CASE 0 WHEN x THEN y ELSE z END
3129  * is transformed by the parser to
3130  * CASE 0 WHEN CaseTestExpr = x THEN y ELSE z END
3131  * which we can simplify to
3132  * CASE WHEN 0 = x THEN y ELSE z END
3133  * It is not necessary for the executor to evaluate the "arg"
3134  * expression when executing the CASE, since any contained
3135  * CaseTestExprs that might have referred to it will have been
3136  * replaced by the constant.
3137  *----------
3138  */
3139  CaseExpr *caseexpr = (CaseExpr *) node;
3140  CaseExpr *newcase;
3141  Node *save_case_val;
3142  Node *newarg;
3143  List *newargs;
3144  bool const_true_cond;
3145  Node *defresult = NULL;
3146  ListCell *arg;
3147 
3148  /* Simplify the test expression, if any */
3149  newarg = eval_const_expressions_mutator((Node *) caseexpr->arg,
3150  context);
3151 
3152  /* Set up for contained CaseTestExpr nodes */
3153  save_case_val = context->case_val;
3154  if (newarg && IsA(newarg, Const))
3155  {
3156  context->case_val = newarg;
3157  newarg = NULL; /* not needed anymore, see above */
3158  }
3159  else
3160  context->case_val = NULL;
3161 
3162  /* Simplify the WHEN clauses */
3163  newargs = NIL;
3164  const_true_cond = false;
3165  foreach(arg, caseexpr->args)
3166  {
3167  CaseWhen *oldcasewhen = lfirst_node(CaseWhen, arg);
3168  Node *casecond;
3169  Node *caseresult;
3170 
3171  /* Simplify this alternative's test condition */
3172  casecond = eval_const_expressions_mutator((Node *) oldcasewhen->expr,
3173  context);
3174 
3175  /*
3176  * If the test condition is constant FALSE (or NULL), then
3177  * drop this WHEN clause completely, without processing
3178  * the result.
3179  */
3180  if (casecond && IsA(casecond, Const))
3181  {
3182  Const *const_input = (Const *) casecond;
3183 
3184  if (const_input->constisnull ||
3185  !DatumGetBool(const_input->constvalue))
3186  continue; /* drop alternative with FALSE cond */
3187  /* Else it's constant TRUE */
3188  const_true_cond = true;
3189  }
3190 
3191  /* Simplify this alternative's result value */
3192  caseresult = eval_const_expressions_mutator((Node *) oldcasewhen->result,
3193  context);
3194 
3195  /* If non-constant test condition, emit a new WHEN node */
3196  if (!const_true_cond)
3197  {
3198  CaseWhen *newcasewhen = makeNode(CaseWhen);
3199 
3200  newcasewhen->expr = (Expr *) casecond;
3201  newcasewhen->result = (Expr *) caseresult;
3202  newcasewhen->location = oldcasewhen->location;
3203  newargs = lappend(newargs, newcasewhen);
3204  continue;
3205  }
3206 
3207  /*
3208  * Found a TRUE condition, so none of the remaining
3209  * alternatives can be reached. We treat the result as
3210  * the default result.
3211  */
3212  defresult = caseresult;
3213  break;
3214  }
3215 
3216  /* Simplify the default result, unless we replaced it above */
3217  if (!const_true_cond)
3218  defresult = eval_const_expressions_mutator((Node *) caseexpr->defresult,
3219  context);
3220 
3221  context->case_val = save_case_val;
3222 
3223  /*
3224  * If no non-FALSE alternatives, CASE reduces to the default
3225  * result
3226  */
3227  if (newargs == NIL)
3228  return defresult;
3229  /* Otherwise we need a new CASE node */
3230  newcase = makeNode(CaseExpr);
3231  newcase->casetype = caseexpr->casetype;
3232  newcase->casecollid = caseexpr->casecollid;
3233  newcase->arg = (Expr *) newarg;
3234  newcase->args = newargs;
3235  newcase->defresult = (Expr *) defresult;
3236  newcase->location = caseexpr->location;
3237  return (Node *) newcase;
3238  }
3239  case T_CaseTestExpr:
3240  {
3241  /*
3242  * If we know a constant test value for the current CASE
3243  * construct, substitute it for the placeholder. Else just
3244  * return the placeholder as-is.
3245  */
3246  if (context->case_val)
3247  return copyObject(context->case_val);
3248  else
3249  return copyObject(node);
3250  }
3251  case T_SubscriptingRef:
3252  case T_ArrayExpr:
3253  case T_RowExpr:
3254  case T_MinMaxExpr:
3255  {
3256  /*
3257  * Generic handling for node types whose own processing is
3258  * known to be immutable, and for which we need no smarts
3259  * beyond "simplify if all inputs are constants".
3260  *
3261  * Treating SubscriptingRef this way assumes that subscripting
3262  * fetch and assignment are both immutable. This constrains
3263  * type-specific subscripting implementations; maybe we should
3264  * relax it someday.
3265  *
3266  * Treating MinMaxExpr this way amounts to assuming that the
3267  * btree comparison function it calls is immutable; see the
3268  * reasoning in contain_mutable_functions_walker.
3269  */
3270 
3271  /* Copy the node and const-simplify its arguments */
3272  node = ece_generic_processing(node);
3273  /* If all arguments are Consts, we can fold to a constant */
3274  if (ece_all_arguments_const(node))
3275  return ece_evaluate_expr(node);
3276  return node;
3277  }
3278  case T_CoalesceExpr:
3279  {
3280  CoalesceExpr *coalesceexpr = (CoalesceExpr *) node;
3281  CoalesceExpr *newcoalesce;
3282  List *newargs;
3283  ListCell *arg;
3284 
3285  newargs = NIL;
3286  foreach(arg, coalesceexpr->args)
3287  {
3288  Node *e;
3289 
3291  context);
3292 
3293  /*
3294  * We can remove null constants from the list. For a
3295  * non-null constant, if it has not been preceded by any
3296  * other non-null-constant expressions then it is the
3297  * result. Otherwise, it's the next argument, but we can
3298  * drop following arguments since they will never be
3299  * reached.
3300  */
3301  if (IsA(e, Const))
3302  {
3303  if (((Const *) e)->constisnull)
3304  continue; /* drop null constant */
3305  if (newargs == NIL)
3306  return e; /* first expr */
3307  newargs = lappend(newargs, e);
3308  break;
3309  }
3310  newargs = lappend(newargs, e);
3311  }
3312 
3313  /*
3314  * If all the arguments were constant null, the result is just
3315  * null
3316  */
3317  if (newargs == NIL)
3318  return (Node *) makeNullConst(coalesceexpr->coalescetype,
3319  -1,
3320  coalesceexpr->coalescecollid);
3321 
3322  newcoalesce = makeNode(CoalesceExpr);
3323  newcoalesce->coalescetype = coalesceexpr->coalescetype;
3324  newcoalesce->coalescecollid = coalesceexpr->coalescecollid;
3325  newcoalesce->args = newargs;
3326  newcoalesce->location = coalesceexpr->location;
3327  return (Node *) newcoalesce;
3328  }
3329  case T_SQLValueFunction:
3330  {
3331  /*
3332  * All variants of SQLValueFunction are stable, so if we are
3333  * estimating the expression's value, we should evaluate the
3334  * current function value. Otherwise just copy.
3335  */
3336  SQLValueFunction *svf = (SQLValueFunction *) node;
3337 
3338  if (context->estimate)
3339  return (Node *) evaluate_expr((Expr *) svf,
3340  svf->type,
3341  svf->typmod,
3342  InvalidOid);
3343  else
3344  return copyObject((Node *) svf);
3345  }
3346  case T_FieldSelect:
3347  {
3348  /*
3349  * We can optimize field selection from a whole-row Var into a
3350  * simple Var. (This case won't be generated directly by the
3351  * parser, because ParseComplexProjection short-circuits it.
3352  * But it can arise while simplifying functions.) Also, we
3353  * can optimize field selection from a RowExpr construct, or
3354  * of course from a constant.
3355  *
3356  * However, replacing a whole-row Var in this way has a
3357  * pitfall: if we've already built the rel targetlist for the
3358  * source relation, then the whole-row Var is scheduled to be
3359  * produced by the relation scan, but the simple Var probably
3360  * isn't, which will lead to a failure in setrefs.c. This is
3361  * not a problem when handling simple single-level queries, in
3362  * which expression simplification always happens first. It
3363  * is a risk for lateral references from subqueries, though.
3364  * To avoid such failures, don't optimize uplevel references.
3365  *
3366  * We must also check that the declared type of the field is
3367  * still the same as when the FieldSelect was created --- this
3368  * can change if someone did ALTER COLUMN TYPE on the rowtype.
3369  * If it isn't, we skip the optimization; the case will
3370  * probably fail at runtime, but that's not our problem here.
3371  */
3372  FieldSelect *fselect = (FieldSelect *) node;
3373  FieldSelect *newfselect;
3374  Node *arg;
3375 
3376  arg = eval_const_expressions_mutator((Node *) fselect->arg,
3377  context);
3378  if (arg && IsA(arg, Var) &&
3379  ((Var *) arg)->varattno == InvalidAttrNumber &&
3380  ((Var *) arg)->varlevelsup == 0)
3381  {
3382  if (rowtype_field_matches(((Var *) arg)->vartype,
3383  fselect->fieldnum,
3384  fselect->resulttype,
3385  fselect->resulttypmod,
3386  fselect->resultcollid))
3387  {
3388  Var *newvar;
3389 
3390  newvar = makeVar(((Var *) arg)->varno,
3391  fselect->fieldnum,
3392  fselect->resulttype,
3393  fselect->resulttypmod,
3394  fselect->resultcollid,
3395  ((Var *) arg)->varlevelsup);
3396  /* New Var is nullable by same rels as the old one */
3397  newvar->varnullingrels = ((Var *) arg)->varnullingrels;
3398  return (Node *) newvar;
3399  }
3400  }
3401  if (arg && IsA(arg, RowExpr))
3402  {
3403  RowExpr *rowexpr = (RowExpr *) arg;
3404 
3405  if (fselect->fieldnum > 0 &&
3406  fselect->fieldnum <= list_length(rowexpr->args))
3407  {
3408  Node *fld = (Node *) list_nth(rowexpr->args,
3409  fselect->fieldnum - 1);
3410 
3411  if (rowtype_field_matches(rowexpr->row_typeid,
3412  fselect->fieldnum,
3413  fselect->resulttype,
3414  fselect->resulttypmod,
3415  fselect->resultcollid) &&
3416  fselect->resulttype == exprType(fld) &&
3417  fselect->resulttypmod == exprTypmod(fld) &&
3418  fselect->resultcollid == exprCollation(fld))
3419  return fld;
3420  }
3421  }
3422  newfselect = makeNode(FieldSelect);
3423  newfselect->arg = (Expr *) arg;
3424  newfselect->fieldnum = fselect->fieldnum;
3425  newfselect->resulttype = fselect->resulttype;
3426  newfselect->resulttypmod = fselect->resulttypmod;
3427  newfselect->resultcollid = fselect->resultcollid;
3428  if (arg && IsA(arg, Const))
3429  {
3430  Const *con = (Const *) arg;
3431 
3433  newfselect->fieldnum,
3434  newfselect->resulttype,
3435  newfselect->resulttypmod,
3436  newfselect->resultcollid))
3437  return ece_evaluate_expr(newfselect);
3438  }
3439  return (Node *) newfselect;
3440  }
3441  case T_NullTest:
3442  {
3443  NullTest *ntest = (NullTest *) node;
3444  NullTest *newntest;
3445  Node *arg;
3446 
3448  context);
3449  if (ntest->argisrow && arg && IsA(arg, RowExpr))
3450  {
3451  /*
3452  * We break ROW(...) IS [NOT] NULL into separate tests on
3453  * its component fields. This form is usually more
3454  * efficient to evaluate, as well as being more amenable
3455  * to optimization.
3456  */
3457  RowExpr *rarg = (RowExpr *) arg;
3458  List *newargs = NIL;
3459  ListCell *l;
3460 
3461  foreach(l, rarg->args)
3462  {
3463  Node *relem = (Node *) lfirst(l);
3464 
3465  /*
3466  * A constant field refutes the whole NullTest if it's
3467  * of the wrong nullness; else we can discard it.
3468  */
3469  if (relem && IsA(relem, Const))
3470  {
3471  Const *carg = (Const *) relem;
3472 
3473  if (carg->constisnull ?
3474  (ntest->nulltesttype == IS_NOT_NULL) :
3475  (ntest->nulltesttype == IS_NULL))
3476  return makeBoolConst(false, false);
3477  continue;
3478  }
3479 
3480  /*
3481  * Else, make a scalar (argisrow == false) NullTest
3482  * for this field. Scalar semantics are required
3483  * because IS [NOT] NULL doesn't recurse; see comments
3484  * in ExecEvalRowNullInt().
3485  */
3486  newntest = makeNode(NullTest);
3487  newntest->arg = (Expr *) relem;
3488  newntest->nulltesttype = ntest->nulltesttype;
3489  newntest->argisrow = false;
3490  newntest->location = ntest->location;
3491  newargs = lappend(newargs, newntest);
3492  }
3493  /* If all the inputs were constants, result is TRUE */
3494  if (newargs == NIL)
3495  return makeBoolConst(true, false);
3496  /* If only one nonconst input, it's the result */
3497  if (list_length(newargs) == 1)
3498  return (Node *) linitial(newargs);
3499  /* Else we need an AND node */
3500  return (Node *) make_andclause(newargs);
3501  }
3502  if (!ntest->argisrow && arg && IsA(arg, Const))
3503  {
3504  Const *carg = (Const *) arg;
3505  bool result;
3506 
3507  switch (ntest->nulltesttype)
3508  {
3509  case IS_NULL:
3510  result = carg->constisnull;
3511  break;
3512  case IS_NOT_NULL:
3513  result = !carg->constisnull;
3514  break;
3515  default:
3516  elog(ERROR, "unrecognized nulltesttype: %d",
3517  (int) ntest->nulltesttype);
3518  result = false; /* keep compiler quiet */
3519  break;
3520  }
3521 
3522  return makeBoolConst(result, false);
3523  }
3524 
3525  newntest = makeNode(NullTest);
3526  newntest->arg = (Expr *) arg;
3527  newntest->nulltesttype = ntest->nulltesttype;
3528  newntest->argisrow = ntest->argisrow;
3529  newntest->location = ntest->location;
3530  return (Node *) newntest;
3531  }
3532  case T_BooleanTest:
3533  {
3534  /*
3535  * This case could be folded into the generic handling used
3536  * for ArrayExpr etc. But because the simplification logic is
3537  * so trivial, applying evaluate_expr() to perform it would be
3538  * a heavy overhead. BooleanTest is probably common enough to
3539  * justify keeping this bespoke implementation.
3540  */
3541  BooleanTest *btest = (BooleanTest *) node;
3542  BooleanTest *newbtest;
3543  Node *arg;
3544 
3546  context);
3547  if (arg && IsA(arg, Const))
3548  {
3549  Const *carg = (Const *) arg;
3550  bool result;
3551 
3552  switch (btest->booltesttype)
3553  {
3554  case IS_TRUE:
3555  result = (!carg->constisnull &&
3556  DatumGetBool(carg->constvalue));
3557  break;
3558  case IS_NOT_TRUE:
3559  result = (carg->constisnull ||
3560  !DatumGetBool(carg->constvalue));
3561  break;
3562  case IS_FALSE:
3563  result = (!carg->constisnull &&
3564  !DatumGetBool(carg->constvalue));
3565  break;
3566  case IS_NOT_FALSE:
3567  result = (carg->constisnull ||
3568  DatumGetBool(carg->constvalue));
3569  break;
3570  case IS_UNKNOWN:
3571  result = carg->constisnull;
3572  break;
3573  case IS_NOT_UNKNOWN:
3574  result = !carg->constisnull;
3575  break;
3576  default:
3577  elog(ERROR, "unrecognized booltesttype: %d",
3578  (int) btest->booltesttype);
3579  result = false; /* keep compiler quiet */
3580  break;
3581  }
3582 
3583  return makeBoolConst(result, false);
3584  }
3585 
3586  newbtest = makeNode(BooleanTest);
3587  newbtest->arg = (Expr *) arg;
3588  newbtest->booltesttype = btest->booltesttype;
3589  newbtest->location = btest->location;
3590  return (Node *) newbtest;
3591  }
3592  case T_CoerceToDomain:
3593  {
3594  /*
3595  * If the domain currently has no constraints, we replace the
3596  * CoerceToDomain node with a simple RelabelType, which is
3597  * both far faster to execute and more amenable to later
3598  * optimization. We must then mark the plan as needing to be
3599  * rebuilt if the domain's constraints change.
3600  *
3601  * Also, in estimation mode, always replace CoerceToDomain
3602  * nodes, effectively assuming that the coercion will succeed.
3603  */
3604  CoerceToDomain *cdomain = (CoerceToDomain *) node;
3605  CoerceToDomain *newcdomain;
3606  Node *arg;
3607 
3608  arg = eval_const_expressions_mutator((Node *) cdomain->arg,
3609  context);
3610  if (context->estimate ||
3611  !DomainHasConstraints(cdomain->resulttype))
3612  {
3613  /* Record dependency, if this isn't estimation mode */
3614  if (context->root && !context->estimate)
3616  cdomain->resulttype);
3617 
3618  /* Generate RelabelType to substitute for CoerceToDomain */
3619  return applyRelabelType(arg,
3620  cdomain->resulttype,
3621  cdomain->resulttypmod,
3622  cdomain->resultcollid,
3623  cdomain->coercionformat,
3624  cdomain->location,
3625  true);
3626  }
3627 
3628  newcdomain = makeNode(CoerceToDomain);
3629  newcdomain->arg = (Expr *) arg;
3630  newcdomain->resulttype = cdomain->resulttype;
3631  newcdomain->resulttypmod = cdomain->resulttypmod;
3632  newcdomain->resultcollid = cdomain->resultcollid;
3633  newcdomain->coercionformat = cdomain->coercionformat;
3634  newcdomain->location = cdomain->location;
3635  return (Node *) newcdomain;
3636  }
3637  case T_PlaceHolderVar:
3638 
3639  /*
3640  * In estimation mode, just strip the PlaceHolderVar node
3641  * altogether; this amounts to estimating that the contained value
3642  * won't be forced to null by an outer join. In regular mode we
3643  * just use the default behavior (ie, simplify the expression but
3644  * leave the PlaceHolderVar node intact).
3645  */
3646  if (context->estimate)
3647  {
3648  PlaceHolderVar *phv = (PlaceHolderVar *) node;
3649 
3650  return eval_const_expressions_mutator((Node *) phv->phexpr,
3651  context);
3652  }
3653  break;
3654  case T_ConvertRowtypeExpr:
3655  {
3657  Node *arg;
3658  ConvertRowtypeExpr *newcre;
3659 
3661  context);
3662 
3663  newcre = makeNode(ConvertRowtypeExpr);
3664  newcre->resulttype = cre->resulttype;
3665  newcre->convertformat = cre->convertformat;
3666  newcre->location = cre->location;
3667 
3668  /*
3669  * In case of a nested ConvertRowtypeExpr, we can convert the
3670  * leaf row directly to the topmost row format without any
3671  * intermediate conversions. (This works because
3672  * ConvertRowtypeExpr is used only for child->parent
3673  * conversion in inheritance trees, which works by exact match
3674  * of column name, and a column absent in an intermediate
3675  * result can't be present in the final result.)
3676  *
3677  * No need to check more than one level deep, because the
3678  * above recursion will have flattened anything else.
3679  */
3680  if (arg != NULL && IsA(arg, ConvertRowtypeExpr))
3681  {
3683 
3684  arg = (Node *) argcre->arg;
3685 
3686  /*
3687  * Make sure an outer implicit conversion can't hide an
3688  * inner explicit one.
3689  */
3690  if (newcre->convertformat == COERCE_IMPLICIT_CAST)
3691  newcre->convertformat = argcre->convertformat;
3692  }
3693 
3694  newcre->arg = (Expr *) arg;
3695 
3696  if (arg != NULL && IsA(arg, Const))
3697  return ece_evaluate_expr((Node *) newcre);
3698  return (Node *) newcre;
3699  }
3700  default:
3701  break;
3702  }
3703 
3704  /*
3705  * For any node type not handled above, copy the node unchanged but
3706  * const-simplify its subexpressions. This is the correct thing for node
3707  * types whose behavior might change between planning and execution, such
3708  * as CurrentOfExpr. It's also a safe default for new node types not
3709  * known to this routine.
3710  */
3711  return ece_generic_processing(node);
3712 }
3713 
3714 /*
3715  * Subroutine for eval_const_expressions: check for non-Const nodes.
3716  *
3717  * We can abort recursion immediately on finding a non-Const node. This is
3718  * critical for performance, else eval_const_expressions_mutator would take
3719  * O(N^2) time on non-simplifiable trees. However, we do need to descend
3720  * into List nodes since expression_tree_walker sometimes invokes the walker
3721  * function directly on List subtrees.
3722  */
3723 static bool
3725 {
3726  if (node == NULL)
3727  return false;
3728  if (IsA(node, Const))
3729  return false;
3730  if (IsA(node, List))
3732  /* Otherwise, abort the tree traversal and return true */
3733  return true;
3734 }
3735 
3736 /*
3737  * Subroutine for eval_const_expressions: check if a function is OK to evaluate
3738  */
3739 static bool
3741 {
3742  char provolatile = func_volatile(funcid);
3743 
3744  /*
3745  * Ordinarily we are only allowed to simplify immutable functions. But for
3746  * purposes of estimation, we consider it okay to simplify functions that
3747  * are merely stable; the risk that the result might change from planning
3748  * time to execution time is worth taking in preference to not being able
3749  * to estimate the value at all.
3750  */
3751  if (provolatile == PROVOLATILE_IMMUTABLE)
3752  return true;
3753  if (context->estimate && provolatile == PROVOLATILE_STABLE)
3754  return true;
3755  return false;
3756 }
3757 
3758 /*
3759  * Subroutine for eval_const_expressions: process arguments of an OR clause
3760  *
3761  * This includes flattening of nested ORs as well as recursion to
3762  * eval_const_expressions to simplify the OR arguments.
3763  *
3764  * After simplification, OR arguments are handled as follows:
3765  * non constant: keep
3766  * FALSE: drop (does not affect result)
3767  * TRUE: force result to TRUE
3768  * NULL: keep only one
3769  * We must keep one NULL input because OR expressions evaluate to NULL when no
3770  * input is TRUE and at least one is NULL. We don't actually include the NULL
3771  * here, that's supposed to be done by the caller.
3772  *
3773  * The output arguments *haveNull and *forceTrue must be initialized false
3774  * by the caller. They will be set true if a NULL constant or TRUE constant,
3775  * respectively, is detected anywhere in the argument list.
3776  */
3777 static List *
3780  bool *haveNull, bool *forceTrue)
3781 {
3782  List *newargs = NIL;
3783  List *unprocessed_args;
3784 
3785  /*
3786  * We want to ensure that any OR immediately beneath another OR gets
3787  * flattened into a single OR-list, so as to simplify later reasoning.
3788  *
3789  * To avoid stack overflow from recursion of eval_const_expressions, we
3790  * resort to some tenseness here: we keep a list of not-yet-processed
3791  * inputs, and handle flattening of nested ORs by prepending to the to-do
3792  * list instead of recursing. Now that the parser generates N-argument
3793  * ORs from simple lists, this complexity is probably less necessary than
3794  * it once was, but we might as well keep the logic.
3795  */
3796  unprocessed_args = list_copy(args);
3797  while (unprocessed_args)
3798  {
3799  Node *arg = (Node *) linitial(unprocessed_args);
3800 
3801  unprocessed_args = list_delete_first(unprocessed_args);
3802 
3803  /* flatten nested ORs as per above comment */
3804  if (is_orclause(arg))
3805  {
3806  List *subargs = ((BoolExpr *) arg)->args;
3807  List *oldlist = unprocessed_args;
3808 
3809  unprocessed_args = list_concat_copy(subargs, unprocessed_args);
3810  /* perhaps-overly-tense code to avoid leaking old lists */
3811  list_free(oldlist);
3812  continue;
3813  }
3814 
3815  /* If it's not an OR, simplify it */
3817 
3818  /*
3819  * It is unlikely but not impossible for simplification of a non-OR
3820  * clause to produce an OR. Recheck, but don't be too tense about it
3821  * since it's not a mainstream case. In particular we don't worry
3822  * about const-simplifying the input twice, nor about list leakage.
3823  */
3824  if (is_orclause(arg))
3825  {
3826  List *subargs = ((BoolExpr *) arg)->args;
3827 
3828  unprocessed_args = list_concat_copy(subargs, unprocessed_args);
3829  continue;
3830  }
3831 
3832  /*
3833  * OK, we have a const-simplified non-OR argument. Process it per
3834  * comments above.
3835  */
3836  if (IsA(arg, Const))
3837  {
3838  Const *const_input = (Const *) arg;
3839 
3840  if (const_input->constisnull)
3841  *haveNull = true;
3842  else if (DatumGetBool(const_input->constvalue))
3843  {
3844  *forceTrue = true;
3845 
3846  /*
3847  * Once we detect a TRUE result we can just exit the loop
3848  * immediately. However, if we ever add a notion of
3849  * non-removable functions, we'd need to keep scanning.
3850  */
3851  return NIL;
3852  }
3853  /* otherwise, we can drop the constant-false input */
3854  continue;
3855  }
3856 
3857  /* else emit the simplified arg into the result list */
3858  newargs = lappend(newargs, arg);
3859  }
3860 
3861  return newargs;
3862 }
3863 
3864 /*
3865  * Subroutine for eval_const_expressions: process arguments of an AND clause
3866  *
3867  * This includes flattening of nested ANDs as well as recursion to
3868  * eval_const_expressions to simplify the AND arguments.
3869  *
3870  * After simplification, AND arguments are handled as follows:
3871  * non constant: keep
3872  * TRUE: drop (does not affect result)
3873  * FALSE: force result to FALSE
3874  * NULL: keep only one
3875  * We must keep one NULL input because AND expressions evaluate to NULL when
3876  * no input is FALSE and at least one is NULL. We don't actually include the
3877  * NULL here, that's supposed to be done by the caller.
3878  *
3879  * The output arguments *haveNull and *forceFalse must be initialized false
3880  * by the caller. They will be set true if a null constant or false constant,
3881  * respectively, is detected anywhere in the argument list.
3882  */
3883 static List *
3886  bool *haveNull, bool *forceFalse)
3887 {
3888  List *newargs = NIL;
3889  List *unprocessed_args;
3890 
3891  /* See comments in simplify_or_arguments */
3892  unprocessed_args = list_copy(args);
3893  while (unprocessed_args)
3894  {
3895  Node *arg = (Node *) linitial(unprocessed_args);
3896 
3897  unprocessed_args = list_delete_first(unprocessed_args);
3898 
3899  /* flatten nested ANDs as per above comment */
3900  if (is_andclause(arg))
3901  {
3902  List *subargs = ((BoolExpr *) arg)->args;
3903  List *oldlist = unprocessed_args;
3904 
3905  unprocessed_args = list_concat_copy(subargs, unprocessed_args);
3906  /* perhaps-overly-tense code to avoid leaking old lists */
3907  list_free(oldlist);
3908  continue;
3909  }
3910 
3911  /* If it's not an AND, simplify it */
3913 
3914  /*
3915  * It is unlikely but not impossible for simplification of a non-AND
3916  * clause to produce an AND. Recheck, but don't be too tense about it
3917  * since it's not a mainstream case. In particular we don't worry
3918  * about const-simplifying the input twice, nor about list leakage.
3919  */
3920  if (is_andclause(arg))
3921  {
3922  List *subargs = ((BoolExpr *) arg)->args;
3923 
3924  unprocessed_args = list_concat_copy(subargs, unprocessed_args);
3925  continue;
3926  }
3927 
3928  /*
3929  * OK, we have a const-simplified non-AND argument. Process it per
3930  * comments above.
3931  */
3932  if (IsA(arg, Const))
3933  {
3934  Const *const_input = (Const *) arg;
3935 
3936  if (const_input->constisnull)
3937  *haveNull = true;
3938  else if (!DatumGetBool(const_input->constvalue))
3939  {
3940  *forceFalse = true;
3941 
3942  /*
3943  * Once we detect a FALSE result we can just exit the loop
3944  * immediately. However, if we ever add a notion of
3945  * non-removable functions, we'd need to keep scanning.
3946  */
3947  return NIL;
3948  }
3949  /* otherwise, we can drop the constant-true input */
3950  continue;
3951  }
3952 
3953  /* else emit the simplified arg into the result list */
3954  newargs = lappend(newargs, arg);
3955  }
3956 
3957  return newargs;
3958 }
3959 
3960 /*
3961  * Subroutine for eval_const_expressions: try to simplify boolean equality
3962  * or inequality condition
3963  *
3964  * Inputs are the operator OID and the simplified arguments to the operator.
3965  * Returns a simplified expression if successful, or NULL if cannot
3966  * simplify the expression.
3967  *
3968  * The idea here is to reduce "x = true" to "x" and "x = false" to "NOT x",
3969  * or similarly "x <> true" to "NOT x" and "x <> false" to "x".
3970  * This is only marginally useful in itself, but doing it in constant folding
3971  * ensures that we will recognize these forms as being equivalent in, for
3972  * example, partial index matching.
3973  *
3974  * We come here only if simplify_function has failed; therefore we cannot
3975  * see two constant inputs, nor a constant-NULL input.
3976  */
3977 static Node *
3979 {
3980  Node *leftop;
3981  Node *rightop;
3982 
3983  Assert(list_length(args) == 2);
3984  leftop = linitial(args);
3985  rightop = lsecond(args);
3986  if (leftop && IsA(leftop, Const))
3987  {
3988  Assert(!((Const *) leftop)->constisnull);
3989  if (opno == BooleanEqualOperator)
3990  {
3991  if (DatumGetBool(((Const *) leftop)->constvalue))
3992  return rightop; /* true = foo */
3993  else
3994  return negate_clause(rightop); /* false = foo */
3995  }
3996  else
3997  {
3998  if (DatumGetBool(((Const *) leftop)->constvalue))
3999  return negate_clause(rightop); /* true <> foo */
4000  else
4001  return rightop; /* false <> foo */
4002  }
4003  }
4004  if (rightop && IsA(rightop, Const))
4005  {
4006  Assert(!((Const *) rightop)->constisnull);
4007  if (opno == BooleanEqualOperator)
4008  {
4009  if (DatumGetBool(((Const *) rightop)->constvalue))
4010  return leftop; /* foo = true */
4011  else
4012  return negate_clause(leftop); /* foo = false */
4013  }
4014  else
4015  {
4016  if (DatumGetBool(((Const *) rightop)->constvalue))
4017  return negate_clause(leftop); /* foo <> true */
4018  else
4019  return leftop; /* foo <> false */
4020  }
4021  }
4022  return NULL;
4023 }
4024 
4025 /*
4026  * Subroutine for eval_const_expressions: try to simplify a function call
4027  * (which might originally have been an operator; we don't care)
4028  *
4029  * Inputs are the function OID, actual result type OID (which is needed for
4030  * polymorphic functions), result typmod, result collation, the input
4031  * collation to use for the function, the original argument list (not
4032  * const-simplified yet, unless process_args is false), and some flags;
4033  * also the context data for eval_const_expressions.
4034  *
4035  * Returns a simplified expression if successful, or NULL if cannot
4036  * simplify the function call.
4037  *
4038  * This function is also responsible for converting named-notation argument
4039  * lists into positional notation and/or adding any needed default argument
4040  * expressions; which is a bit grotty, but it avoids extra fetches of the
4041  * function's pg_proc tuple. For this reason, the args list is
4042  * pass-by-reference. Conversion and const-simplification of the args list
4043  * will be done even if simplification of the function call itself is not
4044  * possible.
4045  */
4046 static Expr *
4047 simplify_function(Oid funcid, Oid result_type, int32 result_typmod,
4048  Oid result_collid, Oid input_collid, List **args_p,
4049  bool funcvariadic, bool process_args, bool allow_non_const,
4051 {
4052  List *args = *args_p;
4053  HeapTuple func_tuple;
4054  Form_pg_proc func_form;
4055  Expr *newexpr;
4056 
4057  /*
4058  * We have three strategies for simplification: execute the function to
4059  * deliver a constant result, use a transform function to generate a
4060  * substitute node tree, or expand in-line the body of the function
4061  * definition (which only works for simple SQL-language functions, but
4062  * that is a common case). Each case needs access to the function's
4063  * pg_proc tuple, so fetch it just once.
4064  *
4065  * Note: the allow_non_const flag suppresses both the second and third
4066  * strategies; so if !allow_non_const, simplify_function can only return a
4067  * Const or NULL. Argument-list rewriting happens anyway, though.
4068  */
4069  func_tuple = SearchSysCache1(PROCOID, ObjectIdGetDatum(funcid));
4070  if (!HeapTupleIsValid(func_tuple))
4071  elog(ERROR, "cache lookup failed for function %u", funcid);
4072  func_form = (Form_pg_proc) GETSTRUCT(func_tuple);
4073 
4074  /*
4075  * Process the function arguments, unless the caller did it already.
4076  *
4077  * Here we must deal with named or defaulted arguments, and then
4078  * recursively apply eval_const_expressions to the whole argument list.
4079  */
4080  if (process_args)
4081  {
4082  args = expand_function_arguments(args, false, result_type, func_tuple);
4085  (void *) context);
4086  /* Argument processing done, give it back to the caller */
4087  *args_p = args;
4088  }
4089 
4090  /* Now attempt simplification of the function call proper. */
4091 
4092  newexpr = evaluate_function(funcid, result_type, result_typmod,
4093  result_collid, input_collid,
4094  args, funcvariadic,
4095  func_tuple, context);
4096 
4097  if (!newexpr && allow_non_const && OidIsValid(func_form->prosupport))
4098  {
4099  /*
4100  * Build a SupportRequestSimplify node to pass to the support
4101  * function, pointing to a dummy FuncExpr node containing the
4102  * simplified arg list. We use this approach to present a uniform
4103  * interface to the support function regardless of how the target
4104  * function is actually being invoked.
4105  */
4107  FuncExpr fexpr;
4108 
4109  fexpr.xpr.type = T_FuncExpr;
4110  fexpr.funcid = funcid;
4111  fexpr.funcresulttype = result_type;
4112  fexpr.funcretset = func_form->proretset;
4113  fexpr.funcvariadic = funcvariadic;
4114  fexpr.funcformat = COERCE_EXPLICIT_CALL;
4115  fexpr.funccollid = result_collid;
4116  fexpr.inputcollid = input_collid;
4117  fexpr.args = args;
4118  fexpr.location = -1;
4119 
4120  req.type = T_SupportRequestSimplify;
4121  req.root = context->root;
4122  req.fcall = &fexpr;
4123 
4124  newexpr = (Expr *)
4125  DatumGetPointer(OidFunctionCall1(func_form->prosupport,
4126  PointerGetDatum(&req)));
4127 
4128  /* catch a possible API misunderstanding */
4129  Assert(newexpr != (Expr *) &fexpr);
4130  }
4131 
4132  if (!newexpr && allow_non_const)
4133  newexpr = inline_function(funcid, result_type, result_collid,
4134  input_collid, args, funcvariadic,
4135  func_tuple, context);
4136 
4137  ReleaseSysCache(func_tuple);
4138 
4139  return newexpr;
4140 }
4141 
4142 /*
4143  * expand_function_arguments: convert named-notation args to positional args
4144  * and/or insert default args, as needed
4145  *
4146  * Returns a possibly-transformed version of the args list.
4147  *
4148  * If include_out_arguments is true, then the args list and the result
4149  * include OUT arguments.
4150  *
4151  * The expected result type of the call must be given, for sanity-checking
4152  * purposes. Also, we ask the caller to provide the function's actual
4153  * pg_proc tuple, not just its OID.
4154  *
4155  * If we need to change anything, the input argument list is copied, not
4156  * modified.
4157  *
4158  * Note: this gets applied to operator argument lists too, even though the
4159  * cases it handles should never occur there. This should be OK since it
4160  * will fall through very quickly if there's nothing to do.
4161  */
4162 List *
4163 expand_function_arguments(List *args, bool include_out_arguments,
4164  Oid result_type, HeapTuple func_tuple)
4165 {
4166  Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4167  Oid *proargtypes = funcform->proargtypes.values;
4168  int pronargs = funcform->pronargs;
4169  bool has_named_args = false;
4170  ListCell *lc;
4171 
4172  /*
4173  * If we are asked to match to OUT arguments, then use the proallargtypes
4174  * array (which includes those); otherwise use proargtypes (which
4175  * doesn't). Of course, if proallargtypes is null, we always use
4176  * proargtypes. (Fetching proallargtypes is annoyingly expensive
4177  * considering that we may have nothing to do here, but fortunately the
4178  * common case is include_out_arguments == false.)
4179  */
4180  if (include_out_arguments)
4181  {
4182  Datum proallargtypes;
4183  bool isNull;
4184 
4185  proallargtypes = SysCacheGetAttr(PROCOID, func_tuple,
4186  Anum_pg_proc_proallargtypes,
4187  &isNull);
4188  if (!isNull)
4189  {
4190  ArrayType *arr = DatumGetArrayTypeP(proallargtypes);
4191 
4192  pronargs = ARR_DIMS(arr)[0];
4193  if (ARR_NDIM(arr) != 1 ||
4194  pronargs < 0 ||
4195  ARR_HASNULL(arr) ||
4196  ARR_ELEMTYPE(arr) != OIDOID)
4197  elog(ERROR, "proallargtypes is not a 1-D Oid array or it contains nulls");
4198  Assert(pronargs >= funcform->pronargs);
4199  proargtypes = (Oid *) ARR_DATA_PTR(arr);
4200  }
4201  }
4202 
4203  /* Do we have any named arguments? */
4204  foreach(lc, args)
4205  {
4206  Node *arg = (Node *) lfirst(lc);
4207 
4208  if (IsA(arg, NamedArgExpr))
4209  {
4210  has_named_args = true;
4211  break;
4212  }
4213  }
4214 
4215  /* If so, we must apply reorder_function_arguments */
4216  if (has_named_args)
4217  {
4218  args = reorder_function_arguments(args, pronargs, func_tuple);
4219  /* Recheck argument types and add casts if needed */
4220  recheck_cast_function_args(args, result_type,
4221  proargtypes, pronargs,
4222  func_tuple);
4223  }
4224  else if (list_length(args) < pronargs)
4225  {
4226  /* No named args, but we seem to be short some defaults */
4227  args = add_function_defaults(args, pronargs, func_tuple);
4228  /* Recheck argument types and add casts if needed */
4229  recheck_cast_function_args(args, result_type,
4230  proargtypes, pronargs,
4231  func_tuple);
4232  }
4233 
4234  return args;
4235 }
4236 
4237 /*
4238  * reorder_function_arguments: convert named-notation args to positional args
4239  *
4240  * This function also inserts default argument values as needed, since it's
4241  * impossible to form a truly valid positional call without that.
4242  */
4243 static List *
4245 {
4246  Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4247  int nargsprovided = list_length(args);
4248  Node *argarray[FUNC_MAX_ARGS];
4249  ListCell *lc;
4250  int i;
4251 
4252  Assert(nargsprovided <= pronargs);
4253  if (pronargs < 0 || pronargs > FUNC_MAX_ARGS)
4254  elog(ERROR, "too many function arguments");
4255  memset(argarray, 0, pronargs * sizeof(Node *));
4256 
4257  /* Deconstruct the argument list into an array indexed by argnumber */
4258  i = 0;
4259  foreach(lc, args)
4260  {
4261  Node *arg = (Node *) lfirst(lc);
4262 
4263  if (!IsA(arg, NamedArgExpr))
4264  {
4265  /* positional argument, assumed to precede all named args */
4266  Assert(argarray[i] == NULL);
4267  argarray[i++] = arg;
4268  }
4269  else
4270  {
4271  NamedArgExpr *na = (NamedArgExpr *) arg;
4272 
4273  Assert(na->argnumber >= 0 && na->argnumber < pronargs);
4274  Assert(argarray[na->argnumber] == NULL);
4275  argarray[na->argnumber] = (Node *) na->arg;
4276  }
4277  }
4278 
4279  /*
4280  * Fetch default expressions, if needed, and insert into array at proper
4281  * locations (they aren't necessarily consecutive or all used)
4282  */
4283  if (nargsprovided < pronargs)
4284  {
4285  List *defaults = fetch_function_defaults(func_tuple);
4286 
4287  i = pronargs - funcform->pronargdefaults;
4288  foreach(lc, defaults)
4289  {
4290  if (argarray[i] == NULL)
4291  argarray[i] = (Node *) lfirst(lc);
4292  i++;
4293  }
4294  }
4295 
4296  /* Now reconstruct the args list in proper order */
4297  args = NIL;
4298  for (i = 0; i < pronargs; i++)
4299  {
4300  Assert(argarray[i] != NULL);
4301  args = lappend(args, argarray[i]);
4302  }
4303 
4304  return args;
4305 }
4306 
4307 /*
4308  * add_function_defaults: add missing function arguments from its defaults
4309  *
4310  * This is used only when the argument list was positional to begin with,
4311  * and so we know we just need to add defaults at the end.
4312  */
4313 static List *
4315 {
4316  int nargsprovided = list_length(args);
4317  List *defaults;
4318  int ndelete;
4319 
4320  /* Get all the default expressions from the pg_proc tuple */
4321  defaults = fetch_function_defaults(func_tuple);
4322 
4323  /* Delete any unused defaults from the list */
4324  ndelete = nargsprovided + list_length(defaults) - pronargs;
4325  if (ndelete < 0)
4326  elog(ERROR, "not enough default arguments");
4327  if (ndelete > 0)
4328  defaults = list_delete_first_n(defaults, ndelete);
4329 
4330  /* And form the combined argument list, not modifying the input list */
4331  return list_concat_copy(args, defaults);
4332 }
4333 
4334 /*
4335  * fetch_function_defaults: get function's default arguments as expression list
4336  */
4337 static List *
4339 {
4340  List *defaults;
4341  Datum proargdefaults;
4342  char *str;
4343 
4344  proargdefaults = SysCacheGetAttrNotNull(PROCOID, func_tuple,
4345  Anum_pg_proc_proargdefaults);
4346  str = TextDatumGetCString(proargdefaults);
4347  defaults = castNode(List, stringToNode(str));
4348  pfree(str);
4349  return defaults;
4350 }
4351 
4352 /*
4353  * recheck_cast_function_args: recheck function args and typecast as needed
4354  * after adding defaults.
4355  *
4356  * It is possible for some of the defaulted arguments to be polymorphic;
4357  * therefore we can't assume that the default expressions have the correct
4358  * data types already. We have to re-resolve polymorphics and do coercion
4359  * just like the parser did.
4360  *
4361  * This should be a no-op if there are no polymorphic arguments,
4362  * but we do it anyway to be sure.
4363  *
4364  * Note: if any casts are needed, the args list is modified in-place;
4365  * caller should have already copied the list structure.
4366  */
4367 static void
4369  Oid *proargtypes, int pronargs,
4370  HeapTuple func_tuple)
4371 {
4372  Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4373  int nargs;
4374  Oid actual_arg_types[FUNC_MAX_ARGS];
4375  Oid declared_arg_types[FUNC_MAX_ARGS];
4376  Oid rettype;
4377  ListCell *lc;
4378 
4380  elog(ERROR, "too many function arguments");
4381  nargs = 0;
4382  foreach(lc, args)
4383  {
4384  actual_arg_types[nargs++] = exprType((Node *) lfirst(lc));
4385  }
4386  Assert(nargs == pronargs);
4387  memcpy(declared_arg_types, proargtypes, pronargs * sizeof(Oid));
4388  rettype = enforce_generic_type_consistency(actual_arg_types,
4389  declared_arg_types,
4390  nargs,
4391  funcform->prorettype,
4392  false);
4393  /* let's just check we got the same answer as the parser did ... */
4394  if (rettype != result_type)
4395  elog(ERROR, "function's resolved result type changed during planning");
4396 
4397  /* perform any necessary typecasting of arguments */
4398  make_fn_arguments(NULL, args, actual_arg_types, declared_arg_types);
4399 }
4400 
4401 /*
4402  * evaluate_function: try to pre-evaluate a function call
4403  *
4404  * We can do this if the function is strict and has any constant-null inputs
4405  * (just return a null constant), or if the function is immutable and has all
4406  * constant inputs (call it and return the result as a Const node). In
4407  * estimation mode we are willing to pre-evaluate stable functions too.
4408  *
4409  * Returns a simplified expression if successful, or NULL if cannot
4410  * simplify the function.
4411  */
4412 static Expr *
4413 evaluate_function(Oid funcid, Oid result_type, int32 result_typmod,
4414  Oid result_collid, Oid input_collid, List *args,
4415  bool funcvariadic,
4416  HeapTuple func_tuple,
4418 {
4419  Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4420  bool has_nonconst_input = false;
4421  bool has_null_input = false;
4422  ListCell *arg;
4423  FuncExpr *newexpr;
4424 
4425  /*
4426  * Can't simplify if it returns a set.
4427  */
4428  if (funcform->proretset)
4429  return NULL;
4430 
4431  /*
4432  * Can't simplify if it returns RECORD. The immediate problem is that it
4433  * will be needing an expected tupdesc which we can't supply here.
4434  *
4435  * In the case where it has OUT parameters, we could build an expected
4436  * tupdesc from those, but there may be other gotchas lurking. In
4437  * particular, if the function were to return NULL, we would produce a
4438  * null constant with no remaining indication of which concrete record
4439  * type it is. For now, seems best to leave the function call unreduced.
4440  */
4441  if (funcform->prorettype == RECORDOID)
4442  return NULL;
4443 
4444  /*
4445  * Check for constant inputs and especially constant-NULL inputs.
4446  */
4447  foreach(arg, args)
4448  {
4449  if (IsA(lfirst(arg), Const))
4450  has_null_input |= ((Const *) lfirst(arg))->constisnull;
4451  else
4452  has_nonconst_input = true;
4453  }
4454 
4455  /*
4456  * If the function is strict and has a constant-NULL input, it will never
4457  * be called at all, so we can replace the call by a NULL constant, even
4458  * if there are other inputs that aren't constant, and even if the
4459  * function is not otherwise immutable.
4460  */
4461  if (funcform->proisstrict && has_null_input)
4462  return (Expr *) makeNullConst(result_type, result_typmod,
4463  result_collid);
4464 
4465  /*
4466  * Otherwise, can simplify only if all inputs are constants. (For a
4467  * non-strict function, constant NULL inputs are treated the same as
4468  * constant non-NULL inputs.)
4469  */
4470  if (has_nonconst_input)
4471  return NULL;
4472 
4473  /*
4474  * Ordinarily we are only allowed to simplify immutable functions. But for
4475  * purposes of estimation, we consider it okay to simplify functions that
4476  * are merely stable; the risk that the result might change from planning
4477  * time to execution time is worth taking in preference to not being able
4478  * to estimate the value at all.
4479  */
4480  if (funcform->provolatile == PROVOLATILE_IMMUTABLE)
4481  /* okay */ ;
4482  else if (context->estimate && funcform->provolatile == PROVOLATILE_STABLE)
4483  /* okay */ ;
4484  else
4485  return NULL;
4486 
4487  /*
4488  * OK, looks like we can simplify this operator/function.
4489  *
4490  * Build a new FuncExpr node containing the already-simplified arguments.
4491  */
4492  newexpr = makeNode(FuncExpr);
4493  newexpr->funcid = funcid;
4494  newexpr->funcresulttype = result_type;
4495  newexpr->funcretset = false;
4496  newexpr->funcvariadic = funcvariadic;
4497  newexpr->funcformat = COERCE_EXPLICIT_CALL; /* doesn't matter */
4498  newexpr->funccollid = result_collid; /* doesn't matter */
4499  newexpr->inputcollid = input_collid;
4500  newexpr->args = args;
4501  newexpr->location = -1;
4502 
4503  return evaluate_expr((Expr *) newexpr, result_type, result_typmod,
4504  result_collid);
4505 }
4506 
4507 /*
4508  * inline_function: try to expand a function call inline
4509  *
4510  * If the function is a sufficiently simple SQL-language function
4511  * (just "SELECT expression"), then we can inline it and avoid the rather
4512  * high per-call overhead of SQL functions. Furthermore, this can expose
4513  * opportunities for constant-folding within the function expression.
4514  *
4515  * We have to beware of some special cases however. A directly or
4516  * indirectly recursive function would cause us to recurse forever,
4517  * so we keep track of which functions we are already expanding and
4518  * do not re-expand them. Also, if a parameter is used more than once
4519  * in the SQL-function body, we require it not to contain any volatile
4520  * functions (volatiles might deliver inconsistent answers) nor to be
4521  * unreasonably expensive to evaluate. The expensiveness check not only
4522  * prevents us from doing multiple evaluations of an expensive parameter
4523  * at runtime, but is a safety value to limit growth of an expression due
4524  * to repeated inlining.
4525  *
4526  * We must also beware of changing the volatility or strictness status of
4527  * functions by inlining them.
4528  *
4529  * Also, at the moment we can't inline functions returning RECORD. This
4530  * doesn't work in the general case because it discards information such
4531  * as OUT-parameter declarations.
4532  *
4533  * Also, context-dependent expression nodes in the argument list are trouble.
4534  *
4535  * Returns a simplified expression if successful, or NULL if cannot
4536  * simplify the function.
4537  */
4538 static Expr *
4539 inline_function(Oid funcid, Oid result_type, Oid result_collid,
4540  Oid input_collid, List *args,
4541  bool funcvariadic,
4542  HeapTuple func_tuple,
4544 {
4545  Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4546  char *src;
4547  Datum tmp;
4548  bool isNull;
4549  MemoryContext oldcxt;
4550  MemoryContext mycxt;
4551  inline_error_callback_arg callback_arg;
4552  ErrorContextCallback sqlerrcontext;
4553  FuncExpr *fexpr;
4555  TupleDesc rettupdesc;
4556  ParseState *pstate;
4557  List *raw_parsetree_list;
4558  List *querytree_list;
4559  Query *querytree;
4560  Node *newexpr;
4561  int *usecounts;
4562  ListCell *arg;
4563  int i;
4564 
4565  /*
4566  * Forget it if the function is not SQL-language or has other showstopper
4567  * properties. (The prokind and nargs checks are just paranoia.)
4568  */
4569  if (funcform->prolang != SQLlanguageId ||
4570  funcform->prokind != PROKIND_FUNCTION ||
4571  funcform->prosecdef ||
4572  funcform->proretset ||
4573  funcform->prorettype == RECORDOID ||
4574  !heap_attisnull(func_tuple, Anum_pg_proc_proconfig, NULL) ||
4575  funcform->pronargs != list_length(args))
4576  return NULL;
4577 
4578  /* Check for recursive function, and give up trying to expand if so */
4579  if (list_member_oid(context->active_fns, funcid))
4580  return NULL;
4581 
4582  /* Check permission to call function (fail later, if not) */
4583  if (object_aclcheck(ProcedureRelationId, funcid, GetUserId(), ACL_EXECUTE) != ACLCHECK_OK)
4584  return NULL;
4585 
4586  /* Check whether a plugin wants to hook function entry/exit */
4587  if (FmgrHookIsNeeded(funcid))
4588  return NULL;
4589 
4590  /*
4591  * Make a temporary memory context, so that we don't leak all the stuff
4592  * that parsing might create.
4593  */
4595  "inline_function",
4597  oldcxt = MemoryContextSwitchTo(mycxt);
4598 
4599  /*
4600  * We need a dummy FuncExpr node containing the already-simplified
4601  * arguments. (In some cases we don't really need it, but building it is
4602  * cheap enough that it's not worth contortions to avoid.)
4603  */
4604  fexpr = makeNode(FuncExpr);
4605  fexpr->funcid = funcid;
4606  fexpr->funcresulttype = result_type;
4607  fexpr->funcretset = false;
4608  fexpr->funcvariadic = funcvariadic;
4609  fexpr->funcformat = COERCE_EXPLICIT_CALL; /* doesn't matter */
4610  fexpr->funccollid = result_collid; /* doesn't matter */
4611  fexpr->inputcollid = input_collid;
4612  fexpr->args = args;
4613  fexpr->location = -1;
4614 
4615  /* Fetch the function body */
4616  tmp = SysCacheGetAttrNotNull(PROCOID, func_tuple, Anum_pg_proc_prosrc);
4617  src = TextDatumGetCString(tmp);
4618 
4619  /*
4620  * Setup error traceback support for ereport(). This is so that we can
4621  * finger the function that bad information came from.
4622  */
4623  callback_arg.proname = NameStr(funcform->proname);
4624  callback_arg.prosrc = src;
4625 
4626  sqlerrcontext.callback = sql_inline_error_callback;
4627  sqlerrcontext.arg = (void *) &callback_arg;
4628  sqlerrcontext.previous = error_context_stack;
4629  error_context_stack = &sqlerrcontext;
4630 
4631  /* If we have prosqlbody, pay attention to that not prosrc */
4632  tmp = SysCacheGetAttr(PROCOID,
4633  func_tuple,
4634  Anum_pg_proc_prosqlbody,
4635  &isNull);
4636  if (!isNull)
4637  {
4638  Node *n;
4639  List *query_list;
4640 
4642  if (IsA(n, List))
4643  query_list = linitial_node(List, castNode(List, n));
4644  else
4645  query_list = list_make1(n);
4646  if (list_length(query_list) != 1)
4647  goto fail;
4648  querytree = linitial(query_list);
4649 
4650  /*
4651  * Because we'll insist below that the querytree have an empty rtable
4652  * and no sublinks, it cannot have any relation references that need
4653  * to be locked or rewritten. So we can omit those steps.
4654  */
4655  }
4656  else
4657  {
4658  /* Set up to handle parameters while parsing the function body. */
4659  pinfo = prepare_sql_fn_parse_info(func_tuple,
4660  (Node *) fexpr,
4661  input_collid);
4662 
4663  /*
4664  * We just do parsing and parse analysis, not rewriting, because
4665  * rewriting will not affect table-free-SELECT-only queries, which is
4666  * all that we care about. Also, we can punt as soon as we detect
4667  * more than one command in the function body.
4668  */
4669  raw_parsetree_list = pg_parse_query(src);
4670  if (list_length(raw_parsetree_list) != 1)
4671  goto fail;
4672 
4673  pstate = make_parsestate(NULL);
4674  pstate->p_sourcetext = src;
4675  sql_fn_parser_setup(pstate, pinfo);
4676 
4677  querytree = transformTopLevelStmt(pstate, linitial(raw_parsetree_list));
4678 
4679  free_parsestate(pstate);
4680  }
4681 
4682  /*
4683  * The single command must be a simple "SELECT expression".
4684  *
4685  * Note: if you change the tests involved in this, see also plpgsql's
4686  * exec_simple_check_plan(). That generally needs to have the same idea
4687  * of what's a "simple expression", so that inlining a function that
4688  * previously wasn't inlined won't change plpgsql's conclusion.
4689  */
4690  if (!IsA(querytree, Query) ||
4691  querytree->commandType != CMD_SELECT ||
4692  querytree->hasAggs ||
4693  querytree->hasWindowFuncs ||
4694  querytree->hasTargetSRFs ||
4695  querytree->hasSubLinks ||
4696  querytree->cteList ||
4697  querytree->rtable ||
4698  querytree->jointree->fromlist ||
4699  querytree->jointree->quals ||
4700  querytree->groupClause ||
4701  querytree->groupingSets ||
4702  querytree->havingQual ||
4703  querytree->windowClause ||
4704  querytree->distinctClause ||
4705  querytree->sortClause ||
4706  querytree->limitOffset ||
4707  querytree->limitCount ||
4708  querytree->setOperations ||
4709  list_length(querytree->targetList) != 1)
4710  goto fail;
4711 
4712  /* If the function result is composite, resolve it */
4713  (void) get_expr_result_type((Node *) fexpr,
4714  NULL,
4715  &rettupdesc);
4716 
4717  /*
4718  * Make sure the function (still) returns what it's declared to. This
4719  * will raise an error if wrong, but that's okay since the function would
4720  * fail at runtime anyway. Note that check_sql_fn_retval will also insert
4721  * a coercion if needed to make the tlist expression match the declared
4722  * type of the function.
4723  *
4724  * Note: we do not try this until we have verified that no rewriting was
4725  * needed; that's probably not important, but let's be careful.
4726  */
4727  querytree_list = list_make1(querytree);
4728  if (check_sql_fn_retval(list_make1(querytree_list),
4729  result_type, rettupdesc,
4730  funcform->prokind,
4731  false, NULL))
4732  goto fail; /* reject whole-tuple-result cases */
4733 
4734  /*
4735  * Given the tests above, check_sql_fn_retval shouldn't have decided to
4736  * inject a projection step, but let's just make sure.
4737  */
4738  if (querytree != linitial(querytree_list))
4739  goto fail;
4740 
4741  /* Now we can grab the tlist expression */
4742  newexpr = (Node *) ((TargetEntry *) linitial(querytree->targetList))->expr;
4743 
4744  /*
4745  * If the SQL function returns VOID, we can only inline it if it is a
4746  * SELECT of an expression returning VOID (ie, it's just a redirection to
4747  * another VOID-returning function). In all non-VOID-returning cases,
4748  * check_sql_fn_retval should ensure that newexpr returns the function's
4749  * declared result type, so this test shouldn't fail otherwise; but we may
4750  * as well cope gracefully if it does.
4751  */
4752  if (exprType(newexpr) != result_type)
4753  goto fail;
4754 
4755  /*
4756  * Additional validity checks on the expression. It mustn't be more
4757  * volatile than the surrounding function (this is to avoid breaking hacks
4758  * that involve pretending a function is immutable when it really ain't).
4759  * If the surrounding function is declared strict, then the expression
4760  * must contain only strict constructs and must use all of the function
4761  * parameters (this is overkill, but an exact analysis is hard).
4762  */
4763  if (funcform->provolatile == PROVOLATILE_IMMUTABLE &&
4764  contain_mutable_functions(newexpr))
4765  goto fail;
4766  else if (funcform->provolatile == PROVOLATILE_STABLE &&
4767  contain_volatile_functions(newexpr))
4768  goto fail;
4769 
4770  if (funcform->proisstrict &&
4771  contain_nonstrict_functions(newexpr))
4772  goto fail;
4773 
4774  /*
4775  * If any parameter expression contains a context-dependent node, we can't
4776  * inline, for fear of putting such a node into the wrong context.
4777  */
4779  goto fail;
4780 
4781  /*
4782  * We may be able to do it; there are still checks on parameter usage to
4783  * make, but those are most easily done in combination with the actual
4784  * substitution of the inputs. So start building expression with inputs
4785  * substituted.
4786  */
4787  usecounts = (int *) palloc0(funcform->pronargs * sizeof(int));
4788  newexpr = substitute_actual_parameters(newexpr, funcform->pronargs,
4789  args, usecounts);
4790 
4791  /* Now check for parameter usage */
4792  i = 0;
4793  foreach(arg, args)
4794  {
4795  Node *param = lfirst(arg);
4796 
4797  if (usecounts[i] == 0)
4798  {
4799  /* Param not used at all: uncool if func is strict */
4800  if (funcform->proisstrict)
4801  goto fail;
4802  }
4803  else if (usecounts[i] != 1)
4804  {
4805  /* Param used multiple times: uncool if expensive or volatile */
4806  QualCost eval_cost;
4807 
4808  /*
4809  * We define "expensive" as "contains any subplan or more than 10
4810  * operators". Note that the subplan search has to be done
4811  * explicitly, since cost_qual_eval() will barf on unplanned
4812  * subselects.
4813  */
4814  if (contain_subplans(param))
4815  goto fail;
4816  cost_qual_eval(&eval_cost, list_make1(param), NULL);
4817  if (eval_cost.startup + eval_cost.per_tuple >
4818  10 * cpu_operator_cost)
4819  goto fail;
4820 
4821  /*
4822  * Check volatility last since this is more expensive than the
4823  * above tests
4824  */
4825  if (contain_volatile_functions(param))
4826  goto fail;
4827  }
4828  i++;
4829  }
4830 
4831  /*
4832  * Whew --- we can make the substitution. Copy the modified expression
4833  * out of the temporary memory context, and clean up.
4834  */
4835  MemoryContextSwitchTo(oldcxt);
4836 
4837  newexpr = copyObject(newexpr);
4838 
4839  MemoryContextDelete(mycxt);
4840 
4841  /*
4842  * If the result is of a collatable type, force the result to expose the
4843  * correct collation. In most cases this does not matter, but it's
4844  * possible that the function result is used directly as a sort key or in
4845  * other places where we expect exprCollation() to tell the truth.
4846  */
4847  if (OidIsValid(result_collid))
4848  {
4849  Oid exprcoll = exprCollation(newexpr);
4850 
4851  if (OidIsValid(exprcoll) && exprcoll != result_collid)
4852  {
4853  CollateExpr *newnode = makeNode(CollateExpr);
4854 
4855  newnode->arg = (Expr *) newexpr;
4856  newnode->collOid = result_collid;
4857  newnode->location = -1;
4858 
4859  newexpr = (Node *) newnode;
4860  }
4861  }
4862 
4863  /*
4864  * Since there is now no trace of the function in the plan tree, we must
4865  * explicitly record the plan's dependency on the function.
4866  */
4867  if (context->root)
4869 
4870  /*
4871  * Recursively try to simplify the modified expression. Here we must add
4872  * the current function to the context list of active functions.
4873  */
4874  context->active_fns = lappend_oid(context->active_fns, funcid);
4875  newexpr = eval_const_expressions_mutator(newexpr, context);
4876  context->active_fns = list_delete_last(context->active_fns);
4877 
4878  error_context_stack = sqlerrcontext.previous;
4879 
4880  return (Expr *) newexpr;
4881 
4882  /* Here if func is not inlinable: release temp memory and return NULL */
4883 fail:
4884  MemoryContextSwitchTo(oldcxt);
4885  MemoryContextDelete(mycxt);
4886  error_context_stack = sqlerrcontext.previous;
4887 
4888  return NULL;
4889 }
4890 
4891 /*
4892  * Replace Param nodes by appropriate actual parameters
4893  */
4894 static Node *
4896  int *usecounts)
4897 {
4899 
4900  context.nargs = nargs;
4901  context.args = args;
4902  context.usecounts = usecounts;
4903 
4905 }
4906 
4907 static Node *
4910 {
4911  if (node == NULL)
4912  return NULL;
4913  if (IsA(node, Param))
4914  {
4915  Param *param = (Param *) node;
4916 
4917  if (param->paramkind != PARAM_EXTERN)
4918  elog(ERROR, "unexpected paramkind: %d", (int) param->paramkind);
4919  if (param->paramid <= 0 || param->paramid > context->nargs)
4920  elog(ERROR, "invalid paramid: %d", param->paramid);
4921 
4922  /* Count usage of parameter */
4923  context->usecounts[param->paramid - 1]++;
4924 
4925  /* Select the appropriate actual arg and replace the Param with it */
4926  /* We don't need to copy at this time (it'll get done later) */
4927  return list_nth(context->args, param->paramid - 1);
4928  }
4930  (void *) context);
4931 }
4932 
4933 /*
4934  * error context callback to let us supply a call-stack traceback
4935  */
4936 static void
4938 {
4940  int syntaxerrposition;
4941 
4942  /* If it's a syntax error, convert to internal syntax error report */
4943  syntaxerrposition = geterrposition();
4944  if (syntaxerrposition > 0)
4945  {
4946  errposition(0);
4947  internalerrposition(syntaxerrposition);
4948  internalerrquery(callback_arg->prosrc);
4949  }
4950 
4951  errcontext("SQL function \"%s\" during inlining", callback_arg->proname);
4952 }
4953 
4954 /*
4955  * evaluate_expr: pre-evaluate a constant expression
4956  *
4957  * We use the executor's routine ExecEvalExpr() to avoid duplication of
4958  * code and ensure we get the same result as the executor would get.
4959  */
4960 Expr *
4961 evaluate_expr(Expr *expr, Oid result_type, int32 result_typmod,
4962  Oid result_collation)
4963 {
4964  EState *estate;
4965  ExprState *exprstate;
4966  MemoryContext oldcontext;
4967  Datum const_val;
4968  bool const_is_null;
4969  int16 resultTypLen;
4970  bool resultTypByVal;
4971 
4972  /*
4973  * To use the executor, we need an EState.
4974  */
4975  estate = CreateExecutorState();
4976 
4977  /* We can use the estate's working context to avoid memory leaks. */
4978  oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
4979 
4980  /* Make sure any opfuncids are filled in. */
4981  fix_opfuncids((Node *) expr);
4982 
4983  /*
4984  * Prepare expr for execution. (Note: we can't use ExecPrepareExpr
4985  * because it'd result in recursively invoking eval_const_expressions.)
4986  */
4987  exprstate = ExecInitExpr(expr, NULL);
4988 
4989  /*
4990  * And evaluate it.
4991  *
4992  * It is OK to use a default econtext because none of the ExecEvalExpr()
4993  * code used in this situation will use econtext. That might seem
4994  * fortuitous, but it's not so unreasonable --- a constant expression does
4995  * not depend on context, by definition, n'est ce pas?
4996  */
4997  const_val = ExecEvalExprSwitchContext(exprstate,
4998  GetPerTupleExprContext(estate),
4999  &const_is_null);
5000 
5001  /* Get info needed about result datatype */
5002  get_typlenbyval(result_type, &resultTypLen, &resultTypByVal);
5003 
5004  /* Get back to outer memory context */
5005  MemoryContextSwitchTo(oldcontext);
5006 
5007  /*
5008  * Must copy result out of sub-context used by expression eval.
5009  *
5010  * Also, if it's varlena, forcibly detoast it. This protects us against
5011  * storing TOAST pointers into plans that might outlive the referenced
5012  * data. (makeConst would handle detoasting anyway, but it's worth a few
5013  * extra lines here so that we can do the copy and detoast in one step.)
5014  */
5015  if (!const_is_null)
5016  {
5017  if (resultTypLen == -1)
5018  const_val = PointerGetDatum(PG_DETOAST_DATUM_COPY(const_val));
5019  else
5020  const_val = datumCopy(const_val, resultTypByVal, resultTypLen);
5021  }
5022 
5023  /* Release all the junk we just created */
5024  FreeExecutorState(estate);
5025 
5026  /*
5027  * Make the constant result node.
5028  */
5029  return (Expr *) makeConst(result_type, result_typmod, result_collation,
5030  resultTypLen,
5031  const_val, const_is_null,
5032  resultTypByVal);
5033 }
5034 
5035 
5036 /*
5037  * inline_set_returning_function
5038  * Attempt to "inline" a set-returning function in the FROM clause.
5039  *
5040  * "rte" is an RTE_FUNCTION rangetable entry. If it represents a call of a
5041  * set-returning SQL function that can safely be inlined, expand the function
5042  * and return the substitute Query structure. Otherwise, return NULL.
5043  *
5044  * We assume that the RTE's expression has already been put through
5045  * eval_const_expressions(), which among other things will take care of
5046  * default arguments and named-argument notation.
5047  *
5048  * This has a good deal of similarity to inline_function(), but that's
5049  * for the non-set-returning case, and there are enough differences to
5050  * justify separate functions.
5051  */
5052 Query *
5054 {
5055  RangeTblFunction *rtfunc;
5056  FuncExpr *fexpr;
5057  Oid func_oid;
5058  HeapTuple func_tuple;
5059  Form_pg_proc funcform;
5060  char *src;
5061  Datum tmp;
5062  bool isNull;
5063  MemoryContext oldcxt;
5064  MemoryContext mycxt;
5065  inline_error_callback_arg callback_arg;
5066  ErrorContextCallback sqlerrcontext;
5068  TypeFuncClass functypclass;
5069  TupleDesc rettupdesc;
5070  List *raw_parsetree_list;
5071  List *querytree_list;
5072  Query *querytree;
5073 
5074  Assert(rte->rtekind == RTE_FUNCTION);
5075 
5076  /*
5077  * It doesn't make a lot of sense for a SQL SRF to refer to itself in its
5078  * own FROM clause, since that must cause infinite recursion at runtime.
5079  * It will cause this code to recurse too, so check for stack overflow.
5080  * (There's no need to do more.)
5081  */
5083 
5084  /* Fail if the RTE has ORDINALITY - we don't implement that here. */
5085  if (rte->funcordinality)
5086  return NULL;
5087 
5088  /* Fail if RTE isn't a single, simple FuncExpr */
5089  if (list_length(rte->functions) != 1)
5090  return NULL;
5091  rtfunc = (RangeTblFunction *) linitial(rte->functions);
5092 
5093  if (!IsA(rtfunc->funcexpr, FuncExpr))
5094  return NULL;
5095  fexpr = (FuncExpr *) rtfunc->funcexpr;
5096 
5097  func_oid = fexpr->funcid;
5098 
5099  /*
5100  * The function must be declared to return a set, else inlining would
5101  * change the results if the contained SELECT didn't return exactly one
5102  * row.
5103  */
5104  if (!fexpr->funcretset)
5105  return NULL;
5106 
5107  /*
5108  * Refuse to inline if the arguments contain any volatile functions or
5109  * sub-selects. Volatile functions are rejected because inlining may
5110  * result in the arguments being evaluated multiple times, risking a
5111  * change in behavior. Sub-selects are rejected partly for implementation
5112  * reasons (pushing them down another level might change their behavior)
5113  * and partly because they're likely to be expensive and so multiple
5114  * evaluation would be bad.
5115  */
5116  if (contain_volatile_functions((Node *) fexpr->args) ||
5117  contain_subplans((Node *) fexpr->args))
5118  return NULL;
5119 
5120  /* Check permission to call function (fail later, if not) */
5121  if (object_aclcheck(ProcedureRelationId, func_oid, GetUserId(), ACL_EXECUTE) != ACLCHECK_OK)
5122  return NULL;
5123 
5124  /* Check whether a plugin wants to hook function entry/exit */
5125  if (FmgrHookIsNeeded(func_oid))
5126  return NULL;
5127 
5128  /*
5129  * OK, let's take a look at the function's pg_proc entry.
5130  */
5131  func_tuple = SearchSysCache1(PROCOID, ObjectIdGetDatum(func_oid));
5132  if (!HeapTupleIsValid(func_tuple))
5133  elog(ERROR, "cache lookup failed for function %u", func_oid);
5134  funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
5135 
5136  /*
5137  * Forget it if the function is not SQL-language or has other showstopper
5138  * properties. In particular it mustn't be declared STRICT, since we
5139  * couldn't enforce that. It also mustn't be VOLATILE, because that is
5140  * supposed to cause it to be executed with its own snapshot, rather than
5141  * sharing the snapshot of the calling query. We also disallow returning
5142  * SETOF VOID, because inlining would result in exposing the actual result
5143  * of the function's last SELECT, which should not happen in that case.
5144  * (Rechecking prokind, proretset, and pronargs is just paranoia.)
5145  */
5146  if (funcform->prolang != SQLlanguageId ||
5147  funcform->prokind != PROKIND_FUNCTION ||
5148  funcform->proisstrict ||
5149  funcform->provolatile == PROVOLATILE_VOLATILE ||
5150  funcform->prorettype == VOIDOID ||
5151  funcform->prosecdef ||
5152  !funcform->proretset ||
5153  list_length(fexpr->args) != funcform->pronargs ||
5154  !heap_attisnull(func_tuple, Anum_pg_proc_proconfig, NULL))
5155  {
5156  ReleaseSysCache(func_tuple);
5157  return NULL;
5158  }
5159 
5160  /*
5161  * Make a temporary memory context, so that we don't leak all the stuff
5162  * that parsing might create.
5163  */
5165  "inline_set_returning_function",
5167  oldcxt = MemoryContextSwitchTo(mycxt);
5168 
5169  /* Fetch the function body */
5170  tmp = SysCacheGetAttrNotNull(PROCOID, func_tuple, Anum_pg_proc_prosrc);
5171  src = TextDatumGetCString(tmp);
5172 
5173  /*
5174  * Setup error traceback support for ereport(). This is so that we can
5175  * finger the function that bad information came from.
5176  */
5177  callback_arg.proname = NameStr(funcform->proname);
5178  callback_arg.prosrc = src;
5179 
5180  sqlerrcontext.callback = sql_inline_error_callback;
5181  sqlerrcontext.arg = (void *) &callback_arg;
5182  sqlerrcontext.previous = error_context_stack;
5183  error_context_stack = &sqlerrcontext;
5184 
5185  /* If we have prosqlbody, pay attention to that not prosrc */
5186  tmp = SysCacheGetAttr(PROCOID,
5187  func_tuple,
5188  Anum_pg_proc_prosqlbody,
5189  &isNull);
5190  if (!isNull)
5191  {
5192  Node *n;
5193 
5195  if (IsA(n, List))
5196  querytree_list = linitial_node(List, castNode(List, n));
5197  else
5198  querytree_list = list_make1(n);
5199  if (list_length(querytree_list) != 1)
5200  goto fail;
5201  querytree = linitial(querytree_list);
5202 
5203  /* Acquire necessary locks, then apply rewriter. */
5204  AcquireRewriteLocks(querytree, true, false);
5205  querytree_list = pg_rewrite_query(querytree);
5206  if (list_length(querytree_list) != 1)
5207  goto fail;
5208  querytree = linitial(querytree_list);
5209  }
5210  else
5211  {
5212  /*
5213  * Set up to handle parameters while parsing the function body. We
5214  * can use the FuncExpr just created as the input for
5215  * prepare_sql_fn_parse_info.
5216  */
5217  pinfo = prepare_sql_fn_parse_info(func_tuple,
5218  (Node *) fexpr,
5219  fexpr->inputcollid);
5220 
5221  /*
5222  * Parse, analyze, and rewrite (unlike inline_function(), we can't
5223  * skip rewriting here). We can fail as soon as we find more than one
5224  * query, though.
5225  */
5226  raw_parsetree_list = pg_parse_query(src);
5227  if (list_length(raw_parsetree_list) != 1)
5228  goto fail;
5229 
5230  querytree_list = pg_analyze_and_rewrite_withcb(linitial(raw_parsetree_list),
5231  src,
5233  pinfo, NULL);
5234  if (list_length(querytree_list) != 1)
5235  goto fail;
5236  querytree = linitial(querytree_list);
5237  }
5238 
5239  /*
5240  * Also resolve the actual function result tupdesc, if composite. If we
5241  * have a coldeflist, believe that; otherwise use get_expr_result_type.
5242  * (This logic should match ExecInitFunctionScan.)
5243  */
5244  if (rtfunc->funccolnames != NIL)
5245  {
5246  functypclass = TYPEFUNC_RECORD;
5247  rettupdesc = BuildDescFromLists(rtfunc->funccolnames,
5248  rtfunc->funccoltypes,
5249  rtfunc->funccoltypmods,
5250  rtfunc->funccolcollations);
5251  }
5252  else
5253  functypclass = get_expr_result_type((Node *) fexpr, NULL, &rettupdesc);
5254 
5255  /*
5256  * The single command must be a plain SELECT.
5257  */
5258  if (!IsA(querytree, Query) ||
5259  querytree->commandType != CMD_SELECT)
5260  goto fail;
5261 
5262  /*
5263  * Make sure the function (still) returns what it's declared to. This
5264  * will raise an error if wrong, but that's okay since the function would
5265  * fail at runtime anyway. Note that check_sql_fn_retval will also insert
5266  * coercions if needed to make the tlist expression(s) match the declared
5267  * type of the function. We also ask it to insert dummy NULL columns for
5268  * any dropped columns in rettupdesc, so that the elements of the modified
5269  * tlist match up to the attribute numbers.
5270  *
5271  * If the function returns a composite type, don't inline unless the check
5272  * shows it's returning a whole tuple result; otherwise what it's
5273  * returning is a single composite column which is not what we need.
5274  */
5275  if (!check_sql_fn_retval(list_make1(querytree_list),
5276  fexpr->funcresulttype, rettupdesc,
5277  funcform->prokind,
5278  true, NULL) &&
5279  (functypclass == TYPEFUNC_COMPOSITE ||
5280  functypclass == TYPEFUNC_COMPOSITE_DOMAIN ||
5281  functypclass == TYPEFUNC_RECORD))
5282  goto fail; /* reject not-whole-tuple-result cases */
5283 
5284  /*
5285  * check_sql_fn_retval might've inserted a projection step, but that's
5286  * fine; just make sure we use the upper Query.
5287  */
5288  querytree = linitial_node(Query, querytree_list);
5289 
5290  /*
5291  * Looks good --- substitute parameters into the query.
5292  */
5294  funcform->pronargs,
5295  fexpr->args);
5296 
5297  /*
5298  * Copy the modified query out of the temporary memory context, and clean
5299  * up.
5300  */
5301  MemoryContextSwitchTo(oldcxt);
5302 
5304 
5305  MemoryContextDelete(mycxt);
5306  error_context_stack = sqlerrcontext.previous;
5307  ReleaseSysCache(func_tuple);
5308 
5309  /*
5310  * We don't have to fix collations here because the upper query is already
5311  * parsed, ie, the collations in the RTE are what count.
5312  */
5313 
5314  /*
5315  * Since there is now no trace of the function in the plan tree, we must
5316  * explicitly record the plan's dependency on the function.
5317  */
5319 
5320  /*
5321  * We must also notice if the inserted query adds a dependency on the
5322  * calling role due to RLS quals.
5323  */
5324  if (querytree->hasRowSecurity)
5325  root->glob->dependsOnRole = true;
5326 
5327  return querytree;
5328 
5329  /* Here if func is not inlinable: release temp memory and return NULL */
5330 fail:
5331  MemoryContextSwitchTo(oldcxt);
5332  MemoryContextDelete(mycxt);
5333  error_context_stack = sqlerrcontext.previous;
5334  ReleaseSysCache(func_tuple);
5335 
5336  return NULL;
5337 }
5338 
5339 /*
5340  * Replace Param nodes by appropriate actual parameters
5341  *
5342  * This is just enough different from substitute_actual_parameters()
5343  * that it needs its own code.
5344  */
5345 static Query *
5347 {
5349 
5350  context.nargs = nargs;
5351  context.args = args;
5352  context.sublevels_up = 1;
5353 
5354  return query_tree_mutator(expr,
5356  &context,
5357  0);
5358 }
5359 
5360 static Node *
5363 {
5364  Node *result;
5365 
5366  if (node == NULL)
5367  return NULL;
5368  if (IsA(node, Query))
5369  {
5370  context->sublevels_up++;
5371  result = (Node *) query_tree_mutator((Query *) node,
5373  (void *) context,
5374  0);
5375  context->sublevels_up--;
5376  return result;
5377  }
5378  if (IsA(node, Param))
5379  {
5380  Param *param = (Param *) node;
5381 
5382  if (param->paramkind == PARAM_EXTERN)
5383  {
5384  if (param->paramid <= 0 || param->paramid > context->nargs)
5385  elog(ERROR, "invalid paramid: %d", param->paramid);
5386 
5387  /*
5388  * Since the parameter is being inserted into a subquery, we must
5389  * adjust levels.
5390  */
5391  result = copyObject(list_nth(context->args, param->paramid - 1));
5392  IncrementVarSublevelsUp(result, context->sublevels_up, 0);
5393  return result;
5394  }
5395  }
5396  return expression_tree_mutator(node,
5398  (void *) context);
5399 }
5400 
5401 /*
5402  * pull_paramids
5403  * Returns a Bitmapset containing the paramids of all Params in 'expr'.
5404  */
5405 Bitmapset *
5407 {
5408  Bitmapset *result = NULL;
5409 
5410  (void) pull_paramids_walker((Node *) expr, &result);
5411 
5412  return result;
5413 }
5414 
5415 static bool
5417 {
5418  if (node == NULL)
5419  return false;
5420  if (IsA(node, Param))
5421  {
5422  Param *param = (Param *) node;
5423 
5424  *context = bms_add_member(*context, param->paramid);
5425  return false;
5426  }
5428  (void *) context);
5429 }
Datum querytree(PG_FUNCTION_ARGS)
Definition: _int_bool.c:666
@ ACLCHECK_OK
Definition: acl.h:183
AclResult object_aclcheck(Oid classid, Oid objectid, Oid roleid, AclMode mode)
Definition: aclchk.c:3890
#define ARR_NDIM(a)
Definition: array.h:290
#define ARR_DATA_PTR(a)
Definition: array.h:322
#define DatumGetArrayTypeP(X)
Definition: array.h:261
#define ARR_ELEMTYPE(a)
Definition: array.h:292
#define ARR_DIMS(a)
Definition: array.h:294
#define ARR_HASNULL(a)
Definition: array.h:291
int ArrayGetNItems(int ndim, const int *dims)
Definition: arrayutils.c:57
#define InvalidAttrNumber
Definition: attnum.h:23
Bitmapset * bms_join(Bitmapset *a, Bitmapset *b)
Definition: bitmapset.c:1230
void bms_free(Bitmapset *a)
Definition: bitmapset.c:239
int bms_num_members(const Bitmapset *a)
Definition: bitmapset.c:751
Bitmapset * bms_make_singleton(int x)
Definition: bitmapset.c:216
Bitmapset * bms_add_member(Bitmapset *a, int x)
Definition: bitmapset.c:815
Bitmapset * bms_add_members(Bitmapset *a, const Bitmapset *b)
Definition: bitmapset.c:917
Bitmapset * bms_del_members(Bitmapset *a, const Bitmapset *b)
Definition: bitmapset.c:1161
Bitmapset * bms_int_members(Bitmapset *a, const Bitmapset *b)
Definition: bitmapset.c:1109
BMS_Membership bms_membership(const Bitmapset *a)
Definition: bitmapset.c:781
#define bms_is_empty(a)
Definition: bitmapset.h:118
@ BMS_SINGLETON
Definition: bitmapset.h:72
#define TextDatumGetCString(d)
Definition: builtins.h:98
#define NameStr(name)
Definition: c.h:746
signed short int16
Definition: c.h:493
signed int int32
Definition: c.h:494
#define Assert(condition)
Definition: c.h:858
unsigned int Index
Definition: c.h:614
#define OidIsValid(objectId)
Definition: c.h:775
static bool contain_subplans_walker(Node *node, void *context)
Definition: clauses.c:336
#define CCDN_CASETESTEXPR_OK
Definition: clauses.c:1186
static List * simplify_or_arguments(List *args, eval_const_expressions_context *context, bool *haveNull, bool *forceTrue)
Definition: clauses.c:3778
static bool is_strict_saop(ScalarArrayOpExpr *expr, bool falseOK)
Definition: clauses.c:2026
bool contain_volatile_functions_not_nextval(Node *clause)
Definition: clauses.c:673
static bool contain_leaked_vars_checker(Oid func_id, void *context)
Definition: clauses.c:1269
static bool rowtype_field_matches(Oid rowtypeid, int fieldnum, Oid expectedtype, int32 expectedtypmod, Oid expectedcollation)
Definition: clauses.c:2186
WindowFuncLists * find_window_functions(Node *clause, Index maxWinRef)
Definition: clauses.c:227
static bool contain_nonstrict_functions_walker(Node *node, void *context)
Definition: clauses.c:1005
static List * add_function_defaults(List *args, int pronargs, HeapTuple func_tuple)
Definition: clauses.c:4314
#define ece_all_arguments_const(node)
Definition: clauses.c:2426
Expr * evaluate_expr(Expr *expr, Oid result_type, int32 result_typmod, Oid result_collation)
Definition: clauses.c:4961
#define ece_evaluate_expr(node)
Definition: clauses.c:2430
static bool max_parallel_hazard_checker(Oid func_id, void *context)
Definition: clauses.c:822
static bool max_parallel_hazard_test(char proparallel, max_parallel_hazard_context *context)
Definition: clauses.c:794
bool contain_agg_clause(Node *clause)
Definition: clauses.c:177
static Node * substitute_actual_srf_parameters_mutator(Node *node, substitute_actual_srf_parameters_context *context)
Definition: clauses.c:5361
static bool contain_agg_clause_walker(Node *node, void *context)
Definition: clauses.c:183
static bool contain_nonstrict_functions_checker(Oid func_id, void *context)
Definition: clauses.c:999
int NumRelids(PlannerInfo *root, Node *clause)
Definition: clauses.c:2130
bool contain_mutable_functions(Node *clause)
Definition: clauses.c:370
bool is_pseudo_constant_clause(Node *clause)
Definition: clauses.c:2088
static bool max_parallel_hazard_walker(Node *node, max_parallel_hazard_context *context)
Definition: clauses.c:829
bool contain_window_function(Node *clause)
Definition: clauses.c:214
#define ece_generic_processing(node)
Definition: clauses.c:2417
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)
Definition: clauses.c:4413
static Node * substitute_actual_parameters_mutator(Node *node, substitute_actual_parameters_context *context)
Definition: clauses.c:4908
static bool contain_mutable_functions_checker(Oid func_id, void *context)
Definition: clauses.c:376
bool is_pseudo_constant_clause_relids(Node *clause, Relids relids)
Definition: clauses.c:2108
List * expand_function_arguments(List *args, bool include_out_arguments, Oid result_type, HeapTuple func_tuple)
Definition: clauses.c:4163
static Query * substitute_actual_srf_parameters(Query *expr, int nargs, List *args)
Definition: clauses.c:5346
Var * find_forced_null_var(Node *node)
Definition: clauses.c:1977
static bool ece_function_is_safe(Oid funcid, eval_const_expressions_context *context)
Definition: clauses.c:3740
static bool contain_volatile_functions_checker(Oid func_id, void *context)
Definition: clauses.c:544
static List * simplify_and_arguments(List *args, eval_const_expressions_context *context, bool *haveNull, bool *forceFalse)
Definition: clauses.c:3884
List * find_forced_null_vars(Node *node)
Definition: clauses.c:1916
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)
Definition: clauses.c:4047
bool contain_volatile_functions_after_planning(Expr *expr)
Definition: clauses.c:659
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)
Definition: clauses.c:4539
static List * reorder_function_arguments(List *args, int pronargs, HeapTuple func_tuple)
Definition: clauses.c:4244
static Node * substitute_actual_parameters(Node *expr, int nargs, List *args, int *usecounts)
Definition: clauses.c:4895
static bool contain_mutable_functions_walker(Node *node, void *context)
Definition: clauses.c:382
Query * inline_set_returning_function(PlannerInfo *root, RangeTblEntry *rte)
Definition: clauses.c:5053
bool contain_mutable_functions_after_planning(Expr *expr)
Definition: clauses.c:490
static bool contain_volatile_functions_walker(Node *node, void *context)
Definition: clauses.c:550
bool contain_leaked_vars(Node *clause)
Definition: clauses.c:1263
static Relids find_nonnullable_rels_walker(Node *node, bool top_level)
Definition: clauses.c:1462
Bitmapset * pull_paramids(Expr *expr)
Definition: clauses.c:5406
void convert_saop_to_hashed_saop(Node *node)
Definition: clauses.c:2287
static void sql_inline_error_callback(void *arg)
Definition: clauses.c:4937
static bool contain_volatile_functions_not_nextval_walker(Node *node, void *context)
Definition: clauses.c:686
static bool contain_leaked_vars_walker(Node *node, void *context)
Definition: clauses.c:1275
static bool contain_non_const_walker(Node *node, void *context)
Definition: clauses.c:3724
static bool contain_context_dependent_node(Node *clause)
Definition: clauses.c:1179
Relids find_nonnullable_rels(Node *clause)
Definition: clauses.c:1456
static void recheck_cast_function_args(List *args, Oid result_type, Oid *proargtypes, int pronargs, HeapTuple func_tuple)
Definition: clauses.c:4368
static bool find_window_functions_walker(Node *node, WindowFuncLists *lists)
Definition: clauses.c:239
List * find_nonnullable_vars(Node *clause)
Definition: clauses.c:1707
char max_parallel_hazard(Query *parse)
Definition: clauses.c:734
bool is_parallel_safe(PlannerInfo *root, Node *node)
Definition: clauses.c:753
bool contain_nonstrict_functions(Node *clause)
Definition: clauses.c:993
static bool contain_volatile_functions_not_nextval_checker(Oid func_id, void *context)
Definition: clauses.c:679
static List * find_nonnullable_vars_walker(Node *node, bool top_level)
Definition: clauses.c:1713
Node * eval_const_expressions(PlannerInfo *root, Node *node)
Definition: clauses.c:2254
bool contain_subplans(Node *clause)
Definition: clauses.c:330
static Node * simplify_boolean_equality(Oid opno, List *args)
Definition: clauses.c:3978
static bool contain_exec_param_walker(Node *node, List *param_ids)
Definition: clauses.c:1143
void CommuteOpExpr(OpExpr *clause)
Definition: clauses.c:2147
static Node * eval_const_expressions_mutator(Node *node, eval_const_expressions_context *context)
Definition: clauses.c:2440
static bool pull_paramids_walker(Node *node, Bitmapset **context)
Definition: clauses.c:5416
static bool convert_saop_to_hashed_saop_walker(Node *node, void *context)
Definition: clauses.c:2293
static List * fetch_function_defaults(HeapTuple func_tuple)
Definition: clauses.c:4338
bool contain_volatile_functions(Node *clause)
Definition: clauses.c:538
double expression_returns_set_rows(PlannerInfo *root, Node *clause)
Definition: clauses.c:289
static bool contain_context_dependent_node_walker(Node *node, int *flags)
Definition: clauses.c:1189
bool contain_exec_param(Node *clause, List *param_ids)
Definition: clauses.c:1137
#define MIN_ARRAY_SIZE_FOR_HASHED_SAOP
Definition: clauses.c:2269
Node * estimate_expression_value(PlannerInfo *root, Node *node)
Definition: clauses.c:2395
double cpu_operator_cost
Definition: costsize.c:123
void cost_qual_eval(QualCost *cost, List *quals, PlannerInfo *root)
Definition: costsize.c:4640
double clamp_row_est(double nrows)
Definition: costsize.c:202
Datum datumCopy(Datum value, bool typByVal, int typLen)
Definition: datum.c:132
int internalerrquery(const char *query)
Definition: elog.c:1482
int internalerrposition(int cursorpos)
Definition: elog.c:1462
ErrorContextCallback * error_context_stack
Definition: elog.c:94
int geterrposition(void)
Definition: elog.c:1578
int errposition(int cursorpos)
Definition: elog.c:1446
#define errcontext
Definition: elog.h:196
#define ERROR
Definition: elog.h:39
#define elog(elevel,...)
Definition: elog.h:224
ExprState * ExecInitExpr(Expr *node, PlanState *parent)
Definition: execExpr.c:134
EState * CreateExecutorState(void)
Definition: execUtils.c:88
void FreeExecutorState(EState *estate)
Definition: execUtils.c:189
#define GetPerTupleExprContext(estate)
Definition: executor.h:550
static Datum ExecEvalExprSwitchContext(ExprState *state, ExprContext *econtext, bool *isNull)
Definition: executor.h:348
#define OidFunctionCall1(functionId, arg1)
Definition: fmgr.h:680
#define PG_DETOAST_DATUM_COPY(datum)
Definition: fmgr.h:242
#define FmgrHookIsNeeded(fn_oid)
Definition: fmgr.h:797
TypeFuncClass get_expr_result_type(Node *expr, Oid *resultTypeId, TupleDesc *resultTupleDesc)
Definition: funcapi.c:299
TypeFuncClass
Definition: funcapi.h:147
@ TYPEFUNC_COMPOSITE
Definition: funcapi.h:149
@ TYPEFUNC_RECORD
Definition: funcapi.h:151
@ TYPEFUNC_COMPOSITE_DOMAIN
Definition: funcapi.h:150
void sql_fn_parser_setup(struct ParseState *pstate, SQLFunctionParseInfoPtr pinfo)
Definition: functions.c:265
bool check_sql_fn_retval(List *queryTreeLists, Oid rettype, TupleDesc rettupdesc, char prokind, bool insertDroppedCols, List **resultTargetList)
Definition: functions.c:1608