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 }
270 
271 
272 /*****************************************************************************
273  * Support for expressions returning sets
274  *****************************************************************************/
275 
276 /*
277  * expression_returns_set_rows
278  * Estimate the number of rows returned by a set-returning expression.
279  * The result is 1 if it's not a set-returning expression.
280  *
281  * We should only examine the top-level function or operator; it used to be
282  * appropriate to recurse, but not anymore. (Even if there are more SRFs in
283  * the function's inputs, their multipliers are accounted for separately.)
284  *
285  * Note: keep this in sync with expression_returns_set() in nodes/nodeFuncs.c.
286  */
287 double
289 {
290  if (clause == NULL)
291  return 1.0;
292  if (IsA(clause, FuncExpr))
293  {
294  FuncExpr *expr = (FuncExpr *) clause;
295 
296  if (expr->funcretset)
297  return clamp_row_est(get_function_rows(root, expr->funcid, clause));
298  }
299  if (IsA(clause, OpExpr))
300  {
301  OpExpr *expr = (OpExpr *) clause;
302 
303  if (expr->opretset)
304  {
305  set_opfuncid(expr);
306  return clamp_row_est(get_function_rows(root, expr->opfuncid, clause));
307  }
308  }
309  return 1.0;
310 }
311 
312 
313 /*****************************************************************************
314  * Subplan clause manipulation
315  *****************************************************************************/
316 
317 /*
318  * contain_subplans
319  * Recursively search for subplan nodes within a clause.
320  *
321  * If we see a SubLink node, we will return true. This is only possible if
322  * the expression tree hasn't yet been transformed by subselect.c. We do not
323  * know whether the node will produce a true subplan or just an initplan,
324  * but we make the conservative assumption that it will be a subplan.
325  *
326  * Returns true if any subplan found.
327  */
328 bool
330 {
331  return contain_subplans_walker(clause, NULL);
332 }
333 
334 static bool
336 {
337  if (node == NULL)
338  return false;
339  if (IsA(node, SubPlan) ||
340  IsA(node, AlternativeSubPlan) ||
341  IsA(node, SubLink))
342  return true; /* abort the tree traversal and return true */
344 }
345 
346 
347 /*****************************************************************************
348  * Check clauses for mutable functions
349  *****************************************************************************/
350 
351 /*
352  * contain_mutable_functions
353  * Recursively search for mutable functions within a clause.
354  *
355  * Returns true if any mutable function (or operator implemented by a
356  * mutable function) is found. This test is needed so that we don't
357  * mistakenly think that something like "WHERE random() < 0.5" can be treated
358  * as a constant qualification.
359  *
360  * This will give the right answer only for clauses that have been put
361  * through expression preprocessing. Callers outside the planner typically
362  * should use contain_mutable_functions_after_planning() instead, for the
363  * reasons given there.
364  *
365  * We will recursively look into Query nodes (i.e., SubLink sub-selects)
366  * but not into SubPlans. See comments for contain_volatile_functions().
367  */
368 bool
370 {
371  return contain_mutable_functions_walker(clause, NULL);
372 }
373 
374 static bool
376 {
377  return (func_volatile(func_id) != PROVOLATILE_IMMUTABLE);
378 }
379 
380 static bool
382 {
383  if (node == NULL)
384  return false;
385  /* Check for mutable functions in node itself */
387  context))
388  return true;
389 
390  if (IsA(node, JsonConstructorExpr))
391  {
392  const JsonConstructorExpr *ctor = (JsonConstructorExpr *) node;
393  ListCell *lc;
394  bool is_jsonb;
395 
396  is_jsonb = ctor->returning->format->format_type == JS_FORMAT_JSONB;
397 
398  /*
399  * Check argument_type => json[b] conversions specifically. We still
400  * recurse to check 'args' below, but here we want to specifically
401  * check whether or not the emitted clause would fail to be immutable
402  * because of TimeZone, for example.
403  */
404  foreach(lc, ctor->args)
405  {
406  Oid typid = exprType(lfirst(lc));
407 
408  if (is_jsonb ?
409  !to_jsonb_is_immutable(typid) :
410  !to_json_is_immutable(typid))
411  return true;
412  }
413 
414  /* Check all subnodes */
415  }
416 
417  if (IsA(node, JsonExpr))
418  {
419  JsonExpr *jexpr = castNode(JsonExpr, node);
420  Const *cnst;
421 
422  if (!IsA(jexpr->path_spec, Const))
423  return true;
424 
425  cnst = castNode(Const, jexpr->path_spec);
426 
427  Assert(cnst->consttype == JSONPATHOID);
428  if (cnst->constisnull)
429  return false;
430 
431  if (jspIsMutable(DatumGetJsonPathP(cnst->constvalue),
432  jexpr->passing_names, jexpr->passing_values))
433  return true;
434  }
435 
436  if (IsA(node, SQLValueFunction))
437  {
438  /* all variants of SQLValueFunction are stable */
439  return true;
440  }
441 
442  if (IsA(node, NextValueExpr))
443  {
444  /* NextValueExpr is volatile */
445  return true;
446  }
447 
448  /*
449  * It should be safe to treat MinMaxExpr as immutable, because it will
450  * depend on a non-cross-type btree comparison function, and those should
451  * always be immutable. Treating XmlExpr as immutable is more dubious,
452  * and treating CoerceToDomain as immutable is outright dangerous. But we
453  * have done so historically, and changing this would probably cause more
454  * problems than it would fix. In practice, if you have a non-immutable
455  * domain constraint you are in for pain anyhow.
456  */
457 
458  /* Recurse to check arguments */
459  if (IsA(node, Query))
460  {
461  /* Recurse into subselects */
462  return query_tree_walker((Query *) node,
464  context, 0);
465  }
467  context);
468 }
469 
470 /*
471  * contain_mutable_functions_after_planning
472  * Test whether given expression contains mutable functions.
473  *
474  * This is a wrapper for contain_mutable_functions() that is safe to use from
475  * outside the planner. The difference is that it first runs the expression
476  * through expression_planner(). There are two key reasons why we need that:
477  *
478  * First, function default arguments will get inserted, which may affect
479  * volatility (consider "default now()").
480  *
481  * Second, inline-able functions will get inlined, which may allow us to
482  * conclude that the function is really less volatile than it's marked.
483  * As an example, polymorphic functions must be marked with the most volatile
484  * behavior that they have for any input type, but once we inline the
485  * function we may be able to conclude that it's not so volatile for the
486  * particular input type we're dealing with.
487  */
488 bool
490 {
491  /* We assume here that expression_planner() won't scribble on its input */
492  expr = expression_planner(expr);
493 
494  /* Now we can search for non-immutable functions */
495  return contain_mutable_functions((Node *) expr);
496 }
497 
498 
499 /*****************************************************************************
500  * Check clauses for volatile functions
501  *****************************************************************************/
502 
503 /*
504  * contain_volatile_functions
505  * Recursively search for volatile functions within a clause.
506  *
507  * Returns true if any volatile function (or operator implemented by a
508  * volatile function) is found. This test prevents, for example,
509  * invalid conversions of volatile expressions into indexscan quals.
510  *
511  * This will give the right answer only for clauses that have been put
512  * through expression preprocessing. Callers outside the planner typically
513  * should use contain_volatile_functions_after_planning() instead, for the
514  * reasons given there.
515  *
516  * We will recursively look into Query nodes (i.e., SubLink sub-selects)
517  * but not into SubPlans. This is a bit odd, but intentional. If we are
518  * looking at a SubLink, we are probably deciding whether a query tree
519  * transformation is safe, and a contained sub-select should affect that;
520  * for example, duplicating a sub-select containing a volatile function
521  * would be bad. However, once we've got to the stage of having SubPlans,
522  * subsequent planning need not consider volatility within those, since
523  * the executor won't change its evaluation rules for a SubPlan based on
524  * volatility.
525  *
526  * For some node types, for example, RestrictInfo and PathTarget, we cache
527  * whether we found any volatile functions or not and reuse that value in any
528  * future checks for that node. All of the logic for determining if the
529  * cached value should be set to VOLATILITY_NOVOLATILE or VOLATILITY_VOLATILE
530  * belongs in this function. Any code which makes changes to these nodes
531  * which could change the outcome this function must set the cached value back
532  * to VOLATILITY_UNKNOWN. That allows this function to redetermine the
533  * correct value during the next call, should we need to redetermine if the
534  * node contains any volatile functions again in the future.
535  */
536 bool
538 {
539  return contain_volatile_functions_walker(clause, NULL);
540 }
541 
542 static bool
544 {
545  return (func_volatile(func_id) == PROVOLATILE_VOLATILE);
546 }
547 
548 static bool
550 {
551  if (node == NULL)
552  return false;
553  /* Check for volatile functions in node itself */
555  context))
556  return true;
557 
558  if (IsA(node, NextValueExpr))
559  {
560  /* NextValueExpr is volatile */
561  return true;
562  }
563 
564  if (IsA(node, RestrictInfo))
565  {
566  RestrictInfo *rinfo = (RestrictInfo *) node;
567 
568  /*
569  * For RestrictInfo, check if we've checked the volatility of it
570  * before. If so, we can just use the cached value and not bother
571  * checking it again. Otherwise, check it and cache if whether we
572  * found any volatile functions.
573  */
574  if (rinfo->has_volatile == VOLATILITY_NOVOLATILE)
575  return false;
576  else if (rinfo->has_volatile == VOLATILITY_VOLATILE)
577  return true;
578  else
579  {
580  bool hasvolatile;
581 
582  hasvolatile = contain_volatile_functions_walker((Node *) rinfo->clause,
583  context);
584  if (hasvolatile)
585  rinfo->has_volatile = VOLATILITY_VOLATILE;
586  else
587  rinfo->has_volatile = VOLATILITY_NOVOLATILE;
588 
589  return hasvolatile;
590  }
591  }
592 
593  if (IsA(node, PathTarget))
594  {
595  PathTarget *target = (PathTarget *) node;
596 
597  /*
598  * We also do caching for PathTarget the same as we do above for
599  * RestrictInfos.
600  */
602  return false;
603  else if (target->has_volatile_expr == VOLATILITY_VOLATILE)
604  return true;
605  else
606  {
607  bool hasvolatile;
608 
609  hasvolatile = contain_volatile_functions_walker((Node *) target->exprs,
610  context);
611 
612  if (hasvolatile)
614  else
616 
617  return hasvolatile;
618  }
619  }
620 
621  /*
622  * See notes in contain_mutable_functions_walker about why we treat
623  * MinMaxExpr, XmlExpr, and CoerceToDomain as immutable, while
624  * SQLValueFunction is stable. Hence, none of them are of interest here.
625  */
626 
627  /* Recurse to check arguments */
628  if (IsA(node, Query))
629  {
630  /* Recurse into subselects */
631  return query_tree_walker((Query *) node,
633  context, 0);
634  }
636  context);
637 }
638 
639 /*
640  * contain_volatile_functions_after_planning
641  * Test whether given expression contains volatile functions.
642  *
643  * This is a wrapper for contain_volatile_functions() that is safe to use from
644  * outside the planner. The difference is that it first runs the expression
645  * through expression_planner(). There are two key reasons why we need that:
646  *
647  * First, function default arguments will get inserted, which may affect
648  * volatility (consider "default random()").
649  *
650  * Second, inline-able functions will get inlined, which may allow us to
651  * conclude that the function is really less volatile than it's marked.
652  * As an example, polymorphic functions must be marked with the most volatile
653  * behavior that they have for any input type, but once we inline the
654  * function we may be able to conclude that it's not so volatile for the
655  * particular input type we're dealing with.
656  */
657 bool
659 {
660  /* We assume here that expression_planner() won't scribble on its input */
661  expr = expression_planner(expr);
662 
663  /* Now we can search for volatile functions */
664  return contain_volatile_functions((Node *) expr);
665 }
666 
667 /*
668  * Special purpose version of contain_volatile_functions() for use in COPY:
669  * ignore nextval(), but treat all other functions normally.
670  */
671 bool
673 {
675 }
676 
677 static bool
679 {
680  return (func_id != F_NEXTVAL &&
681  func_volatile(func_id) == PROVOLATILE_VOLATILE);
682 }
683 
684 static bool
686 {
687  if (node == NULL)
688  return false;
689  /* Check for volatile functions in node itself */
690  if (check_functions_in_node(node,
692  context))
693  return true;
694 
695  /*
696  * See notes in contain_mutable_functions_walker about why we treat
697  * MinMaxExpr, XmlExpr, and CoerceToDomain as immutable, while
698  * SQLValueFunction is stable. Hence, none of them are of interest here.
699  * Also, since we're intentionally ignoring nextval(), presumably we
700  * should ignore NextValueExpr.
701  */
702 
703  /* Recurse to check arguments */
704  if (IsA(node, Query))
705  {
706  /* Recurse into subselects */
707  return query_tree_walker((Query *) node,
709  context, 0);
710  }
711  return expression_tree_walker(node,
713  context);
714 }
715 
716 
717 /*****************************************************************************
718  * Check queries for parallel unsafe and/or restricted constructs
719  *****************************************************************************/
720 
721 /*
722  * max_parallel_hazard
723  * Find the worst parallel-hazard level in the given query
724  *
725  * Returns the worst function hazard property (the earliest in this list:
726  * PROPARALLEL_UNSAFE, PROPARALLEL_RESTRICTED, PROPARALLEL_SAFE) that can
727  * be found in the given parsetree. We use this to find out whether the query
728  * can be parallelized at all. The caller will also save the result in
729  * PlannerGlobal so as to short-circuit checks of portions of the querytree
730  * later, in the common case where everything is SAFE.
731  */
732 char
734 {
736 
737  context.max_hazard = PROPARALLEL_SAFE;
738  context.max_interesting = PROPARALLEL_UNSAFE;
739  context.safe_param_ids = NIL;
741  return context.max_hazard;
742 }
743 
744 /*
745  * is_parallel_safe
746  * Detect whether the given expr contains only parallel-safe functions
747  *
748  * root->glob->maxParallelHazard must previously have been set to the
749  * result of max_parallel_hazard() on the whole query.
750  */
751 bool
753 {
755  PlannerInfo *proot;
756  ListCell *l;
757 
758  /*
759  * Even if the original querytree contained nothing unsafe, we need to
760  * search the expression if we have generated any PARAM_EXEC Params while
761  * planning, because those are parallel-restricted and there might be one
762  * in this expression. But otherwise we don't need to look.
763  */
764  if (root->glob->maxParallelHazard == PROPARALLEL_SAFE &&
765  root->glob->paramExecTypes == NIL)
766  return true;
767  /* Else use max_parallel_hazard's search logic, but stop on RESTRICTED */
768  context.max_hazard = PROPARALLEL_SAFE;
769  context.max_interesting = PROPARALLEL_RESTRICTED;
770  context.safe_param_ids = NIL;
771 
772  /*
773  * The params that refer to the same or parent query level are considered
774  * parallel-safe. The idea is that we compute such params at Gather or
775  * Gather Merge node and pass their value to workers.
776  */
777  for (proot = root; proot != NULL; proot = proot->parent_root)
778  {
779  foreach(l, proot->init_plans)
780  {
781  SubPlan *initsubplan = (SubPlan *) lfirst(l);
782 
783  context.safe_param_ids = list_concat(context.safe_param_ids,
784  initsubplan->setParam);
785  }
786  }
787 
788  return !max_parallel_hazard_walker(node, &context);
789 }
790 
791 /* core logic for all parallel-hazard checks */
792 static bool
794 {
795  switch (proparallel)
796  {
797  case PROPARALLEL_SAFE:
798  /* nothing to see here, move along */
799  break;
800  case PROPARALLEL_RESTRICTED:
801  /* increase max_hazard to RESTRICTED */
802  Assert(context->max_hazard != PROPARALLEL_UNSAFE);
803  context->max_hazard = proparallel;
804  /* done if we are not expecting any unsafe functions */
805  if (context->max_interesting == proparallel)
806  return true;
807  break;
808  case PROPARALLEL_UNSAFE:
809  context->max_hazard = proparallel;
810  /* we're always done at the first unsafe construct */
811  return true;
812  default:
813  elog(ERROR, "unrecognized proparallel value \"%c\"", proparallel);
814  break;
815  }
816  return false;
817 }
818 
819 /* check_functions_in_node callback */
820 static bool
822 {
823  return max_parallel_hazard_test(func_parallel(func_id),
825 }
826 
827 static bool
829 {
830  if (node == NULL)
831  return false;
832 
833  /* Check for hazardous functions in node itself */
835  context))
836  return true;
837 
838  /*
839  * It should be OK to treat MinMaxExpr as parallel-safe, since btree
840  * opclass support functions are generally parallel-safe. XmlExpr is a
841  * bit more dubious but we can probably get away with it. We err on the
842  * side of caution by treating CoerceToDomain as parallel-restricted.
843  * (Note: in principle that's wrong because a domain constraint could
844  * contain a parallel-unsafe function; but useful constraints probably
845  * never would have such, and assuming they do would cripple use of
846  * parallel query in the presence of domain types.) SQLValueFunction
847  * should be safe in all cases. NextValueExpr is parallel-unsafe.
848  */
849  if (IsA(node, CoerceToDomain))
850  {
851  if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context))
852  return true;
853  }
854 
855  else if (IsA(node, NextValueExpr))
856  {
857  if (max_parallel_hazard_test(PROPARALLEL_UNSAFE, context))
858  return true;
859  }
860 
861  /*
862  * Treat window functions as parallel-restricted because we aren't sure
863  * whether the input row ordering is fully deterministic, and the output
864  * of window functions might vary across workers if not. (In some cases,
865  * like where the window frame orders by a primary key, we could relax
866  * this restriction. But it doesn't currently seem worth expending extra
867  * effort to do so.)
868  */
869  else if (IsA(node, WindowFunc))
870  {
871  if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context))
872  return true;
873  }
874 
875  /*
876  * As a notational convenience for callers, look through RestrictInfo.
877  */
878  else if (IsA(node, RestrictInfo))
879  {
880  RestrictInfo *rinfo = (RestrictInfo *) node;
881 
882  return max_parallel_hazard_walker((Node *) rinfo->clause, context);
883  }
884 
885  /*
886  * Really we should not see SubLink during a max_interesting == restricted
887  * scan, but if we do, return true.
888  */
889  else if (IsA(node, SubLink))
890  {
891  if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context))
892  return true;
893  }
894 
895  /*
896  * Only parallel-safe SubPlans can be sent to workers. Within the
897  * testexpr of the SubPlan, Params representing the output columns of the
898  * subplan can be treated as parallel-safe, so temporarily add their IDs
899  * to the safe_param_ids list while examining the testexpr.
900  */
901  else if (IsA(node, SubPlan))
902  {
903  SubPlan *subplan = (SubPlan *) node;
904  List *save_safe_param_ids;
905 
906  if (!subplan->parallel_safe &&
907  max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context))
908  return true;
909  save_safe_param_ids = context->safe_param_ids;
910  context->safe_param_ids = list_concat_copy(context->safe_param_ids,
911  subplan->paramIds);
913  return true; /* no need to restore safe_param_ids */
914  list_free(context->safe_param_ids);
915  context->safe_param_ids = save_safe_param_ids;
916  /* we must also check args, but no special Param treatment there */
917  if (max_parallel_hazard_walker((Node *) subplan->args, context))
918  return true;
919  /* don't want to recurse normally, so we're done */
920  return false;
921  }
922 
923  /*
924  * We can't pass Params to workers at the moment either, so they are also
925  * parallel-restricted, unless they are PARAM_EXTERN Params or are
926  * PARAM_EXEC Params listed in safe_param_ids, meaning they could be
927  * either generated within workers or can be computed by the leader and
928  * then their value can be passed to workers.
929  */
930  else if (IsA(node, Param))
931  {
932  Param *param = (Param *) node;
933 
934  if (param->paramkind == PARAM_EXTERN)
935  return false;
936 
937  if (param->paramkind != PARAM_EXEC ||
938  !list_member_int(context->safe_param_ids, param->paramid))
939  {
940  if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context))
941  return true;
942  }
943  return false; /* nothing to recurse to */
944  }
945 
946  /*
947  * When we're first invoked on a completely unplanned tree, we must
948  * recurse into subqueries so to as to locate parallel-unsafe constructs
949  * anywhere in the tree.
950  */
951  else if (IsA(node, Query))
952  {
953  Query *query = (Query *) node;
954 
955  /* SELECT FOR UPDATE/SHARE must be treated as unsafe */
956  if (query->rowMarks != NULL)
957  {
958  context->max_hazard = PROPARALLEL_UNSAFE;
959  return true;
960  }
961 
962  /* Recurse into subselects */
963  return query_tree_walker(query,
965  context, 0);
966  }
967 
968  /* Recurse to check arguments */
969  return expression_tree_walker(node,
971  context);
972 }
973 
974 
975 /*****************************************************************************
976  * Check clauses for nonstrict functions
977  *****************************************************************************/
978 
979 /*
980  * contain_nonstrict_functions
981  * Recursively search for nonstrict functions within a clause.
982  *
983  * Returns true if any nonstrict construct is found --- ie, anything that
984  * could produce non-NULL output with a NULL input.
985  *
986  * The idea here is that the caller has verified that the expression contains
987  * one or more Var or Param nodes (as appropriate for the caller's need), and
988  * now wishes to prove that the expression result will be NULL if any of these
989  * inputs is NULL. If we return false, then the proof succeeded.
990  */
991 bool
993 {
994  return contain_nonstrict_functions_walker(clause, NULL);
995 }
996 
997 static bool
999 {
1000  return !func_strict(func_id);
1001 }
1002 
1003 static bool
1005 {
1006  if (node == NULL)
1007  return false;
1008  if (IsA(node, Aggref))
1009  {
1010  /* an aggregate could return non-null with null input */
1011  return true;
1012  }
1013  if (IsA(node, GroupingFunc))
1014  {
1015  /*
1016  * A GroupingFunc doesn't evaluate its arguments, and therefore must
1017  * be treated as nonstrict.
1018  */
1019  return true;
1020  }
1021  if (IsA(node, WindowFunc))
1022  {
1023  /* a window function could return non-null with null input */
1024  return true;
1025  }
1026  if (IsA(node, SubscriptingRef))
1027  {
1028  SubscriptingRef *sbsref = (SubscriptingRef *) node;
1029  const SubscriptRoutines *sbsroutines;
1030 
1031  /* Subscripting assignment is always presumed nonstrict */
1032  if (sbsref->refassgnexpr != NULL)
1033  return true;
1034  /* Otherwise we must look up the subscripting support methods */
1035  sbsroutines = getSubscriptingRoutines(sbsref->refcontainertype, NULL);
1036  if (!(sbsroutines && sbsroutines->fetch_strict))
1037  return true;
1038  /* else fall through to check args */
1039  }
1040  if (IsA(node, DistinctExpr))
1041  {
1042  /* IS DISTINCT FROM is inherently non-strict */
1043  return true;
1044  }
1045  if (IsA(node, NullIfExpr))
1046  {
1047  /* NULLIF is inherently non-strict */
1048  return true;
1049  }
1050  if (IsA(node, BoolExpr))
1051  {
1052  BoolExpr *expr = (BoolExpr *) node;
1053 
1054  switch (expr->boolop)
1055  {
1056  case AND_EXPR:
1057  case OR_EXPR:
1058  /* AND, OR are inherently non-strict */
1059  return true;
1060  default:
1061  break;
1062  }
1063  }
1064  if (IsA(node, SubLink))
1065  {
1066  /* In some cases a sublink might be strict, but in general not */
1067  return true;
1068  }
1069  if (IsA(node, SubPlan))
1070  return true;
1071  if (IsA(node, AlternativeSubPlan))
1072  return true;
1073  if (IsA(node, FieldStore))
1074  return true;
1075  if (IsA(node, CoerceViaIO))
1076  {
1077  /*
1078  * CoerceViaIO is strict regardless of whether the I/O functions are,
1079  * so just go look at its argument; asking check_functions_in_node is
1080  * useless expense and could deliver the wrong answer.
1081  */
1082  return contain_nonstrict_functions_walker((Node *) ((CoerceViaIO *) node)->arg,
1083  context);
1084  }
1085  if (IsA(node, ArrayCoerceExpr))
1086  {
1087  /*
1088  * ArrayCoerceExpr is strict at the array level, regardless of what
1089  * the per-element expression is; so we should ignore elemexpr and
1090  * recurse only into the arg.
1091  */
1093  context);
1094  }
1095  if (IsA(node, CaseExpr))
1096  return true;
1097  if (IsA(node, ArrayExpr))
1098  return true;
1099  if (IsA(node, RowExpr))
1100  return true;
1101  if (IsA(node, RowCompareExpr))
1102  return true;
1103  if (IsA(node, CoalesceExpr))
1104  return true;
1105  if (IsA(node, MinMaxExpr))
1106  return true;
1107  if (IsA(node, XmlExpr))
1108  return true;
1109  if (IsA(node, NullTest))
1110  return true;
1111  if (IsA(node, BooleanTest))
1112  return true;
1113 
1114  /* Check other function-containing nodes */
1116  context))
1117  return true;
1118 
1120  context);
1121 }
1122 
1123 /*****************************************************************************
1124  * Check clauses for Params
1125  *****************************************************************************/
1126 
1127 /*
1128  * contain_exec_param
1129  * Recursively search for PARAM_EXEC Params within a clause.
1130  *
1131  * Returns true if the clause contains any PARAM_EXEC Param with a paramid
1132  * appearing in the given list of Param IDs. Does not descend into
1133  * subqueries!
1134  */
1135 bool
1136 contain_exec_param(Node *clause, List *param_ids)
1137 {
1138  return contain_exec_param_walker(clause, param_ids);
1139 }
1140 
1141 static bool
1143 {
1144  if (node == NULL)
1145  return false;
1146  if (IsA(node, Param))
1147  {
1148  Param *p = (Param *) node;
1149 
1150  if (p->paramkind == PARAM_EXEC &&
1151  list_member_int(param_ids, p->paramid))
1152  return true;
1153  }
1154  return expression_tree_walker(node, contain_exec_param_walker, param_ids);
1155 }
1156 
1157 /*****************************************************************************
1158  * Check clauses for context-dependent nodes
1159  *****************************************************************************/
1160 
1161 /*
1162  * contain_context_dependent_node
1163  * Recursively search for context-dependent nodes within a clause.
1164  *
1165  * CaseTestExpr nodes must appear directly within the corresponding CaseExpr,
1166  * not nested within another one, or they'll see the wrong test value. If one
1167  * appears "bare" in the arguments of a SQL function, then we can't inline the
1168  * SQL function for fear of creating such a situation. The same applies for
1169  * CaseTestExpr used within the elemexpr of an ArrayCoerceExpr.
1170  *
1171  * CoerceToDomainValue would have the same issue if domain CHECK expressions
1172  * could get inlined into larger expressions, but presently that's impossible.
1173  * Still, it might be allowed in future, or other node types with similar
1174  * issues might get invented. So give this function a generic name, and set
1175  * up the recursion state to allow multiple flag bits.
1176  */
1177 static bool
1179 {
1180  int flags = 0;
1181 
1182  return contain_context_dependent_node_walker(clause, &flags);
1183 }
1184 
1185 #define CCDN_CASETESTEXPR_OK 0x0001 /* CaseTestExpr okay here? */
1186 
1187 static bool
1189 {
1190  if (node == NULL)
1191  return false;
1192  if (IsA(node, CaseTestExpr))
1193  return !(*flags & CCDN_CASETESTEXPR_OK);
1194  else if (IsA(node, CaseExpr))
1195  {
1196  CaseExpr *caseexpr = (CaseExpr *) node;
1197 
1198  /*
1199  * If this CASE doesn't have a test expression, then it doesn't create
1200  * a context in which CaseTestExprs should appear, so just fall
1201  * through and treat it as a generic expression node.
1202  */
1203  if (caseexpr->arg)
1204  {
1205  int save_flags = *flags;
1206  bool res;
1207 
1208  /*
1209  * Note: in principle, we could distinguish the various sub-parts
1210  * of a CASE construct and set the flag bit only for some of them,
1211  * since we are only expecting CaseTestExprs to appear in the
1212  * "expr" subtree of the CaseWhen nodes. But it doesn't really
1213  * seem worth any extra code. If there are any bare CaseTestExprs
1214  * elsewhere in the CASE, something's wrong already.
1215  */
1216  *flags |= CCDN_CASETESTEXPR_OK;
1217  res = expression_tree_walker(node,
1219  flags);
1220  *flags = save_flags;
1221  return res;
1222  }
1223  }
1224  else if (IsA(node, ArrayCoerceExpr))
1225  {
1226  ArrayCoerceExpr *ac = (ArrayCoerceExpr *) node;
1227  int save_flags;
1228  bool res;
1229 
1230  /* Check the array expression */
1231  if (contain_context_dependent_node_walker((Node *) ac->arg, flags))
1232  return true;
1233 
1234  /* Check the elemexpr, which is allowed to contain CaseTestExpr */
1235  save_flags = *flags;
1236  *flags |= CCDN_CASETESTEXPR_OK;
1238  flags);
1239  *flags = save_flags;
1240  return res;
1241  }
1243  flags);
1244 }
1245 
1246 /*****************************************************************************
1247  * Check clauses for Vars passed to non-leakproof functions
1248  *****************************************************************************/
1249 
1250 /*
1251  * contain_leaked_vars
1252  * Recursively scan a clause to discover whether it contains any Var
1253  * nodes (of the current query level) that are passed as arguments to
1254  * leaky functions.
1255  *
1256  * Returns true if the clause contains any non-leakproof functions that are
1257  * passed Var nodes of the current query level, and which might therefore leak
1258  * data. Such clauses must be applied after any lower-level security barrier
1259  * clauses.
1260  */
1261 bool
1263 {
1264  return contain_leaked_vars_walker(clause, NULL);
1265 }
1266 
1267 static bool
1269 {
1270  return !get_func_leakproof(func_id);
1271 }
1272 
1273 static bool
1275 {
1276  if (node == NULL)
1277  return false;
1278 
1279  switch (nodeTag(node))
1280  {
1281  case T_Var:
1282  case T_Const:
1283  case T_Param:
1284  case T_ArrayExpr:
1285  case T_FieldSelect:
1286  case T_FieldStore:
1287  case T_NamedArgExpr:
1288  case T_BoolExpr:
1289  case T_RelabelType:
1290  case T_CollateExpr:
1291  case T_CaseExpr:
1292  case T_CaseTestExpr:
1293  case T_RowExpr:
1294  case T_SQLValueFunction:
1295  case T_NullTest:
1296  case T_BooleanTest:
1297  case T_NextValueExpr:
1298  case T_List:
1299 
1300  /*
1301  * We know these node types don't contain function calls; but
1302  * something further down in the node tree might.
1303  */
1304  break;
1305 
1306  case T_FuncExpr:
1307  case T_OpExpr:
1308  case T_DistinctExpr:
1309  case T_NullIfExpr:
1310  case T_ScalarArrayOpExpr:
1311  case T_CoerceViaIO:
1312  case T_ArrayCoerceExpr:
1313 
1314  /*
1315  * If node contains a leaky function call, and there's any Var
1316  * underneath it, reject.
1317  */
1319  context) &&
1320  contain_var_clause(node))
1321  return true;
1322  break;
1323 
1324  case T_SubscriptingRef:
1325  {
1326  SubscriptingRef *sbsref = (SubscriptingRef *) node;
1327  const SubscriptRoutines *sbsroutines;
1328 
1329  /* Consult the subscripting support method info */
1330  sbsroutines = getSubscriptingRoutines(sbsref->refcontainertype,
1331  NULL);
1332  if (!sbsroutines ||
1333  !(sbsref->refassgnexpr != NULL ?
1334  sbsroutines->store_leakproof :
1335  sbsroutines->fetch_leakproof))
1336  {
1337  /* Node is leaky, so reject if it contains Vars */
1338  if (contain_var_clause(node))
1339  return true;
1340  }
1341  }
1342  break;
1343 
1344  case T_RowCompareExpr:
1345  {
1346  /*
1347  * It's worth special-casing this because a leaky comparison
1348  * function only compromises one pair of row elements, which
1349  * might not contain Vars while others do.
1350  */
1351  RowCompareExpr *rcexpr = (RowCompareExpr *) node;
1352  ListCell *opid;
1353  ListCell *larg;
1354  ListCell *rarg;
1355 
1356  forthree(opid, rcexpr->opnos,
1357  larg, rcexpr->largs,
1358  rarg, rcexpr->rargs)
1359  {
1360  Oid funcid = get_opcode(lfirst_oid(opid));
1361 
1362  if (!get_func_leakproof(funcid) &&
1363  (contain_var_clause((Node *) lfirst(larg)) ||
1364  contain_var_clause((Node *) lfirst(rarg))))
1365  return true;
1366  }
1367  }
1368  break;
1369 
1370  case T_MinMaxExpr:
1371  {
1372  /*
1373  * MinMaxExpr is leakproof if the comparison function it calls
1374  * is leakproof.
1375  */
1376  MinMaxExpr *minmaxexpr = (MinMaxExpr *) node;
1377  TypeCacheEntry *typentry;
1378  bool leakproof;
1379 
1380  /* Look up the btree comparison function for the datatype */
1381  typentry = lookup_type_cache(minmaxexpr->minmaxtype,
1383  if (OidIsValid(typentry->cmp_proc))
1384  leakproof = get_func_leakproof(typentry->cmp_proc);
1385  else
1386  {
1387  /*
1388  * The executor will throw an error, but here we just
1389  * treat the missing function as leaky.
1390  */
1391  leakproof = false;
1392  }
1393 
1394  if (!leakproof &&
1395  contain_var_clause((Node *) minmaxexpr->args))
1396  return true;
1397  }
1398  break;
1399 
1400  case T_CurrentOfExpr:
1401 
1402  /*
1403  * WHERE CURRENT OF doesn't contain leaky function calls.
1404  * Moreover, it is essential that this is considered non-leaky,
1405  * since the planner must always generate a TID scan when CURRENT
1406  * OF is present -- cf. cost_tidscan.
1407  */
1408  return false;
1409 
1410  default:
1411 
1412  /*
1413  * If we don't recognize the node tag, assume it might be leaky.
1414  * This prevents an unexpected security hole if someone adds a new
1415  * node type that can call a function.
1416  */
1417  return true;
1418  }
1420  context);
1421 }
1422 
1423 /*
1424  * find_nonnullable_rels
1425  * Determine which base rels are forced nonnullable by given clause.
1426  *
1427  * Returns the set of all Relids that are referenced in the clause in such
1428  * a way that the clause cannot possibly return TRUE if any of these Relids
1429  * is an all-NULL row. (It is OK to err on the side of conservatism; hence
1430  * the analysis here is simplistic.)
1431  *
1432  * The semantics here are subtly different from contain_nonstrict_functions:
1433  * that function is concerned with NULL results from arbitrary expressions,
1434  * but here we assume that the input is a Boolean expression, and wish to
1435  * see if NULL inputs will provably cause a FALSE-or-NULL result. We expect
1436  * the expression to have been AND/OR flattened and converted to implicit-AND
1437  * format.
1438  *
1439  * Note: this function is largely duplicative of find_nonnullable_vars().
1440  * The reason not to simplify this function into a thin wrapper around
1441  * find_nonnullable_vars() is that the tested conditions really are different:
1442  * a clause like "t1.v1 IS NOT NULL OR t1.v2 IS NOT NULL" does not prove
1443  * that either v1 or v2 can't be NULL, but it does prove that the t1 row
1444  * as a whole can't be all-NULL. Also, the behavior for PHVs is different.
1445  *
1446  * top_level is true while scanning top-level AND/OR structure; here, showing
1447  * the result is either FALSE or NULL is good enough. top_level is false when
1448  * we have descended below a NOT or a strict function: now we must be able to
1449  * prove that the subexpression goes to NULL.
1450  *
1451  * We don't use expression_tree_walker here because we don't want to descend
1452  * through very many kinds of nodes; only the ones we can be sure are strict.
1453  */
1454 Relids
1456 {
1457  return find_nonnullable_rels_walker(clause, true);
1458 }
1459 
1460 static Relids
1461 find_nonnullable_rels_walker(Node *node, bool top_level)
1462 {
1463  Relids result = NULL;
1464  ListCell *l;
1465 
1466  if (node == NULL)
1467  return NULL;
1468  if (IsA(node, Var))
1469  {
1470  Var *var = (Var *) node;
1471 
1472  if (var->varlevelsup == 0)
1473  result = bms_make_singleton(var->varno);
1474  }
1475  else if (IsA(node, List))
1476  {
1477  /*
1478  * At top level, we are examining an implicit-AND list: if any of the
1479  * arms produces FALSE-or-NULL then the result is FALSE-or-NULL. If
1480  * not at top level, we are examining the arguments of a strict
1481  * function: if any of them produce NULL then the result of the
1482  * function must be NULL. So in both cases, the set of nonnullable
1483  * rels is the union of those found in the arms, and we pass down the
1484  * top_level flag unmodified.
1485  */
1486  foreach(l, (List *) node)
1487  {
1488  result = bms_join(result,
1490  top_level));
1491  }
1492  }
1493  else if (IsA(node, FuncExpr))
1494  {
1495  FuncExpr *expr = (FuncExpr *) node;
1496 
1497  if (func_strict(expr->funcid))
1498  result = find_nonnullable_rels_walker((Node *) expr->args, false);
1499  }
1500  else if (IsA(node, OpExpr))
1501  {
1502  OpExpr *expr = (OpExpr *) node;
1503 
1504  set_opfuncid(expr);
1505  if (func_strict(expr->opfuncid))
1506  result = find_nonnullable_rels_walker((Node *) expr->args, false);
1507  }
1508  else if (IsA(node, ScalarArrayOpExpr))
1509  {
1510  ScalarArrayOpExpr *expr = (ScalarArrayOpExpr *) node;
1511 
1512  if (is_strict_saop(expr, true))
1513  result = find_nonnullable_rels_walker((Node *) expr->args, false);
1514  }
1515  else if (IsA(node, BoolExpr))
1516  {
1517  BoolExpr *expr = (BoolExpr *) node;
1518 
1519  switch (expr->boolop)
1520  {
1521  case AND_EXPR:
1522  /* At top level we can just recurse (to the List case) */
1523  if (top_level)
1524  {
1525  result = find_nonnullable_rels_walker((Node *) expr->args,
1526  top_level);
1527  break;
1528  }
1529 
1530  /*
1531  * Below top level, even if one arm produces NULL, the result
1532  * could be FALSE (hence not NULL). However, if *all* the
1533  * arms produce NULL then the result is NULL, so we can take
1534  * the intersection of the sets of nonnullable rels, just as
1535  * for OR. Fall through to share code.
1536  */
1537  /* FALL THRU */
1538  case OR_EXPR:
1539 
1540  /*
1541  * OR is strict if all of its arms are, so we can take the
1542  * intersection of the sets of nonnullable rels for each arm.
1543  * This works for both values of top_level.
1544  */
1545  foreach(l, expr->args)
1546  {
1547  Relids subresult;
1548 
1549  subresult = find_nonnullable_rels_walker(lfirst(l),
1550  top_level);
1551  if (result == NULL) /* first subresult? */
1552  result = subresult;
1553  else
1554  result = bms_int_members(result, subresult);
1555 
1556  /*
1557  * If the intersection is empty, we can stop looking. This
1558  * also justifies the test for first-subresult above.
1559  */
1560  if (bms_is_empty(result))
1561  break;
1562  }
1563  break;
1564  case NOT_EXPR:
1565  /* NOT will return null if its arg is null */
1566  result = find_nonnullable_rels_walker((Node *) expr->args,
1567  false);
1568  break;
1569  default:
1570  elog(ERROR, "unrecognized boolop: %d", (int) expr->boolop);
1571  break;
1572  }
1573  }
1574  else if (IsA(node, RelabelType))
1575  {
1576  RelabelType *expr = (RelabelType *) node;
1577 
1578  result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
1579  }
1580  else if (IsA(node, CoerceViaIO))
1581  {
1582  /* not clear this is useful, but it can't hurt */
1583  CoerceViaIO *expr = (CoerceViaIO *) node;
1584 
1585  result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
1586  }
1587  else if (IsA(node, ArrayCoerceExpr))
1588  {
1589  /* ArrayCoerceExpr is strict at the array level; ignore elemexpr */
1590  ArrayCoerceExpr *expr = (ArrayCoerceExpr *) node;
1591 
1592  result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
1593  }
1594  else if (IsA(node, ConvertRowtypeExpr))
1595  {
1596  /* not clear this is useful, but it can't hurt */
1597  ConvertRowtypeExpr *expr = (ConvertRowtypeExpr *) node;
1598 
1599  result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
1600  }
1601  else if (IsA(node, CollateExpr))
1602  {
1603  CollateExpr *expr = (CollateExpr *) node;
1604 
1605  result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
1606  }
1607  else if (IsA(node, NullTest))
1608  {
1609  /* IS NOT NULL can be considered strict, but only at top level */
1610  NullTest *expr = (NullTest *) node;
1611 
1612  if (top_level && expr->nulltesttype == IS_NOT_NULL && !expr->argisrow)
1613  result = find_nonnullable_rels_walker((Node *) expr->arg, false);
1614  }
1615  else if (IsA(node, BooleanTest))
1616  {
1617  /* Boolean tests that reject NULL are strict at top level */
1618  BooleanTest *expr = (BooleanTest *) node;
1619 
1620  if (top_level &&
1621  (expr->booltesttype == IS_TRUE ||
1622  expr->booltesttype == IS_FALSE ||
1623  expr->booltesttype == IS_NOT_UNKNOWN))
1624  result = find_nonnullable_rels_walker((Node *) expr->arg, false);
1625  }
1626  else if (IsA(node, SubPlan))
1627  {
1628  SubPlan *splan = (SubPlan *) node;
1629 
1630  /*
1631  * For some types of SubPlan, we can infer strictness from Vars in the
1632  * testexpr (the LHS of the original SubLink).
1633  *
1634  * For ANY_SUBLINK, if the subquery produces zero rows, the result is
1635  * always FALSE. If the subquery produces more than one row, the
1636  * per-row results of the testexpr are combined using OR semantics.
1637  * Hence ANY_SUBLINK can be strict only at top level, but there it's
1638  * as strict as the testexpr is.
1639  *
1640  * For ROWCOMPARE_SUBLINK, if the subquery produces zero rows, the
1641  * result is always NULL. Otherwise, the result is as strict as the
1642  * testexpr is. So we can check regardless of top_level.
1643  *
1644  * We can't prove anything for other sublink types (in particular,
1645  * note that ALL_SUBLINK will return TRUE if the subquery is empty).
1646  */
1647  if ((top_level && splan->subLinkType == ANY_SUBLINK) ||
1648  splan->subLinkType == ROWCOMPARE_SUBLINK)
1649  result = find_nonnullable_rels_walker(splan->testexpr, top_level);
1650  }
1651  else if (IsA(node, PlaceHolderVar))
1652  {
1653  PlaceHolderVar *phv = (PlaceHolderVar *) node;
1654 
1655  /*
1656  * If the contained expression forces any rels non-nullable, so does
1657  * the PHV.
1658  */
1659  result = find_nonnullable_rels_walker((Node *) phv->phexpr, top_level);
1660 
1661  /*
1662  * If the PHV's syntactic scope is exactly one rel, it will be forced
1663  * to be evaluated at that rel, and so it will behave like a Var of
1664  * that rel: if the rel's entire output goes to null, so will the PHV.
1665  * (If the syntactic scope is a join, we know that the PHV will go to
1666  * null if the whole join does; but that is AND semantics while we
1667  * need OR semantics for find_nonnullable_rels' result, so we can't do
1668  * anything with the knowledge.)
1669  */
1670  if (phv->phlevelsup == 0 &&
1671  bms_membership(phv->phrels) == BMS_SINGLETON)
1672  result = bms_add_members(result, phv->phrels);
1673  }
1674  return result;
1675 }
1676 
1677 /*
1678  * find_nonnullable_vars
1679  * Determine which Vars are forced nonnullable by given clause.
1680  *
1681  * Returns the set of all level-zero Vars that are referenced in the clause in
1682  * such a way that the clause cannot possibly return TRUE if any of these Vars
1683  * is NULL. (It is OK to err on the side of conservatism; hence the analysis
1684  * here is simplistic.)
1685  *
1686  * The semantics here are subtly different from contain_nonstrict_functions:
1687  * that function is concerned with NULL results from arbitrary expressions,
1688  * but here we assume that the input is a Boolean expression, and wish to
1689  * see if NULL inputs will provably cause a FALSE-or-NULL result. We expect
1690  * the expression to have been AND/OR flattened and converted to implicit-AND
1691  * format.
1692  *
1693  * Attnos of the identified Vars are returned in a multibitmapset (a List of
1694  * Bitmapsets). List indexes correspond to relids (varnos), while the per-rel
1695  * Bitmapsets hold varattnos offset by FirstLowInvalidHeapAttributeNumber.
1696  *
1697  * top_level is true while scanning top-level AND/OR structure; here, showing
1698  * the result is either FALSE or NULL is good enough. top_level is false when
1699  * we have descended below a NOT or a strict function: now we must be able to
1700  * prove that the subexpression goes to NULL.
1701  *
1702  * We don't use expression_tree_walker here because we don't want to descend
1703  * through very many kinds of nodes; only the ones we can be sure are strict.
1704  */
1705 List *
1707 {
1708  return find_nonnullable_vars_walker(clause, true);
1709 }
1710 
1711 static List *
1712 find_nonnullable_vars_walker(Node *node, bool top_level)
1713 {
1714  List *result = NIL;
1715  ListCell *l;
1716 
1717  if (node == NULL)
1718  return NIL;
1719  if (IsA(node, Var))
1720  {
1721  Var *var = (Var *) node;
1722 
1723  if (var->varlevelsup == 0)
1724  result = mbms_add_member(result,
1725  var->varno,
1727  }
1728  else if (IsA(node, List))
1729  {
1730  /*
1731  * At top level, we are examining an implicit-AND list: if any of the
1732  * arms produces FALSE-or-NULL then the result is FALSE-or-NULL. If
1733  * not at top level, we are examining the arguments of a strict
1734  * function: if any of them produce NULL then the result of the
1735  * function must be NULL. So in both cases, the set of nonnullable
1736  * vars is the union of those found in the arms, and we pass down the
1737  * top_level flag unmodified.
1738  */
1739  foreach(l, (List *) node)
1740  {
1741  result = mbms_add_members(result,
1743  top_level));
1744  }
1745  }
1746  else if (IsA(node, FuncExpr))
1747  {
1748  FuncExpr *expr = (FuncExpr *) node;
1749 
1750  if (func_strict(expr->funcid))
1751  result = find_nonnullable_vars_walker((Node *) expr->args, false);
1752  }
1753  else if (IsA(node, OpExpr))
1754  {
1755  OpExpr *expr = (OpExpr *) node;
1756 
1757  set_opfuncid(expr);
1758  if (func_strict(expr->opfuncid))
1759  result = find_nonnullable_vars_walker((Node *) expr->args, false);
1760  }
1761  else if (IsA(node, ScalarArrayOpExpr))
1762  {
1763  ScalarArrayOpExpr *expr = (ScalarArrayOpExpr *) node;
1764 
1765  if (is_strict_saop(expr, true))
1766  result = find_nonnullable_vars_walker((Node *) expr->args, false);
1767  }
1768  else if (IsA(node, BoolExpr))
1769  {
1770  BoolExpr *expr = (BoolExpr *) node;
1771 
1772  switch (expr->boolop)
1773  {
1774  case AND_EXPR:
1775 
1776  /*
1777  * At top level we can just recurse (to the List case), since
1778  * the result should be the union of what we can prove in each
1779  * arm.
1780  */
1781  if (top_level)
1782  {
1783  result = find_nonnullable_vars_walker((Node *) expr->args,
1784  top_level);
1785  break;
1786  }
1787 
1788  /*
1789  * Below top level, even if one arm produces NULL, the result
1790  * could be FALSE (hence not NULL). However, if *all* the
1791  * arms produce NULL then the result is NULL, so we can take
1792  * the intersection of the sets of nonnullable vars, just as
1793  * for OR. Fall through to share code.
1794  */
1795  /* FALL THRU */
1796  case OR_EXPR:
1797 
1798  /*
1799  * OR is strict if all of its arms are, so we can take the
1800  * intersection of the sets of nonnullable vars for each arm.
1801  * This works for both values of top_level.
1802  */
1803  foreach(l, expr->args)
1804  {
1805  List *subresult;
1806 
1807  subresult = find_nonnullable_vars_walker(lfirst(l),
1808  top_level);
1809  if (result == NIL) /* first subresult? */
1810  result = subresult;
1811  else
1812  result = mbms_int_members(result, subresult);
1813 
1814  /*
1815  * If the intersection is empty, we can stop looking. This
1816  * also justifies the test for first-subresult above.
1817  */
1818  if (result == NIL)
1819  break;
1820  }
1821  break;
1822  case NOT_EXPR:
1823  /* NOT will return null if its arg is null */
1824  result = find_nonnullable_vars_walker((Node *) expr->args,
1825  false);
1826  break;
1827  default:
1828  elog(ERROR, "unrecognized boolop: %d", (int) expr->boolop);
1829  break;
1830  }
1831  }
1832  else if (IsA(node, RelabelType))
1833  {
1834  RelabelType *expr = (RelabelType *) node;
1835 
1836  result = find_nonnullable_vars_walker((Node *) expr->arg, top_level);
1837  }
1838  else if (IsA(node, CoerceViaIO))
1839  {
1840  /* not clear this is useful, but it can't hurt */
1841  CoerceViaIO *expr = (CoerceViaIO *) node;
1842 
1843  result = find_nonnullable_vars_walker((Node *) expr->arg, false);
1844  }
1845  else if (IsA(node, ArrayCoerceExpr))
1846  {
1847  /* ArrayCoerceExpr is strict at the array level; ignore elemexpr */
1848  ArrayCoerceExpr *expr = (ArrayCoerceExpr *) node;
1849 
1850  result = find_nonnullable_vars_walker((Node *) expr->arg, top_level);
1851  }
1852  else if (IsA(node, ConvertRowtypeExpr))
1853  {
1854  /* not clear this is useful, but it can't hurt */
1855  ConvertRowtypeExpr *expr = (ConvertRowtypeExpr *) node;
1856 
1857  result = find_nonnullable_vars_walker((Node *) expr->arg, top_level);
1858  }
1859  else if (IsA(node, CollateExpr))
1860  {
1861  CollateExpr *expr = (CollateExpr *) node;
1862 
1863  result = find_nonnullable_vars_walker((Node *) expr->arg, top_level);
1864  }
1865  else if (IsA(node, NullTest))
1866  {
1867  /* IS NOT NULL can be considered strict, but only at top level */
1868  NullTest *expr = (NullTest *) node;
1869 
1870  if (top_level && expr->nulltesttype == IS_NOT_NULL && !expr->argisrow)
1871  result = find_nonnullable_vars_walker((Node *) expr->arg, false);
1872  }
1873  else if (IsA(node, BooleanTest))
1874  {
1875  /* Boolean tests that reject NULL are strict at top level */
1876  BooleanTest *expr = (BooleanTest *) node;
1877 
1878  if (top_level &&
1879  (expr->booltesttype == IS_TRUE ||
1880  expr->booltesttype == IS_FALSE ||
1881  expr->booltesttype == IS_NOT_UNKNOWN))
1882  result = find_nonnullable_vars_walker((Node *) expr->arg, false);
1883  }
1884  else if (IsA(node, SubPlan))
1885  {
1886  SubPlan *splan = (SubPlan *) node;
1887 
1888  /* See analysis in find_nonnullable_rels_walker */
1889  if ((top_level && splan->subLinkType == ANY_SUBLINK) ||
1890  splan->subLinkType == ROWCOMPARE_SUBLINK)
1891  result = find_nonnullable_vars_walker(splan->testexpr, top_level);
1892  }
1893  else if (IsA(node, PlaceHolderVar))
1894  {
1895  PlaceHolderVar *phv = (PlaceHolderVar *) node;
1896 
1897  result = find_nonnullable_vars_walker((Node *) phv->phexpr, top_level);
1898  }
1899  return result;
1900 }
1901 
1902 /*
1903  * find_forced_null_vars
1904  * Determine which Vars must be NULL for the given clause to return TRUE.
1905  *
1906  * This is the complement of find_nonnullable_vars: find the level-zero Vars
1907  * that must be NULL for the clause to return TRUE. (It is OK to err on the
1908  * side of conservatism; hence the analysis here is simplistic. In fact,
1909  * we only detect simple "var IS NULL" tests at the top level.)
1910  *
1911  * As with find_nonnullable_vars, we return the varattnos of the identified
1912  * Vars in a multibitmapset.
1913  */
1914 List *
1916 {
1917  List *result = NIL;
1918  Var *var;
1919  ListCell *l;
1920 
1921  if (node == NULL)
1922  return NIL;
1923  /* Check single-clause cases using subroutine */
1924  var = find_forced_null_var(node);
1925  if (var)
1926  {
1927  result = mbms_add_member(result,
1928  var->varno,
1930  }
1931  /* Otherwise, handle AND-conditions */
1932  else if (IsA(node, List))
1933  {
1934  /*
1935  * At top level, we are examining an implicit-AND list: if any of the
1936  * arms produces FALSE-or-NULL then the result is FALSE-or-NULL.
1937  */
1938  foreach(l, (List *) node)
1939  {
1940  result = mbms_add_members(result,
1942  }
1943  }
1944  else if (IsA(node, BoolExpr))
1945  {
1946  BoolExpr *expr = (BoolExpr *) node;
1947 
1948  /*
1949  * We don't bother considering the OR case, because it's fairly
1950  * unlikely anyone would write "v1 IS NULL OR v1 IS NULL". Likewise,
1951  * the NOT case isn't worth expending code on.
1952  */
1953  if (expr->boolop == AND_EXPR)
1954  {
1955  /* At top level we can just recurse (to the List case) */
1956  result = find_forced_null_vars((Node *) expr->args);
1957  }
1958  }
1959  return result;
1960 }
1961 
1962 /*
1963  * find_forced_null_var
1964  * Return the Var forced null by the given clause, or NULL if it's
1965  * not an IS NULL-type clause. For success, the clause must enforce
1966  * *only* nullness of the particular Var, not any other conditions.
1967  *
1968  * This is just the single-clause case of find_forced_null_vars(), without
1969  * any allowance for AND conditions. It's used by initsplan.c on individual
1970  * qual clauses. The reason for not just applying find_forced_null_vars()
1971  * is that if an AND of an IS NULL clause with something else were to somehow
1972  * survive AND/OR flattening, initsplan.c might get fooled into discarding
1973  * the whole clause when only the IS NULL part of it had been proved redundant.
1974  */
1975 Var *
1977 {
1978  if (node == NULL)
1979  return NULL;
1980  if (IsA(node, NullTest))
1981  {
1982  /* check for var IS NULL */
1983  NullTest *expr = (NullTest *) node;
1984 
1985  if (expr->nulltesttype == IS_NULL && !expr->argisrow)
1986  {
1987  Var *var = (Var *) expr->arg;
1988 
1989  if (var && IsA(var, Var) &&
1990  var->varlevelsup == 0)
1991  return var;
1992  }
1993  }
1994  else if (IsA(node, BooleanTest))
1995  {
1996  /* var IS UNKNOWN is equivalent to var IS NULL */
1997  BooleanTest *expr = (BooleanTest *) node;
1998 
1999  if (expr->booltesttype == IS_UNKNOWN)
2000  {
2001  Var *var = (Var *) expr->arg;
2002 
2003  if (var && IsA(var, Var) &&
2004  var->varlevelsup == 0)
2005  return var;
2006  }
2007  }
2008  return NULL;
2009 }
2010 
2011 /*
2012  * Can we treat a ScalarArrayOpExpr as strict?
2013  *
2014  * If "falseOK" is true, then a "false" result can be considered strict,
2015  * else we need to guarantee an actual NULL result for NULL input.
2016  *
2017  * "foo op ALL array" is strict if the op is strict *and* we can prove
2018  * that the array input isn't an empty array. We can check that
2019  * for the cases of an array constant and an ARRAY[] construct.
2020  *
2021  * "foo op ANY array" is strict in the falseOK sense if the op is strict.
2022  * If not falseOK, the test is the same as for "foo op ALL array".
2023  */
2024 static bool
2025 is_strict_saop(ScalarArrayOpExpr *expr, bool falseOK)
2026 {
2027  Node *rightop;
2028 
2029  /* The contained operator must be strict. */
2030  set_sa_opfuncid(expr);
2031  if (!func_strict(expr->opfuncid))
2032  return false;
2033  /* If ANY and falseOK, that's all we need to check. */
2034  if (expr->useOr && falseOK)
2035  return true;
2036  /* Else, we have to see if the array is provably non-empty. */
2037  Assert(list_length(expr->args) == 2);
2038  rightop = (Node *) lsecond(expr->args);
2039  if (rightop && IsA(rightop, Const))
2040  {
2041  Datum arraydatum = ((Const *) rightop)->constvalue;
2042  bool arrayisnull = ((Const *) rightop)->constisnull;
2043  ArrayType *arrayval;
2044  int nitems;
2045 
2046  if (arrayisnull)
2047  return false;
2048  arrayval = DatumGetArrayTypeP(arraydatum);
2049  nitems = ArrayGetNItems(ARR_NDIM(arrayval), ARR_DIMS(arrayval));
2050  if (nitems > 0)
2051  return true;
2052  }
2053  else if (rightop && IsA(rightop, ArrayExpr))
2054  {
2055  ArrayExpr *arrayexpr = (ArrayExpr *) rightop;
2056 
2057  if (arrayexpr->elements != NIL && !arrayexpr->multidims)
2058  return true;
2059  }
2060  return false;
2061 }
2062 
2063 
2064 /*****************************************************************************
2065  * Check for "pseudo-constant" clauses
2066  *****************************************************************************/
2067 
2068 /*
2069  * is_pseudo_constant_clause
2070  * Detect whether an expression is "pseudo constant", ie, it contains no
2071  * variables of the current query level and no uses of volatile functions.
2072  * Such an expr is not necessarily a true constant: it can still contain
2073  * Params and outer-level Vars, not to mention functions whose results
2074  * may vary from one statement to the next. However, the expr's value
2075  * will be constant over any one scan of the current query, so it can be
2076  * used as, eg, an indexscan key. (Actually, the condition for indexscan
2077  * keys is weaker than this; see is_pseudo_constant_for_index().)
2078  *
2079  * CAUTION: this function omits to test for one very important class of
2080  * not-constant expressions, namely aggregates (Aggrefs). In current usage
2081  * this is only applied to WHERE clauses and so a check for Aggrefs would be
2082  * a waste of cycles; but be sure to also check contain_agg_clause() if you
2083  * want to know about pseudo-constness in other contexts. The same goes
2084  * for window functions (WindowFuncs).
2085  */
2086 bool
2088 {
2089  /*
2090  * We could implement this check in one recursive scan. But since the
2091  * check for volatile functions is both moderately expensive and unlikely
2092  * to fail, it seems better to look for Vars first and only check for
2093  * volatile functions if we find no Vars.
2094  */
2095  if (!contain_var_clause(clause) &&
2096  !contain_volatile_functions(clause))
2097  return true;
2098  return false;
2099 }
2100 
2101 /*
2102  * is_pseudo_constant_clause_relids
2103  * Same as above, except caller already has available the var membership
2104  * of the expression; this lets us avoid the contain_var_clause() scan.
2105  */
2106 bool
2108 {
2109  if (bms_is_empty(relids) &&
2110  !contain_volatile_functions(clause))
2111  return true;
2112  return false;
2113 }
2114 
2115 
2116 /*****************************************************************************
2117  * *
2118  * General clause-manipulating routines *
2119  * *
2120  *****************************************************************************/
2121 
2122 /*
2123  * NumRelids
2124  * (formerly clause_relids)
2125  *
2126  * Returns the number of different base relations referenced in 'clause'.
2127  */
2128 int
2130 {
2131  int result;
2132  Relids varnos = pull_varnos(root, clause);
2133 
2134  varnos = bms_del_members(varnos, root->outer_join_rels);
2135  result = bms_num_members(varnos);
2136  bms_free(varnos);
2137  return result;
2138 }
2139 
2140 /*
2141  * CommuteOpExpr: commute a binary operator clause
2142  *
2143  * XXX the clause is destructively modified!
2144  */
2145 void
2147 {
2148  Oid opoid;
2149  Node *temp;
2150 
2151  /* Sanity checks: caller is at fault if these fail */
2152  if (!is_opclause(clause) ||
2153  list_length(clause->args) != 2)
2154  elog(ERROR, "cannot commute non-binary-operator clause");
2155 
2156  opoid = get_commutator(clause->opno);
2157 
2158  if (!OidIsValid(opoid))
2159  elog(ERROR, "could not find commutator for operator %u",
2160  clause->opno);
2161 
2162  /*
2163  * modify the clause in-place!
2164  */
2165  clause->opno = opoid;
2166  clause->opfuncid = InvalidOid;
2167  /* opresulttype, opretset, opcollid, inputcollid need not change */
2168 
2169  temp = linitial(clause->args);
2170  linitial(clause->args) = lsecond(clause->args);
2171  lsecond(clause->args) = temp;
2172 }
2173 
2174 /*
2175  * Helper for eval_const_expressions: check that datatype of an attribute
2176  * is still what it was when the expression was parsed. This is needed to
2177  * guard against improper simplification after ALTER COLUMN TYPE. (XXX we
2178  * may well need to make similar checks elsewhere?)
2179  *
2180  * rowtypeid may come from a whole-row Var, and therefore it can be a domain
2181  * over composite, but for this purpose we only care about checking the type
2182  * of a contained field.
2183  */
2184 static bool
2185 rowtype_field_matches(Oid rowtypeid, int fieldnum,
2186  Oid expectedtype, int32 expectedtypmod,
2187  Oid expectedcollation)
2188 {
2189  TupleDesc tupdesc;
2190  Form_pg_attribute attr;
2191 
2192  /* No issue for RECORD, since there is no way to ALTER such a type */
2193  if (rowtypeid == RECORDOID)
2194  return true;
2195  tupdesc = lookup_rowtype_tupdesc_domain(rowtypeid, -1, false);
2196  if (fieldnum <= 0 || fieldnum > tupdesc->natts)
2197  {
2198  ReleaseTupleDesc(tupdesc);
2199  return false;
2200  }
2201  attr = TupleDescAttr(tupdesc, fieldnum - 1);
2202  if (attr->attisdropped ||
2203  attr->atttypid != expectedtype ||
2204  attr->atttypmod != expectedtypmod ||
2205  attr->attcollation != expectedcollation)
2206  {
2207  ReleaseTupleDesc(tupdesc);
2208  return false;
2209  }
2210  ReleaseTupleDesc(tupdesc);
2211  return true;
2212 }
2213 
2214 
2215 /*--------------------
2216  * eval_const_expressions
2217  *
2218  * Reduce any recognizably constant subexpressions of the given
2219  * expression tree, for example "2 + 2" => "4". More interestingly,
2220  * we can reduce certain boolean expressions even when they contain
2221  * non-constant subexpressions: "x OR true" => "true" no matter what
2222  * the subexpression x is. (XXX We assume that no such subexpression
2223  * will have important side-effects, which is not necessarily a good
2224  * assumption in the presence of user-defined functions; do we need a
2225  * pg_proc flag that prevents discarding the execution of a function?)
2226  *
2227  * We do understand that certain functions may deliver non-constant
2228  * results even with constant inputs, "nextval()" being the classic
2229  * example. Functions that are not marked "immutable" in pg_proc
2230  * will not be pre-evaluated here, although we will reduce their
2231  * arguments as far as possible.
2232  *
2233  * Whenever a function is eliminated from the expression by means of
2234  * constant-expression evaluation or inlining, we add the function to
2235  * root->glob->invalItems. This ensures the plan is known to depend on
2236  * such functions, even though they aren't referenced anymore.
2237  *
2238  * We assume that the tree has already been type-checked and contains
2239  * only operators and functions that are reasonable to try to execute.
2240  *
2241  * NOTE: "root" can be passed as NULL if the caller never wants to do any
2242  * Param substitutions nor receive info about inlined functions.
2243  *
2244  * NOTE: the planner assumes that this will always flatten nested AND and
2245  * OR clauses into N-argument form. See comments in prepqual.c.
2246  *
2247  * NOTE: another critical effect is that any function calls that require
2248  * default arguments will be expanded, and named-argument calls will be
2249  * converted to positional notation. The executor won't handle either.
2250  *--------------------
2251  */
2252 Node *
2254 {
2256 
2257  if (root)
2258  context.boundParams = root->glob->boundParams; /* bound Params */
2259  else
2260  context.boundParams = NULL;
2261  context.root = root; /* for inlined-function dependencies */
2262  context.active_fns = NIL; /* nothing being recursively simplified */
2263  context.case_val = NULL; /* no CASE being examined */
2264  context.estimate = false; /* safe transformations only */
2265  return eval_const_expressions_mutator(node, &context);
2266 }
2267 
2268 #define MIN_ARRAY_SIZE_FOR_HASHED_SAOP 9
2269 /*--------------------
2270  * convert_saop_to_hashed_saop
2271  *
2272  * Recursively search 'node' for ScalarArrayOpExprs and fill in the hash
2273  * function for any ScalarArrayOpExpr that looks like it would be useful to
2274  * evaluate using a hash table rather than a linear search.
2275  *
2276  * We'll use a hash table if all of the following conditions are met:
2277  * 1. The 2nd argument of the array contain only Consts.
2278  * 2. useOr is true or there is a valid negator operator for the
2279  * ScalarArrayOpExpr's opno.
2280  * 3. There's valid hash function for both left and righthand operands and
2281  * these hash functions are the same.
2282  * 4. If the array contains enough elements for us to consider it to be
2283  * worthwhile using a hash table rather than a linear search.
2284  */
2285 void
2287 {
2288  (void) convert_saop_to_hashed_saop_walker(node, NULL);
2289 }
2290 
2291 static bool
2293 {
2294  if (node == NULL)
2295  return false;
2296 
2297  if (IsA(node, ScalarArrayOpExpr))
2298  {
2299  ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) node;
2300  Expr *arrayarg = (Expr *) lsecond(saop->args);
2301  Oid lefthashfunc;
2302  Oid righthashfunc;
2303 
2304  if (arrayarg && IsA(arrayarg, Const) &&
2305  !((Const *) arrayarg)->constisnull)
2306  {
2307  if (saop->useOr)
2308  {
2309  if (get_op_hash_functions(saop->opno, &lefthashfunc, &righthashfunc) &&
2310  lefthashfunc == righthashfunc)
2311  {
2312  Datum arrdatum = ((Const *) arrayarg)->constvalue;
2313  ArrayType *arr = (ArrayType *) DatumGetPointer(arrdatum);
2314  int nitems;
2315 
2316  /*
2317  * Only fill in the hash functions if the array looks
2318  * large enough for it to be worth hashing instead of
2319  * doing a linear search.
2320  */
2321  nitems = ArrayGetNItems(ARR_NDIM(arr), ARR_DIMS(arr));
2322 
2324  {
2325  /* Looks good. Fill in the hash functions */
2326  saop->hashfuncid = lefthashfunc;
2327  }
2328  return true;
2329  }
2330  }
2331  else /* !saop->useOr */
2332  {
2333  Oid negator = get_negator(saop->opno);
2334 
2335  /*
2336  * Check if this is a NOT IN using an operator whose negator
2337  * is hashable. If so we can still build a hash table and
2338  * just ensure the lookup items are not in the hash table.
2339  */
2340  if (OidIsValid(negator) &&
2341  get_op_hash_functions(negator, &lefthashfunc, &righthashfunc) &&
2342  lefthashfunc == righthashfunc)
2343  {
2344  Datum arrdatum = ((Const *) arrayarg)->constvalue;
2345  ArrayType *arr = (ArrayType *) DatumGetPointer(arrdatum);
2346  int nitems;
2347 
2348  /*
2349  * Only fill in the hash functions if the array looks
2350  * large enough for it to be worth hashing instead of
2351  * doing a linear search.
2352  */
2353  nitems = ArrayGetNItems(ARR_NDIM(arr), ARR_DIMS(arr));
2354 
2356  {
2357  /* Looks good. Fill in the hash functions */
2358  saop->hashfuncid = lefthashfunc;
2359 
2360  /*
2361  * Also set the negfuncid. The executor will need
2362  * that to perform hashtable lookups.
2363  */
2364  saop->negfuncid = get_opcode(negator);
2365  }
2366  return true;
2367  }
2368  }
2369  }
2370  }
2371 
2373 }
2374 
2375 
2376 /*--------------------
2377  * estimate_expression_value
2378  *
2379  * This function attempts to estimate the value of an expression for
2380  * planning purposes. It is in essence a more aggressive version of
2381  * eval_const_expressions(): we will perform constant reductions that are
2382  * not necessarily 100% safe, but are reasonable for estimation purposes.
2383  *
2384  * Currently the extra steps that are taken in this mode are:
2385  * 1. Substitute values for Params, where a bound Param value has been made
2386  * available by the caller of planner(), even if the Param isn't marked
2387  * constant. This effectively means that we plan using the first supplied
2388  * value of the Param.
2389  * 2. Fold stable, as well as immutable, functions to constants.
2390  * 3. Reduce PlaceHolderVar nodes to their contained expressions.
2391  *--------------------
2392  */
2393 Node *
2395 {
2397 
2398  context.boundParams = root->glob->boundParams; /* bound Params */
2399  /* we do not need to mark the plan as depending on inlined functions */
2400  context.root = NULL;
2401  context.active_fns = NIL; /* nothing being recursively simplified */
2402  context.case_val = NULL; /* no CASE being examined */
2403  context.estimate = true; /* unsafe transformations OK */
2404  return eval_const_expressions_mutator(node, &context);
2405 }
2406 
2407 /*
2408  * The generic case in eval_const_expressions_mutator is to recurse using
2409  * expression_tree_mutator, which will copy the given node unchanged but
2410  * const-simplify its arguments (if any) as far as possible. If the node
2411  * itself does immutable processing, and each of its arguments were reduced
2412  * to a Const, we can then reduce it to a Const using evaluate_expr. (Some
2413  * node types need more complicated logic; for example, a CASE expression
2414  * might be reducible to a constant even if not all its subtrees are.)
2415  */
2416 #define ece_generic_processing(node) \
2417  expression_tree_mutator((Node *) (node), eval_const_expressions_mutator, \
2418  context)
2419 
2420 /*
2421  * Check whether all arguments of the given node were reduced to Consts.
2422  * By going directly to expression_tree_walker, contain_non_const_walker
2423  * is not applied to the node itself, only to its children.
2424  */
2425 #define ece_all_arguments_const(node) \
2426  (!expression_tree_walker((Node *) (node), contain_non_const_walker, NULL))
2427 
2428 /* Generic macro for applying evaluate_expr */
2429 #define ece_evaluate_expr(node) \
2430  ((Node *) evaluate_expr((Expr *) (node), \
2431  exprType((Node *) (node)), \
2432  exprTypmod((Node *) (node)), \
2433  exprCollation((Node *) (node))))
2434 
2435 /*
2436  * Recursive guts of eval_const_expressions/estimate_expression_value
2437  */
2438 static Node *
2441 {
2442 
2443  /* since this function recurses, it could be driven to stack overflow */
2445 
2446  if (node == NULL)
2447  return NULL;
2448  switch (nodeTag(node))
2449  {
2450  case T_Param:
2451  {
2452  Param *param = (Param *) node;
2453  ParamListInfo paramLI = context->boundParams;
2454 
2455  /* Look to see if we've been given a value for this Param */
2456  if (param->paramkind == PARAM_EXTERN &&
2457  paramLI != NULL &&
2458  param->paramid > 0 &&
2459  param->paramid <= paramLI->numParams)
2460  {
2461  ParamExternData *prm;
2462  ParamExternData prmdata;
2463 
2464  /*
2465  * Give hook a chance in case parameter is dynamic. Tell
2466  * it that this fetch is speculative, so it should avoid
2467  * erroring out if parameter is unavailable.
2468  */
2469  if (paramLI->paramFetch != NULL)
2470  prm = paramLI->paramFetch(paramLI, param->paramid,
2471  true, &prmdata);
2472  else
2473  prm = &paramLI->params[param->paramid - 1];
2474 
2475  /*
2476  * We don't just check OidIsValid, but insist that the
2477  * fetched type match the Param, just in case the hook did
2478  * something unexpected. No need to throw an error here
2479  * though; leave that for runtime.
2480  */
2481  if (OidIsValid(prm->ptype) &&
2482  prm->ptype == param->paramtype)
2483  {
2484  /* OK to substitute parameter value? */
2485  if (context->estimate ||
2486  (prm->pflags & PARAM_FLAG_CONST))
2487  {
2488  /*
2489  * Return a Const representing the param value.
2490  * Must copy pass-by-ref datatypes, since the
2491  * Param might be in a memory context
2492  * shorter-lived than our output plan should be.
2493  */
2494  int16 typLen;
2495  bool typByVal;
2496  Datum pval;
2497  Const *con;
2498 
2499  get_typlenbyval(param->paramtype,
2500  &typLen, &typByVal);
2501  if (prm->isnull || typByVal)
2502  pval = prm->value;
2503  else
2504  pval = datumCopy(prm->value, typByVal, typLen);
2505  con = makeConst(param->paramtype,
2506  param->paramtypmod,
2507  param->paramcollid,
2508  (int) typLen,
2509  pval,
2510  prm->isnull,
2511  typByVal);
2512  con->location = param->location;
2513  return (Node *) con;
2514  }
2515  }
2516  }
2517 
2518  /*
2519  * Not replaceable, so just copy the Param (no need to
2520  * recurse)
2521  */
2522  return (Node *) copyObject(param);
2523  }
2524  case T_WindowFunc:
2525  {
2526  WindowFunc *expr = (WindowFunc *) node;
2527  Oid funcid = expr->winfnoid;
2528  List *args;
2529  Expr *aggfilter;
2530  HeapTuple func_tuple;
2531  WindowFunc *newexpr;
2532 
2533  /*
2534  * We can't really simplify a WindowFunc node, but we mustn't
2535  * just fall through to the default processing, because we
2536  * have to apply expand_function_arguments to its argument
2537  * list. That takes care of inserting default arguments and
2538  * expanding named-argument notation.
2539  */
2540  func_tuple = SearchSysCache1(PROCOID, ObjectIdGetDatum(funcid));
2541  if (!HeapTupleIsValid(func_tuple))
2542  elog(ERROR, "cache lookup failed for function %u", funcid);
2543 
2545  false, expr->wintype,
2546  func_tuple);
2547 
2548  ReleaseSysCache(func_tuple);
2549 
2550  /* Now, recursively simplify the args (which are a List) */
2551  args = (List *)
2554  context);
2555  /* ... and the filter expression, which isn't */
2556  aggfilter = (Expr *)
2558  context);
2559 
2560  /* And build the replacement WindowFunc node */
2561  newexpr = makeNode(WindowFunc);
2562  newexpr->winfnoid = expr->winfnoid;
2563  newexpr->wintype = expr->wintype;
2564  newexpr->wincollid = expr->wincollid;
2565  newexpr->inputcollid = expr->inputcollid;
2566  newexpr->args = args;
2567  newexpr->aggfilter = aggfilter;
2568  newexpr->runCondition = expr->runCondition;
2569  newexpr->winref = expr->winref;
2570  newexpr->winstar = expr->winstar;
2571  newexpr->winagg = expr->winagg;
2572  newexpr->location = expr->location;
2573 
2574  return (Node *) newexpr;
2575  }
2576  case T_FuncExpr:
2577  {
2578  FuncExpr *expr = (FuncExpr *) node;
2579  List *args = expr->args;
2580  Expr *simple;
2581  FuncExpr *newexpr;
2582 
2583  /*
2584  * Code for op/func reduction is pretty bulky, so split it out
2585  * as a separate function. Note: exprTypmod normally returns
2586  * -1 for a FuncExpr, but not when the node is recognizably a
2587  * length coercion; we want to preserve the typmod in the
2588  * eventual Const if so.
2589  */
2590  simple = simplify_function(expr->funcid,
2591  expr->funcresulttype,
2592  exprTypmod(node),
2593  expr->funccollid,
2594  expr->inputcollid,
2595  &args,
2596  expr->funcvariadic,
2597  true,
2598  true,
2599  context);
2600  if (simple) /* successfully simplified it */
2601  return (Node *) simple;
2602 
2603  /*
2604  * The expression cannot be simplified any further, so build
2605  * and return a replacement FuncExpr node using the
2606  * possibly-simplified arguments. Note that we have also
2607  * converted the argument list to positional notation.
2608  */
2609  newexpr = makeNode(FuncExpr);
2610  newexpr->funcid = expr->funcid;
2611  newexpr->funcresulttype = expr->funcresulttype;
2612  newexpr->funcretset = expr->funcretset;
2613  newexpr->funcvariadic = expr->funcvariadic;
2614  newexpr->funcformat = expr->funcformat;
2615  newexpr->funccollid = expr->funccollid;
2616  newexpr->inputcollid = expr->inputcollid;
2617  newexpr->args = args;
2618  newexpr->location = expr->location;
2619  return (Node *) newexpr;
2620  }
2621  case T_OpExpr:
2622  {
2623  OpExpr *expr = (OpExpr *) node;
2624  List *args = expr->args;
2625  Expr *simple;
2626  OpExpr *newexpr;
2627 
2628  /*
2629  * Need to get OID of underlying function. Okay to scribble
2630  * on input to this extent.
2631  */
2632  set_opfuncid(expr);
2633 
2634  /*
2635  * Code for op/func reduction is pretty bulky, so split it out
2636  * as a separate function.
2637  */
2638  simple = simplify_function(expr->opfuncid,
2639  expr->opresulttype, -1,
2640  expr->opcollid,
2641  expr->inputcollid,
2642  &args,
2643  false,
2644  true,
2645  true,
2646  context);
2647  if (simple) /* successfully simplified it */
2648  return (Node *) simple;
2649 
2650  /*
2651  * If the operator is boolean equality or inequality, we know
2652  * how to simplify cases involving one constant and one
2653  * non-constant argument.
2654  */
2655  if (expr->opno == BooleanEqualOperator ||
2656  expr->opno == BooleanNotEqualOperator)
2657  {
2658  simple = (Expr *) simplify_boolean_equality(expr->opno,
2659  args);
2660  if (simple) /* successfully simplified it */
2661  return (Node *) simple;
2662  }
2663 
2664  /*
2665  * The expression cannot be simplified any further, so build
2666  * and return a replacement OpExpr node using the
2667  * possibly-simplified arguments.
2668  */
2669  newexpr = makeNode(OpExpr);
2670  newexpr->opno = expr->opno;
2671  newexpr->opfuncid = expr->opfuncid;
2672  newexpr->opresulttype = expr->opresulttype;
2673  newexpr->opretset = expr->opretset;
2674  newexpr->opcollid = expr->opcollid;
2675  newexpr->inputcollid = expr->inputcollid;
2676  newexpr->args = args;
2677  newexpr->location = expr->location;
2678  return (Node *) newexpr;
2679  }
2680  case T_DistinctExpr:
2681  {
2682  DistinctExpr *expr = (DistinctExpr *) node;
2683  List *args;
2684  ListCell *arg;
2685  bool has_null_input = false;
2686  bool all_null_input = true;
2687  bool has_nonconst_input = false;
2688  Expr *simple;
2689  DistinctExpr *newexpr;
2690 
2691  /*
2692  * Reduce constants in the DistinctExpr's arguments. We know
2693  * args is either NIL or a List node, so we can call
2694  * expression_tree_mutator directly rather than recursing to
2695  * self.
2696  */
2697  args = (List *) expression_tree_mutator((Node *) expr->args,
2699  context);
2700 
2701  /*
2702  * We must do our own check for NULLs because DistinctExpr has
2703  * different results for NULL input than the underlying
2704  * operator does.
2705  */
2706  foreach(arg, args)
2707  {
2708  if (IsA(lfirst(arg), Const))
2709  {
2710  has_null_input |= ((Const *) lfirst(arg))->constisnull;
2711  all_null_input &= ((Const *) lfirst(arg))->constisnull;
2712  }
2713  else
2714  has_nonconst_input = true;
2715  }
2716 
2717  /* all constants? then can optimize this out */
2718  if (!has_nonconst_input)
2719  {
2720  /* all nulls? then not distinct */
2721  if (all_null_input)
2722  return makeBoolConst(false, false);
2723 
2724  /* one null? then distinct */
2725  if (has_null_input)
2726  return makeBoolConst(true, false);
2727 
2728  /* otherwise try to evaluate the '=' operator */
2729  /* (NOT okay to try to inline it, though!) */
2730 
2731  /*
2732  * Need to get OID of underlying function. Okay to
2733  * scribble on input to this extent.
2734  */
2735  set_opfuncid((OpExpr *) expr); /* rely on struct
2736  * equivalence */
2737 
2738  /*
2739  * Code for op/func reduction is pretty bulky, so split it
2740  * out as a separate function.
2741  */
2742  simple = simplify_function(expr->opfuncid,
2743  expr->opresulttype, -1,
2744  expr->opcollid,
2745  expr->inputcollid,
2746  &args,
2747  false,
2748  false,
2749  false,
2750  context);
2751  if (simple) /* successfully simplified it */
2752  {
2753  /*
2754  * Since the underlying operator is "=", must negate
2755  * its result
2756  */
2757  Const *csimple = castNode(Const, simple);
2758 
2759  csimple->constvalue =
2760  BoolGetDatum(!DatumGetBool(csimple->constvalue));
2761  return (Node *) csimple;
2762  }
2763  }
2764 
2765  /*
2766  * The expression cannot be simplified any further, so build
2767  * and return a replacement DistinctExpr node using the
2768  * possibly-simplified arguments.
2769  */
2770  newexpr = makeNode(DistinctExpr);
2771  newexpr->opno = expr->opno;
2772  newexpr->opfuncid = expr->opfuncid;
2773  newexpr->opresulttype = expr->opresulttype;
2774  newexpr->opretset = expr->opretset;
2775  newexpr->opcollid = expr->opcollid;
2776  newexpr->inputcollid = expr->inputcollid;
2777  newexpr->args = args;
2778  newexpr->location = expr->location;
2779  return (Node *) newexpr;
2780  }
2781  case T_NullIfExpr:
2782  {
2783  NullIfExpr *expr;
2784  ListCell *arg;
2785  bool has_nonconst_input = false;
2786 
2787  /* Copy the node and const-simplify its arguments */
2788  expr = (NullIfExpr *) ece_generic_processing(node);
2789 
2790  /* If either argument is NULL they can't be equal */
2791  foreach(arg, expr->args)
2792  {
2793  if (!IsA(lfirst(arg), Const))
2794  has_nonconst_input = true;
2795  else if (((Const *) lfirst(arg))->constisnull)
2796  return (Node *) linitial(expr->args);
2797  }
2798 
2799  /*
2800  * Need to get OID of underlying function before checking if
2801  * the function is OK to evaluate.
2802  */
2803  set_opfuncid((OpExpr *) expr);
2804 
2805  if (!has_nonconst_input &&
2806  ece_function_is_safe(expr->opfuncid, context))
2807  return ece_evaluate_expr(expr);
2808 
2809  return (Node *) expr;
2810  }
2811  case T_ScalarArrayOpExpr:
2812  {
2813  ScalarArrayOpExpr *saop;
2814 
2815  /* Copy the node and const-simplify its arguments */
2816  saop = (ScalarArrayOpExpr *) ece_generic_processing(node);
2817 
2818  /* Make sure we know underlying function */
2819  set_sa_opfuncid(saop);
2820 
2821  /*
2822  * If all arguments are Consts, and it's a safe function, we
2823  * can fold to a constant
2824  */
2825  if (ece_all_arguments_const(saop) &&
2826  ece_function_is_safe(saop->opfuncid, context))
2827  return ece_evaluate_expr(saop);
2828  return (Node *) saop;
2829  }
2830  case T_BoolExpr:
2831  {
2832  BoolExpr *expr = (BoolExpr *) node;
2833 
2834  switch (expr->boolop)
2835  {
2836  case OR_EXPR:
2837  {
2838  List *newargs;
2839  bool haveNull = false;
2840  bool forceTrue = false;
2841 
2842  newargs = simplify_or_arguments(expr->args,
2843  context,
2844  &haveNull,
2845  &forceTrue);
2846  if (forceTrue)
2847  return makeBoolConst(true, false);
2848  if (haveNull)
2849  newargs = lappend(newargs,
2850  makeBoolConst(false, true));
2851  /* If all the inputs are FALSE, result is FALSE */
2852  if (newargs == NIL)
2853  return makeBoolConst(false, false);
2854 
2855  /*
2856  * If only one nonconst-or-NULL input, it's the
2857  * result
2858  */
2859  if (list_length(newargs) == 1)
2860  return (Node *) linitial(newargs);
2861  /* Else we still need an OR node */
2862  return (Node *) make_orclause(newargs);
2863  }
2864  case AND_EXPR:
2865  {
2866  List *newargs;
2867  bool haveNull = false;
2868  bool forceFalse = false;
2869 
2870  newargs = simplify_and_arguments(expr->args,
2871  context,
2872  &haveNull,
2873  &forceFalse);
2874  if (forceFalse)
2875  return makeBoolConst(false, false);
2876  if (haveNull)
2877  newargs = lappend(newargs,
2878  makeBoolConst(false, true));
2879  /* If all the inputs are TRUE, result is TRUE */
2880  if (newargs == NIL)
2881  return makeBoolConst(true, false);
2882 
2883  /*
2884  * If only one nonconst-or-NULL input, it's the
2885  * result
2886  */
2887  if (list_length(newargs) == 1)
2888  return (Node *) linitial(newargs);
2889  /* Else we still need an AND node */
2890  return (Node *) make_andclause(newargs);
2891  }
2892  case NOT_EXPR:
2893  {
2894  Node *arg;
2895 
2896  Assert(list_length(expr->args) == 1);
2898  context);
2899 
2900  /*
2901  * Use negate_clause() to see if we can simplify
2902  * away the NOT.
2903  */
2904  return negate_clause(arg);
2905  }
2906  default:
2907  elog(ERROR, "unrecognized boolop: %d",
2908  (int) expr->boolop);
2909  break;
2910  }
2911  break;
2912  }
2913 
2914  case T_JsonValueExpr:
2915  {
2916  JsonValueExpr *jve = (JsonValueExpr *) node;
2917  Node *raw_expr = (Node *) jve->raw_expr;
2918  Node *formatted_expr = (Node *) jve->formatted_expr;
2919 
2920  /*
2921  * If we can fold formatted_expr to a constant, we can elide
2922  * the JsonValueExpr altogether. Otherwise we must process
2923  * raw_expr too. But JsonFormat is a flat node and requires
2924  * no simplification, only copying.
2925  */
2926  formatted_expr = eval_const_expressions_mutator(formatted_expr,
2927  context);
2928  if (formatted_expr && IsA(formatted_expr, Const))
2929  return formatted_expr;
2930 
2931  raw_expr = eval_const_expressions_mutator(raw_expr, context);
2932 
2933  return (Node *) makeJsonValueExpr((Expr *) raw_expr,
2934  (Expr *) formatted_expr,
2935  copyObject(jve->format));
2936  }
2937 
2938  case T_SubPlan:
2939  case T_AlternativeSubPlan:
2940 
2941  /*
2942  * Return a SubPlan unchanged --- too late to do anything with it.
2943  *
2944  * XXX should we ereport() here instead? Probably this routine
2945  * should never be invoked after SubPlan creation.
2946  */
2947  return node;
2948  case T_RelabelType:
2949  {
2950  RelabelType *relabel = (RelabelType *) node;
2951  Node *arg;
2952 
2953  /* Simplify the input ... */
2954  arg = eval_const_expressions_mutator((Node *) relabel->arg,
2955  context);
2956  /* ... and attach a new RelabelType node, if needed */
2957  return applyRelabelType(arg,
2958  relabel->resulttype,
2959  relabel->resulttypmod,
2960  relabel->resultcollid,
2961  relabel->relabelformat,
2962  relabel->location,
2963  true);
2964  }
2965  case T_CoerceViaIO:
2966  {
2967  CoerceViaIO *expr = (CoerceViaIO *) node;
2968  List *args;
2969  Oid outfunc;
2970  bool outtypisvarlena;
2971  Oid infunc;
2972  Oid intypioparam;
2973  Expr *simple;
2974  CoerceViaIO *newexpr;
2975 
2976  /* Make a List so we can use simplify_function */
2977  args = list_make1(expr->arg);
2978 
2979  /*
2980  * CoerceViaIO represents calling the source type's output
2981  * function then the result type's input function. So, try to
2982  * simplify it as though it were a stack of two such function
2983  * calls. First we need to know what the functions are.
2984  *
2985  * Note that the coercion functions are assumed not to care
2986  * about input collation, so we just pass InvalidOid for that.
2987  */
2988  getTypeOutputInfo(exprType((Node *) expr->arg),
2989  &outfunc, &outtypisvarlena);
2991  &infunc, &intypioparam);
2992 
2993  simple = simplify_function(outfunc,
2994  CSTRINGOID, -1,
2995  InvalidOid,
2996  InvalidOid,
2997  &args,
2998  false,
2999  true,
3000  true,
3001  context);
3002  if (simple) /* successfully simplified output fn */
3003  {
3004  /*
3005  * Input functions may want 1 to 3 arguments. We always
3006  * supply all three, trusting that nothing downstream will
3007  * complain.
3008  */
3009  args = list_make3(simple,
3010  makeConst(OIDOID,
3011  -1,
3012  InvalidOid,
3013  sizeof(Oid),
3014  ObjectIdGetDatum(intypioparam),
3015  false,
3016  true),
3017  makeConst(INT4OID,
3018  -1,
3019  InvalidOid,
3020  sizeof(int32),
3021  Int32GetDatum(-1),
3022  false,
3023  true));
3024 
3025  simple = simplify_function(infunc,
3026  expr->resulttype, -1,
3027  expr->resultcollid,
3028  InvalidOid,
3029  &args,
3030  false,
3031  false,
3032  true,
3033  context);
3034  if (simple) /* successfully simplified input fn */
3035  return (Node *) simple;
3036  }
3037 
3038  /*
3039  * The expression cannot be simplified any further, so build
3040  * and return a replacement CoerceViaIO node using the
3041  * possibly-simplified argument.
3042  */
3043  newexpr = makeNode(CoerceViaIO);
3044  newexpr->arg = (Expr *) linitial(args);
3045  newexpr->resulttype = expr->resulttype;
3046  newexpr->resultcollid = expr->resultcollid;
3047  newexpr->coerceformat = expr->coerceformat;
3048  newexpr->location = expr->location;
3049  return (Node *) newexpr;
3050  }
3051  case T_ArrayCoerceExpr:
3052  {
3054  Node *save_case_val;
3055 
3056  /*
3057  * Copy the node and const-simplify its arguments. We can't
3058  * use ece_generic_processing() here because we need to mess
3059  * with case_val only while processing the elemexpr.
3060  */
3061  memcpy(ac, node, sizeof(ArrayCoerceExpr));
3062  ac->arg = (Expr *)
3064  context);
3065 
3066  /*
3067  * Set up for the CaseTestExpr node contained in the elemexpr.
3068  * We must prevent it from absorbing any outer CASE value.
3069  */
3070  save_case_val = context->case_val;
3071  context->case_val = NULL;
3072 
3073  ac->elemexpr = (Expr *)
3075  context);
3076 
3077  context->case_val = save_case_val;
3078 
3079  /*
3080  * If constant argument and the per-element expression is
3081  * immutable, we can simplify the whole thing to a constant.
3082  * Exception: although contain_mutable_functions considers
3083  * CoerceToDomain immutable for historical reasons, let's not
3084  * do so here; this ensures coercion to an array-over-domain
3085  * does not apply the domain's constraints until runtime.
3086  */
3087  if (ac->arg && IsA(ac->arg, Const) &&
3088  ac->elemexpr && !IsA(ac->elemexpr, CoerceToDomain) &&
3090  return ece_evaluate_expr(ac);
3091 
3092  return (Node *) ac;
3093  }
3094  case T_CollateExpr:
3095  {
3096  /*
3097  * We replace CollateExpr with RelabelType, so as to improve
3098  * uniformity of expression representation and thus simplify
3099  * comparison of expressions. Hence this looks very nearly
3100  * the same as the RelabelType case, and we can apply the same
3101  * optimizations to avoid unnecessary RelabelTypes.
3102  */
3103  CollateExpr *collate = (CollateExpr *) node;
3104  Node *arg;
3105 
3106  /* Simplify the input ... */
3107  arg = eval_const_expressions_mutator((Node *) collate->arg,
3108  context);
3109  /* ... and attach a new RelabelType node, if needed */
3110  return applyRelabelType(arg,
3111  exprType(arg),
3112  exprTypmod(arg),
3113  collate->collOid,
3115  collate->location,
3116  true);
3117  }
3118  case T_CaseExpr:
3119  {
3120  /*----------
3121  * CASE expressions can be simplified if there are constant
3122  * condition clauses:
3123  * FALSE (or NULL): drop the alternative
3124  * TRUE: drop all remaining alternatives
3125  * If the first non-FALSE alternative is a constant TRUE,
3126  * we can simplify the entire CASE to that alternative's
3127  * expression. If there are no non-FALSE alternatives,
3128  * we simplify the entire CASE to the default result (ELSE).
3129  *
3130  * If we have a simple-form CASE with constant test
3131  * expression, we substitute the constant value for contained
3132  * CaseTestExpr placeholder nodes, so that we have the
3133  * opportunity to reduce constant test conditions. For
3134  * example this allows
3135  * CASE 0 WHEN 0 THEN 1 ELSE 1/0 END
3136  * to reduce to 1 rather than drawing a divide-by-0 error.
3137  * Note that when the test expression is constant, we don't
3138  * have to include it in the resulting CASE; for example
3139  * CASE 0 WHEN x THEN y ELSE z END
3140  * is transformed by the parser to
3141  * CASE 0 WHEN CaseTestExpr = x THEN y ELSE z END
3142  * which we can simplify to
3143  * CASE WHEN 0 = x THEN y ELSE z END
3144  * It is not necessary for the executor to evaluate the "arg"
3145  * expression when executing the CASE, since any contained
3146  * CaseTestExprs that might have referred to it will have been
3147  * replaced by the constant.
3148  *----------
3149  */
3150  CaseExpr *caseexpr = (CaseExpr *) node;
3151  CaseExpr *newcase;
3152  Node *save_case_val;
3153  Node *newarg;
3154  List *newargs;
3155  bool const_true_cond;
3156  Node *defresult = NULL;
3157  ListCell *arg;
3158 
3159  /* Simplify the test expression, if any */
3160  newarg = eval_const_expressions_mutator((Node *) caseexpr->arg,
3161  context);
3162 
3163  /* Set up for contained CaseTestExpr nodes */
3164  save_case_val = context->case_val;
3165  if (newarg && IsA(newarg, Const))
3166  {
3167  context->case_val = newarg;
3168  newarg = NULL; /* not needed anymore, see above */
3169  }
3170  else
3171  context->case_val = NULL;
3172 
3173  /* Simplify the WHEN clauses */
3174  newargs = NIL;
3175  const_true_cond = false;
3176  foreach(arg, caseexpr->args)
3177  {
3178  CaseWhen *oldcasewhen = lfirst_node(CaseWhen, arg);
3179  Node *casecond;
3180  Node *caseresult;
3181 
3182  /* Simplify this alternative's test condition */
3183  casecond = eval_const_expressions_mutator((Node *) oldcasewhen->expr,
3184  context);
3185 
3186  /*
3187  * If the test condition is constant FALSE (or NULL), then
3188  * drop this WHEN clause completely, without processing
3189  * the result.
3190  */
3191  if (casecond && IsA(casecond, Const))
3192  {
3193  Const *const_input = (Const *) casecond;
3194 
3195  if (const_input->constisnull ||
3196  !DatumGetBool(const_input->constvalue))
3197  continue; /* drop alternative with FALSE cond */
3198  /* Else it's constant TRUE */
3199  const_true_cond = true;
3200  }
3201 
3202  /* Simplify this alternative's result value */
3203  caseresult = eval_const_expressions_mutator((Node *) oldcasewhen->result,
3204  context);
3205 
3206  /* If non-constant test condition, emit a new WHEN node */
3207  if (!const_true_cond)
3208  {
3209  CaseWhen *newcasewhen = makeNode(CaseWhen);
3210 
3211  newcasewhen->expr = (Expr *) casecond;
3212  newcasewhen->result = (Expr *) caseresult;
3213  newcasewhen->location = oldcasewhen->location;
3214  newargs = lappend(newargs, newcasewhen);
3215  continue;
3216  }
3217 
3218  /*
3219  * Found a TRUE condition, so none of the remaining
3220  * alternatives can be reached. We treat the result as
3221  * the default result.
3222  */
3223  defresult = caseresult;
3224  break;
3225  }
3226 
3227  /* Simplify the default result, unless we replaced it above */
3228  if (!const_true_cond)
3229  defresult = eval_const_expressions_mutator((Node *) caseexpr->defresult,
3230  context);
3231 
3232  context->case_val = save_case_val;
3233 
3234  /*
3235  * If no non-FALSE alternatives, CASE reduces to the default
3236  * result
3237  */
3238  if (newargs == NIL)
3239  return defresult;
3240  /* Otherwise we need a new CASE node */
3241  newcase = makeNode(CaseExpr);
3242  newcase->casetype = caseexpr->casetype;
3243  newcase->casecollid = caseexpr->casecollid;
3244  newcase->arg = (Expr *) newarg;
3245  newcase->args = newargs;
3246  newcase->defresult = (Expr *) defresult;
3247  newcase->location = caseexpr->location;
3248  return (Node *) newcase;
3249  }
3250  case T_CaseTestExpr:
3251  {
3252  /*
3253  * If we know a constant test value for the current CASE
3254  * construct, substitute it for the placeholder. Else just
3255  * return the placeholder as-is.
3256  */
3257  if (context->case_val)
3258  return copyObject(context->case_val);
3259  else
3260  return copyObject(node);
3261  }
3262  case T_SubscriptingRef:
3263  case T_ArrayExpr:
3264  case T_RowExpr:
3265  case T_MinMaxExpr:
3266  {
3267  /*
3268  * Generic handling for node types whose own processing is
3269  * known to be immutable, and for which we need no smarts
3270  * beyond "simplify if all inputs are constants".
3271  *
3272  * Treating SubscriptingRef this way assumes that subscripting
3273  * fetch and assignment are both immutable. This constrains
3274  * type-specific subscripting implementations; maybe we should
3275  * relax it someday.
3276  *
3277  * Treating MinMaxExpr this way amounts to assuming that the
3278  * btree comparison function it calls is immutable; see the
3279  * reasoning in contain_mutable_functions_walker.
3280  */
3281 
3282  /* Copy the node and const-simplify its arguments */
3283  node = ece_generic_processing(node);
3284  /* If all arguments are Consts, we can fold to a constant */
3285  if (ece_all_arguments_const(node))
3286  return ece_evaluate_expr(node);
3287  return node;
3288  }
3289  case T_CoalesceExpr:
3290  {
3291  CoalesceExpr *coalesceexpr = (CoalesceExpr *) node;
3292  CoalesceExpr *newcoalesce;
3293  List *newargs;
3294  ListCell *arg;
3295 
3296  newargs = NIL;
3297  foreach(arg, coalesceexpr->args)
3298  {
3299  Node *e;
3300 
3302  context);
3303 
3304  /*
3305  * We can remove null constants from the list. For a
3306  * non-null constant, if it has not been preceded by any
3307  * other non-null-constant expressions then it is the
3308  * result. Otherwise, it's the next argument, but we can
3309  * drop following arguments since they will never be
3310  * reached.
3311  */
3312  if (IsA(e, Const))
3313  {
3314  if (((Const *) e)->constisnull)
3315  continue; /* drop null constant */
3316  if (newargs == NIL)
3317  return e; /* first expr */
3318  newargs = lappend(newargs, e);
3319  break;
3320  }
3321  newargs = lappend(newargs, e);
3322  }
3323 
3324  /*
3325  * If all the arguments were constant null, the result is just
3326  * null
3327  */
3328  if (newargs == NIL)
3329  return (Node *) makeNullConst(coalesceexpr->coalescetype,
3330  -1,
3331  coalesceexpr->coalescecollid);
3332 
3333  newcoalesce = makeNode(CoalesceExpr);
3334  newcoalesce->coalescetype = coalesceexpr->coalescetype;
3335  newcoalesce->coalescecollid = coalesceexpr->coalescecollid;
3336  newcoalesce->args = newargs;
3337  newcoalesce->location = coalesceexpr->location;
3338  return (Node *) newcoalesce;
3339  }
3340  case T_SQLValueFunction:
3341  {
3342  /*
3343  * All variants of SQLValueFunction are stable, so if we are
3344  * estimating the expression's value, we should evaluate the
3345  * current function value. Otherwise just copy.
3346  */
3347  SQLValueFunction *svf = (SQLValueFunction *) node;
3348 
3349  if (context->estimate)
3350  return (Node *) evaluate_expr((Expr *) svf,
3351  svf->type,
3352  svf->typmod,
3353  InvalidOid);
3354  else
3355  return copyObject((Node *) svf);
3356  }
3357  case T_FieldSelect:
3358  {
3359  /*
3360  * We can optimize field selection from a whole-row Var into a
3361  * simple Var. (This case won't be generated directly by the
3362  * parser, because ParseComplexProjection short-circuits it.
3363  * But it can arise while simplifying functions.) Also, we
3364  * can optimize field selection from a RowExpr construct, or
3365  * of course from a constant.
3366  *
3367  * However, replacing a whole-row Var in this way has a
3368  * pitfall: if we've already built the rel targetlist for the
3369  * source relation, then the whole-row Var is scheduled to be
3370  * produced by the relation scan, but the simple Var probably
3371  * isn't, which will lead to a failure in setrefs.c. This is
3372  * not a problem when handling simple single-level queries, in
3373  * which expression simplification always happens first. It
3374  * is a risk for lateral references from subqueries, though.
3375  * To avoid such failures, don't optimize uplevel references.
3376  *
3377  * We must also check that the declared type of the field is
3378  * still the same as when the FieldSelect was created --- this
3379  * can change if someone did ALTER COLUMN TYPE on the rowtype.
3380  * If it isn't, we skip the optimization; the case will
3381  * probably fail at runtime, but that's not our problem here.
3382  */
3383  FieldSelect *fselect = (FieldSelect *) node;
3384  FieldSelect *newfselect;
3385  Node *arg;
3386 
3387  arg = eval_const_expressions_mutator((Node *) fselect->arg,
3388  context);
3389  if (arg && IsA(arg, Var) &&
3390  ((Var *) arg)->varattno == InvalidAttrNumber &&
3391  ((Var *) arg)->varlevelsup == 0)
3392  {
3393  if (rowtype_field_matches(((Var *) arg)->vartype,
3394  fselect->fieldnum,
3395  fselect->resulttype,
3396  fselect->resulttypmod,
3397  fselect->resultcollid))
3398  {
3399  Var *newvar;
3400 
3401  newvar = makeVar(((Var *) arg)->varno,
3402  fselect->fieldnum,
3403  fselect->resulttype,
3404  fselect->resulttypmod,
3405  fselect->resultcollid,
3406  ((Var *) arg)->varlevelsup);
3407  /* New Var is nullable by same rels as the old one */
3408  newvar->varnullingrels = ((Var *) arg)->varnullingrels;
3409  return (Node *) newvar;
3410  }
3411  }
3412  if (arg && IsA(arg, RowExpr))
3413  {
3414  RowExpr *rowexpr = (RowExpr *) arg;
3415 
3416  if (fselect->fieldnum > 0 &&
3417  fselect->fieldnum <= list_length(rowexpr->args))
3418  {
3419  Node *fld = (Node *) list_nth(rowexpr->args,
3420  fselect->fieldnum - 1);
3421 
3422  if (rowtype_field_matches(rowexpr->row_typeid,
3423  fselect->fieldnum,
3424  fselect->resulttype,
3425  fselect->resulttypmod,
3426  fselect->resultcollid) &&
3427  fselect->resulttype == exprType(fld) &&
3428  fselect->resulttypmod == exprTypmod(fld) &&
3429  fselect->resultcollid == exprCollation(fld))
3430  return fld;
3431  }
3432  }
3433  newfselect = makeNode(FieldSelect);
3434  newfselect->arg = (Expr *) arg;
3435  newfselect->fieldnum = fselect->fieldnum;
3436  newfselect->resulttype = fselect->resulttype;
3437  newfselect->resulttypmod = fselect->resulttypmod;
3438  newfselect->resultcollid = fselect->resultcollid;
3439  if (arg && IsA(arg, Const))
3440  {
3441  Const *con = (Const *) arg;
3442 
3444  newfselect->fieldnum,
3445  newfselect->resulttype,
3446  newfselect->resulttypmod,
3447  newfselect->resultcollid))
3448  return ece_evaluate_expr(newfselect);
3449  }
3450  return (Node *) newfselect;
3451  }
3452  case T_NullTest:
3453  {
3454  NullTest *ntest = (NullTest *) node;
3455  NullTest *newntest;
3456  Node *arg;
3457 
3459  context);
3460  if (ntest->argisrow && arg && IsA(arg, RowExpr))
3461  {
3462  /*
3463  * We break ROW(...) IS [NOT] NULL into separate tests on
3464  * its component fields. This form is usually more
3465  * efficient to evaluate, as well as being more amenable
3466  * to optimization.
3467  */
3468  RowExpr *rarg = (RowExpr *) arg;
3469  List *newargs = NIL;
3470  ListCell *l;
3471 
3472  foreach(l, rarg->args)
3473  {
3474  Node *relem = (Node *) lfirst(l);
3475 
3476  /*
3477  * A constant field refutes the whole NullTest if it's
3478  * of the wrong nullness; else we can discard it.
3479  */
3480  if (relem && IsA(relem, Const))
3481  {
3482  Const *carg = (Const *) relem;
3483 
3484  if (carg->constisnull ?
3485  (ntest->nulltesttype == IS_NOT_NULL) :
3486  (ntest->nulltesttype == IS_NULL))
3487  return makeBoolConst(false, false);
3488  continue;
3489  }
3490 
3491  /*
3492  * Else, make a scalar (argisrow == false) NullTest
3493  * for this field. Scalar semantics are required
3494  * because IS [NOT] NULL doesn't recurse; see comments
3495  * in ExecEvalRowNullInt().
3496  */
3497  newntest = makeNode(NullTest);
3498  newntest->arg = (Expr *) relem;
3499  newntest->nulltesttype = ntest->nulltesttype;
3500  newntest->argisrow = false;
3501  newntest->location = ntest->location;
3502  newargs = lappend(newargs, newntest);
3503  }
3504  /* If all the inputs were constants, result is TRUE */
3505  if (newargs == NIL)
3506  return makeBoolConst(true, false);
3507  /* If only one nonconst input, it's the result */
3508  if (list_length(newargs) == 1)
3509  return (Node *) linitial(newargs);
3510  /* Else we need an AND node */
3511  return (Node *) make_andclause(newargs);
3512  }
3513  if (!ntest->argisrow && arg && IsA(arg, Const))
3514  {
3515  Const *carg = (Const *) arg;
3516  bool result;
3517 
3518  switch (ntest->nulltesttype)
3519  {
3520  case IS_NULL:
3521  result = carg->constisnull;
3522  break;
3523  case IS_NOT_NULL:
3524  result = !carg->constisnull;
3525  break;
3526  default:
3527  elog(ERROR, "unrecognized nulltesttype: %d",
3528  (int) ntest->nulltesttype);
3529  result = false; /* keep compiler quiet */
3530  break;
3531  }
3532 
3533  return makeBoolConst(result, false);
3534  }
3535 
3536  newntest = makeNode(NullTest);
3537  newntest->arg = (Expr *) arg;
3538  newntest->nulltesttype = ntest->nulltesttype;
3539  newntest->argisrow = ntest->argisrow;
3540  newntest->location = ntest->location;
3541  return (Node *) newntest;
3542  }
3543  case T_BooleanTest:
3544  {
3545  /*
3546  * This case could be folded into the generic handling used
3547  * for ArrayExpr etc. But because the simplification logic is
3548  * so trivial, applying evaluate_expr() to perform it would be
3549  * a heavy overhead. BooleanTest is probably common enough to
3550  * justify keeping this bespoke implementation.
3551  */
3552  BooleanTest *btest = (BooleanTest *) node;
3553  BooleanTest *newbtest;
3554  Node *arg;
3555 
3557  context);
3558  if (arg && IsA(arg, Const))
3559  {
3560  Const *carg = (Const *) arg;
3561  bool result;
3562 
3563  switch (btest->booltesttype)
3564  {
3565  case IS_TRUE:
3566  result = (!carg->constisnull &&
3567  DatumGetBool(carg->constvalue));
3568  break;
3569  case IS_NOT_TRUE:
3570  result = (carg->constisnull ||
3571  !DatumGetBool(carg->constvalue));
3572  break;
3573  case IS_FALSE:
3574  result = (!carg->constisnull &&
3575  !DatumGetBool(carg->constvalue));
3576  break;
3577  case IS_NOT_FALSE:
3578  result = (carg->constisnull ||
3579  DatumGetBool(carg->constvalue));
3580  break;
3581  case IS_UNKNOWN:
3582  result = carg->constisnull;
3583  break;
3584  case IS_NOT_UNKNOWN:
3585  result = !carg->constisnull;
3586  break;
3587  default:
3588  elog(ERROR, "unrecognized booltesttype: %d",
3589  (int) btest->booltesttype);
3590  result = false; /* keep compiler quiet */
3591  break;
3592  }
3593 
3594  return makeBoolConst(result, false);
3595  }
3596 
3597  newbtest = makeNode(BooleanTest);
3598  newbtest->arg = (Expr *) arg;
3599  newbtest->booltesttype = btest->booltesttype;
3600  newbtest->location = btest->location;
3601  return (Node *) newbtest;
3602  }
3603  case T_CoerceToDomain:
3604  {
3605  /*
3606  * If the domain currently has no constraints, we replace the
3607  * CoerceToDomain node with a simple RelabelType, which is
3608  * both far faster to execute and more amenable to later
3609  * optimization. We must then mark the plan as needing to be
3610  * rebuilt if the domain's constraints change.
3611  *
3612  * Also, in estimation mode, always replace CoerceToDomain
3613  * nodes, effectively assuming that the coercion will succeed.
3614  */
3615  CoerceToDomain *cdomain = (CoerceToDomain *) node;
3616  CoerceToDomain *newcdomain;
3617  Node *arg;
3618 
3619  arg = eval_const_expressions_mutator((Node *) cdomain->arg,
3620  context);
3621  if (context->estimate ||
3622  !DomainHasConstraints(cdomain->resulttype))
3623  {
3624  /* Record dependency, if this isn't estimation mode */
3625  if (context->root && !context->estimate)
3627  cdomain->resulttype);
3628 
3629  /* Generate RelabelType to substitute for CoerceToDomain */
3630  return applyRelabelType(arg,
3631  cdomain->resulttype,
3632  cdomain->resulttypmod,
3633  cdomain->resultcollid,
3634  cdomain->coercionformat,
3635  cdomain->location,
3636  true);
3637  }
3638 
3639  newcdomain = makeNode(CoerceToDomain);
3640  newcdomain->arg = (Expr *) arg;
3641  newcdomain->resulttype = cdomain->resulttype;
3642  newcdomain->resulttypmod = cdomain->resulttypmod;
3643  newcdomain->resultcollid = cdomain->resultcollid;
3644  newcdomain->coercionformat = cdomain->coercionformat;
3645  newcdomain->location = cdomain->location;
3646  return (Node *) newcdomain;
3647  }
3648  case T_PlaceHolderVar:
3649 
3650  /*
3651  * In estimation mode, just strip the PlaceHolderVar node
3652  * altogether; this amounts to estimating that the contained value
3653  * won't be forced to null by an outer join. In regular mode we
3654  * just use the default behavior (ie, simplify the expression but
3655  * leave the PlaceHolderVar node intact).
3656  */
3657  if (context->estimate)
3658  {
3659  PlaceHolderVar *phv = (PlaceHolderVar *) node;
3660 
3661  return eval_const_expressions_mutator((Node *) phv->phexpr,
3662  context);
3663  }
3664  break;
3665  case T_ConvertRowtypeExpr:
3666  {
3668  Node *arg;
3669  ConvertRowtypeExpr *newcre;
3670 
3672  context);
3673 
3674  newcre = makeNode(ConvertRowtypeExpr);
3675  newcre->resulttype = cre->resulttype;
3676  newcre->convertformat = cre->convertformat;
3677  newcre->location = cre->location;
3678 
3679  /*
3680  * In case of a nested ConvertRowtypeExpr, we can convert the
3681  * leaf row directly to the topmost row format without any
3682  * intermediate conversions. (This works because
3683  * ConvertRowtypeExpr is used only for child->parent
3684  * conversion in inheritance trees, which works by exact match
3685  * of column name, and a column absent in an intermediate
3686  * result can't be present in the final result.)
3687  *
3688  * No need to check more than one level deep, because the
3689  * above recursion will have flattened anything else.
3690  */
3691  if (arg != NULL && IsA(arg, ConvertRowtypeExpr))
3692  {
3694 
3695  arg = (Node *) argcre->arg;
3696 
3697  /*
3698  * Make sure an outer implicit conversion can't hide an
3699  * inner explicit one.
3700  */
3701  if (newcre->convertformat == COERCE_IMPLICIT_CAST)
3702  newcre->convertformat = argcre->convertformat;
3703  }
3704 
3705  newcre->arg = (Expr *) arg;
3706 
3707  if (arg != NULL && IsA(arg, Const))
3708  return ece_evaluate_expr((Node *) newcre);
3709  return (Node *) newcre;
3710  }
3711  default:
3712  break;
3713  }
3714 
3715  /*
3716  * For any node type not handled above, copy the node unchanged but
3717  * const-simplify its subexpressions. This is the correct thing for node
3718  * types whose behavior might change between planning and execution, such
3719  * as CurrentOfExpr. It's also a safe default for new node types not
3720  * known to this routine.
3721  */
3722  return ece_generic_processing(node);
3723 }
3724 
3725 /*
3726  * Subroutine for eval_const_expressions: check for non-Const nodes.
3727  *
3728  * We can abort recursion immediately on finding a non-Const node. This is
3729  * critical for performance, else eval_const_expressions_mutator would take
3730  * O(N^2) time on non-simplifiable trees. However, we do need to descend
3731  * into List nodes since expression_tree_walker sometimes invokes the walker
3732  * function directly on List subtrees.
3733  */
3734 static bool
3736 {
3737  if (node == NULL)
3738  return false;
3739  if (IsA(node, Const))
3740  return false;
3741  if (IsA(node, List))
3743  /* Otherwise, abort the tree traversal and return true */
3744  return true;
3745 }
3746 
3747 /*
3748  * Subroutine for eval_const_expressions: check if a function is OK to evaluate
3749  */
3750 static bool
3752 {
3753  char provolatile = func_volatile(funcid);
3754 
3755  /*
3756  * Ordinarily we are only allowed to simplify immutable functions. But for
3757  * purposes of estimation, we consider it okay to simplify functions that
3758  * are merely stable; the risk that the result might change from planning
3759  * time to execution time is worth taking in preference to not being able
3760  * to estimate the value at all.
3761  */
3762  if (provolatile == PROVOLATILE_IMMUTABLE)
3763  return true;
3764  if (context->estimate && provolatile == PROVOLATILE_STABLE)
3765  return true;
3766  return false;
3767 }
3768 
3769 /*
3770  * Subroutine for eval_const_expressions: process arguments of an OR clause
3771  *
3772  * This includes flattening of nested ORs as well as recursion to
3773  * eval_const_expressions to simplify the OR arguments.
3774  *
3775  * After simplification, OR arguments are handled as follows:
3776  * non constant: keep
3777  * FALSE: drop (does not affect result)
3778  * TRUE: force result to TRUE
3779  * NULL: keep only one
3780  * We must keep one NULL input because OR expressions evaluate to NULL when no
3781  * input is TRUE and at least one is NULL. We don't actually include the NULL
3782  * here, that's supposed to be done by the caller.
3783  *
3784  * The output arguments *haveNull and *forceTrue must be initialized false
3785  * by the caller. They will be set true if a NULL constant or TRUE constant,
3786  * respectively, is detected anywhere in the argument list.
3787  */
3788 static List *
3791  bool *haveNull, bool *forceTrue)
3792 {
3793  List *newargs = NIL;
3794  List *unprocessed_args;
3795 
3796  /*
3797  * We want to ensure that any OR immediately beneath another OR gets
3798  * flattened into a single OR-list, so as to simplify later reasoning.
3799  *
3800  * To avoid stack overflow from recursion of eval_const_expressions, we
3801  * resort to some tenseness here: we keep a list of not-yet-processed
3802  * inputs, and handle flattening of nested ORs by prepending to the to-do
3803  * list instead of recursing. Now that the parser generates N-argument
3804  * ORs from simple lists, this complexity is probably less necessary than
3805  * it once was, but we might as well keep the logic.
3806  */
3807  unprocessed_args = list_copy(args);
3808  while (unprocessed_args)
3809  {
3810  Node *arg = (Node *) linitial(unprocessed_args);
3811 
3812  unprocessed_args = list_delete_first(unprocessed_args);
3813 
3814  /* flatten nested ORs as per above comment */
3815  if (is_orclause(arg))
3816  {
3817  List *subargs = ((BoolExpr *) arg)->args;
3818  List *oldlist = unprocessed_args;
3819 
3820  unprocessed_args = list_concat_copy(subargs, unprocessed_args);
3821  /* perhaps-overly-tense code to avoid leaking old lists */
3822  list_free(oldlist);
3823  continue;
3824  }
3825 
3826  /* If it's not an OR, simplify it */
3828 
3829  /*
3830  * It is unlikely but not impossible for simplification of a non-OR
3831  * clause to produce an OR. Recheck, but don't be too tense about it
3832  * since it's not a mainstream case. In particular we don't worry
3833  * about const-simplifying the input twice, nor about list leakage.
3834  */
3835  if (is_orclause(arg))
3836  {
3837  List *subargs = ((BoolExpr *) arg)->args;
3838 
3839  unprocessed_args = list_concat_copy(subargs, unprocessed_args);
3840  continue;
3841  }
3842 
3843  /*
3844  * OK, we have a const-simplified non-OR argument. Process it per
3845  * comments above.
3846  */
3847  if (IsA(arg, Const))
3848  {
3849  Const *const_input = (Const *) arg;
3850 
3851  if (const_input->constisnull)
3852  *haveNull = true;
3853  else if (DatumGetBool(const_input->constvalue))
3854  {
3855  *forceTrue = true;
3856 
3857  /*
3858  * Once we detect a TRUE result we can just exit the loop
3859  * immediately. However, if we ever add a notion of
3860  * non-removable functions, we'd need to keep scanning.
3861  */
3862  return NIL;
3863  }
3864  /* otherwise, we can drop the constant-false input */
3865  continue;
3866  }
3867 
3868  /* else emit the simplified arg into the result list */
3869  newargs = lappend(newargs, arg);
3870  }
3871 
3872  return newargs;
3873 }
3874 
3875 /*
3876  * Subroutine for eval_const_expressions: process arguments of an AND clause
3877  *
3878  * This includes flattening of nested ANDs as well as recursion to
3879  * eval_const_expressions to simplify the AND arguments.
3880  *
3881  * After simplification, AND arguments are handled as follows:
3882  * non constant: keep
3883  * TRUE: drop (does not affect result)
3884  * FALSE: force result to FALSE
3885  * NULL: keep only one
3886  * We must keep one NULL input because AND expressions evaluate to NULL when
3887  * no input is FALSE and at least one is NULL. We don't actually include the
3888  * NULL here, that's supposed to be done by the caller.
3889  *
3890  * The output arguments *haveNull and *forceFalse must be initialized false
3891  * by the caller. They will be set true if a null constant or false constant,
3892  * respectively, is detected anywhere in the argument list.
3893  */
3894 static List *
3897  bool *haveNull, bool *forceFalse)
3898 {
3899  List *newargs = NIL;
3900  List *unprocessed_args;
3901 
3902  /* See comments in simplify_or_arguments */
3903  unprocessed_args = list_copy(args);
3904  while (unprocessed_args)
3905  {
3906  Node *arg = (Node *) linitial(unprocessed_args);
3907 
3908  unprocessed_args = list_delete_first(unprocessed_args);
3909 
3910  /* flatten nested ANDs as per above comment */
3911  if (is_andclause(arg))
3912  {
3913  List *subargs = ((BoolExpr *) arg)->args;
3914  List *oldlist = unprocessed_args;
3915 
3916  unprocessed_args = list_concat_copy(subargs, unprocessed_args);
3917  /* perhaps-overly-tense code to avoid leaking old lists */
3918  list_free(oldlist);
3919  continue;
3920  }
3921 
3922  /* If it's not an AND, simplify it */
3924 
3925  /*
3926  * It is unlikely but not impossible for simplification of a non-AND
3927  * clause to produce an AND. Recheck, but don't be too tense about it
3928  * since it's not a mainstream case. In particular we don't worry
3929  * about const-simplifying the input twice, nor about list leakage.
3930  */
3931  if (is_andclause(arg))
3932  {
3933  List *subargs = ((BoolExpr *) arg)->args;
3934 
3935  unprocessed_args = list_concat_copy(subargs, unprocessed_args);
3936  continue;
3937  }
3938 
3939  /*
3940  * OK, we have a const-simplified non-AND argument. Process it per
3941  * comments above.
3942  */
3943  if (IsA(arg, Const))
3944  {
3945  Const *const_input = (Const *) arg;
3946 
3947  if (const_input->constisnull)
3948  *haveNull = true;
3949  else if (!DatumGetBool(const_input->constvalue))
3950  {
3951  *forceFalse = true;
3952 
3953  /*
3954  * Once we detect a FALSE result we can just exit the loop
3955  * immediately. However, if we ever add a notion of
3956  * non-removable functions, we'd need to keep scanning.
3957  */
3958  return NIL;
3959  }
3960  /* otherwise, we can drop the constant-true input */
3961  continue;
3962  }
3963 
3964  /* else emit the simplified arg into the result list */
3965  newargs = lappend(newargs, arg);
3966  }
3967 
3968  return newargs;
3969 }
3970 
3971 /*
3972  * Subroutine for eval_const_expressions: try to simplify boolean equality
3973  * or inequality condition
3974  *
3975  * Inputs are the operator OID and the simplified arguments to the operator.
3976  * Returns a simplified expression if successful, or NULL if cannot
3977  * simplify the expression.
3978  *
3979  * The idea here is to reduce "x = true" to "x" and "x = false" to "NOT x",
3980  * or similarly "x <> true" to "NOT x" and "x <> false" to "x".
3981  * This is only marginally useful in itself, but doing it in constant folding
3982  * ensures that we will recognize these forms as being equivalent in, for
3983  * example, partial index matching.
3984  *
3985  * We come here only if simplify_function has failed; therefore we cannot
3986  * see two constant inputs, nor a constant-NULL input.
3987  */
3988 static Node *
3990 {
3991  Node *leftop;
3992  Node *rightop;
3993 
3994  Assert(list_length(args) == 2);
3995  leftop = linitial(args);
3996  rightop = lsecond(args);
3997  if (leftop && IsA(leftop, Const))
3998  {
3999  Assert(!((Const *) leftop)->constisnull);
4000  if (opno == BooleanEqualOperator)
4001  {
4002  if (DatumGetBool(((Const *) leftop)->constvalue))
4003  return rightop; /* true = foo */
4004  else
4005  return negate_clause(rightop); /* false = foo */
4006  }
4007  else
4008  {
4009  if (DatumGetBool(((Const *) leftop)->constvalue))
4010  return negate_clause(rightop); /* true <> foo */
4011  else
4012  return rightop; /* false <> foo */
4013  }
4014  }
4015  if (rightop && IsA(rightop, Const))
4016  {
4017  Assert(!((Const *) rightop)->constisnull);
4018  if (opno == BooleanEqualOperator)
4019  {
4020  if (DatumGetBool(((Const *) rightop)->constvalue))
4021  return leftop; /* foo = true */
4022  else
4023  return negate_clause(leftop); /* foo = false */
4024  }
4025  else
4026  {
4027  if (DatumGetBool(((Const *) rightop)->constvalue))
4028  return negate_clause(leftop); /* foo <> true */
4029  else
4030  return leftop; /* foo <> false */
4031  }
4032  }
4033  return NULL;
4034 }
4035 
4036 /*
4037  * Subroutine for eval_const_expressions: try to simplify a function call
4038  * (which might originally have been an operator; we don't care)
4039  *
4040  * Inputs are the function OID, actual result type OID (which is needed for
4041  * polymorphic functions), result typmod, result collation, the input
4042  * collation to use for the function, the original argument list (not
4043  * const-simplified yet, unless process_args is false), and some flags;
4044  * also the context data for eval_const_expressions.
4045  *
4046  * Returns a simplified expression if successful, or NULL if cannot
4047  * simplify the function call.
4048  *
4049  * This function is also responsible for converting named-notation argument
4050  * lists into positional notation and/or adding any needed default argument
4051  * expressions; which is a bit grotty, but it avoids extra fetches of the
4052  * function's pg_proc tuple. For this reason, the args list is
4053  * pass-by-reference. Conversion and const-simplification of the args list
4054  * will be done even if simplification of the function call itself is not
4055  * possible.
4056  */
4057 static Expr *
4058 simplify_function(Oid funcid, Oid result_type, int32 result_typmod,
4059  Oid result_collid, Oid input_collid, List **args_p,
4060  bool funcvariadic, bool process_args, bool allow_non_const,
4062 {
4063  List *args = *args_p;
4064  HeapTuple func_tuple;
4065  Form_pg_proc func_form;
4066  Expr *newexpr;
4067 
4068  /*
4069  * We have three strategies for simplification: execute the function to
4070  * deliver a constant result, use a transform function to generate a
4071  * substitute node tree, or expand in-line the body of the function
4072  * definition (which only works for simple SQL-language functions, but
4073  * that is a common case). Each case needs access to the function's
4074  * pg_proc tuple, so fetch it just once.
4075  *
4076  * Note: the allow_non_const flag suppresses both the second and third
4077  * strategies; so if !allow_non_const, simplify_function can only return a
4078  * Const or NULL. Argument-list rewriting happens anyway, though.
4079  */
4080  func_tuple = SearchSysCache1(PROCOID, ObjectIdGetDatum(funcid));
4081  if (!HeapTupleIsValid(func_tuple))
4082  elog(ERROR, "cache lookup failed for function %u", funcid);
4083  func_form = (Form_pg_proc) GETSTRUCT(func_tuple);
4084 
4085  /*
4086  * Process the function arguments, unless the caller did it already.
4087  *
4088  * Here we must deal with named or defaulted arguments, and then
4089  * recursively apply eval_const_expressions to the whole argument list.
4090  */
4091  if (process_args)
4092  {
4093  args = expand_function_arguments(args, false, result_type, func_tuple);
4096  context);
4097  /* Argument processing done, give it back to the caller */
4098  *args_p = args;
4099  }
4100 
4101  /* Now attempt simplification of the function call proper. */
4102 
4103  newexpr = evaluate_function(funcid, result_type, result_typmod,
4104  result_collid, input_collid,
4105  args, funcvariadic,
4106  func_tuple, context);
4107 
4108  if (!newexpr && allow_non_const && OidIsValid(func_form->prosupport))
4109  {
4110  /*
4111  * Build a SupportRequestSimplify node to pass to the support
4112  * function, pointing to a dummy FuncExpr node containing the
4113  * simplified arg list. We use this approach to present a uniform
4114  * interface to the support function regardless of how the target
4115  * function is actually being invoked.
4116  */
4118  FuncExpr fexpr;
4119 
4120  fexpr.xpr.type = T_FuncExpr;
4121  fexpr.funcid = funcid;
4122  fexpr.funcresulttype = result_type;
4123  fexpr.funcretset = func_form->proretset;
4124  fexpr.funcvariadic = funcvariadic;
4125  fexpr.funcformat = COERCE_EXPLICIT_CALL;
4126  fexpr.funccollid = result_collid;
4127  fexpr.inputcollid = input_collid;
4128  fexpr.args = args;
4129  fexpr.location = -1;
4130 
4131  req.type = T_SupportRequestSimplify;
4132  req.root = context->root;
4133  req.fcall = &fexpr;
4134 
4135  newexpr = (Expr *)
4136  DatumGetPointer(OidFunctionCall1(func_form->prosupport,
4137  PointerGetDatum(&req)));
4138 
4139  /* catch a possible API misunderstanding */
4140  Assert(newexpr != (Expr *) &fexpr);
4141  }
4142 
4143  if (!newexpr && allow_non_const)
4144  newexpr = inline_function(funcid, result_type, result_collid,
4145  input_collid, args, funcvariadic,
4146  func_tuple, context);
4147 
4148  ReleaseSysCache(func_tuple);
4149 
4150  return newexpr;
4151 }
4152 
4153 /*
4154  * expand_function_arguments: convert named-notation args to positional args
4155  * and/or insert default args, as needed
4156  *
4157  * Returns a possibly-transformed version of the args list.
4158  *
4159  * If include_out_arguments is true, then the args list and the result
4160  * include OUT arguments.
4161  *
4162  * The expected result type of the call must be given, for sanity-checking
4163  * purposes. Also, we ask the caller to provide the function's actual
4164  * pg_proc tuple, not just its OID.
4165  *
4166  * If we need to change anything, the input argument list is copied, not
4167  * modified.
4168  *
4169  * Note: this gets applied to operator argument lists too, even though the
4170  * cases it handles should never occur there. This should be OK since it
4171  * will fall through very quickly if there's nothing to do.
4172  */
4173 List *
4174 expand_function_arguments(List *args, bool include_out_arguments,
4175  Oid result_type, HeapTuple func_tuple)
4176 {
4177  Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4178  Oid *proargtypes = funcform->proargtypes.values;
4179  int pronargs = funcform->pronargs;
4180  bool has_named_args = false;
4181  ListCell *lc;
4182 
4183  /*
4184  * If we are asked to match to OUT arguments, then use the proallargtypes
4185  * array (which includes those); otherwise use proargtypes (which
4186  * doesn't). Of course, if proallargtypes is null, we always use
4187  * proargtypes. (Fetching proallargtypes is annoyingly expensive
4188  * considering that we may have nothing to do here, but fortunately the
4189  * common case is include_out_arguments == false.)
4190  */
4191  if (include_out_arguments)
4192  {
4193  Datum proallargtypes;
4194  bool isNull;
4195 
4196  proallargtypes = SysCacheGetAttr(PROCOID, func_tuple,
4197  Anum_pg_proc_proallargtypes,
4198  &isNull);
4199  if (!isNull)
4200  {
4201  ArrayType *arr = DatumGetArrayTypeP(proallargtypes);
4202 
4203  pronargs = ARR_DIMS(arr)[0];
4204  if (ARR_NDIM(arr) != 1 ||
4205  pronargs < 0 ||
4206  ARR_HASNULL(arr) ||
4207  ARR_ELEMTYPE(arr) != OIDOID)
4208  elog(ERROR, "proallargtypes is not a 1-D Oid array or it contains nulls");
4209  Assert(pronargs >= funcform->pronargs);
4210  proargtypes = (Oid *) ARR_DATA_PTR(arr);
4211  }
4212  }
4213 
4214  /* Do we have any named arguments? */
4215  foreach(lc, args)
4216  {
4217  Node *arg = (Node *) lfirst(lc);
4218 
4219  if (IsA(arg, NamedArgExpr))
4220  {
4221  has_named_args = true;
4222  break;
4223  }
4224  }
4225 
4226  /* If so, we must apply reorder_function_arguments */
4227  if (has_named_args)
4228  {
4229  args = reorder_function_arguments(args, pronargs, func_tuple);
4230  /* Recheck argument types and add casts if needed */
4231  recheck_cast_function_args(args, result_type,
4232  proargtypes, pronargs,
4233  func_tuple);
4234  }
4235  else if (list_length(args) < pronargs)
4236  {
4237  /* No named args, but we seem to be short some defaults */
4238  args = add_function_defaults(args, pronargs, func_tuple);
4239  /* Recheck argument types and add casts if needed */
4240  recheck_cast_function_args(args, result_type,
4241  proargtypes, pronargs,
4242  func_tuple);
4243  }
4244 
4245  return args;
4246 }
4247 
4248 /*
4249  * reorder_function_arguments: convert named-notation args to positional args
4250  *
4251  * This function also inserts default argument values as needed, since it's
4252  * impossible to form a truly valid positional call without that.
4253  */
4254 static List *
4256 {
4257  Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4258  int nargsprovided = list_length(args);
4259  Node *argarray[FUNC_MAX_ARGS];
4260  ListCell *lc;
4261  int i;
4262 
4263  Assert(nargsprovided <= pronargs);
4264  if (pronargs < 0 || pronargs > FUNC_MAX_ARGS)
4265  elog(ERROR, "too many function arguments");
4266  memset(argarray, 0, pronargs * sizeof(Node *));
4267 
4268  /* Deconstruct the argument list into an array indexed by argnumber */
4269  i = 0;
4270  foreach(lc, args)
4271  {
4272  Node *arg = (Node *) lfirst(lc);
4273 
4274  if (!IsA(arg, NamedArgExpr))
4275  {
4276  /* positional argument, assumed to precede all named args */
4277  Assert(argarray[i] == NULL);
4278  argarray[i++] = arg;
4279  }
4280  else
4281  {
4282  NamedArgExpr *na = (NamedArgExpr *) arg;
4283 
4284  Assert(na->argnumber >= 0 && na->argnumber < pronargs);
4285  Assert(argarray[na->argnumber] == NULL);
4286  argarray[na->argnumber] = (Node *) na->arg;
4287  }
4288  }
4289 
4290  /*
4291  * Fetch default expressions, if needed, and insert into array at proper
4292  * locations (they aren't necessarily consecutive or all used)
4293  */
4294  if (nargsprovided < pronargs)
4295  {
4296  List *defaults = fetch_function_defaults(func_tuple);
4297 
4298  i = pronargs - funcform->pronargdefaults;
4299  foreach(lc, defaults)
4300  {
4301  if (argarray[i] == NULL)
4302  argarray[i] = (Node *) lfirst(lc);
4303  i++;
4304  }
4305  }
4306 
4307  /* Now reconstruct the args list in proper order */
4308  args = NIL;
4309  for (i = 0; i < pronargs; i++)
4310  {
4311  Assert(argarray[i] != NULL);
4312  args = lappend(args, argarray[i]);
4313  }
4314 
4315  return args;
4316 }
4317 
4318 /*
4319  * add_function_defaults: add missing function arguments from its defaults
4320  *
4321  * This is used only when the argument list was positional to begin with,
4322  * and so we know we just need to add defaults at the end.
4323  */
4324 static List *
4326 {
4327  int nargsprovided = list_length(args);
4328  List *defaults;
4329  int ndelete;
4330 
4331  /* Get all the default expressions from the pg_proc tuple */
4332  defaults = fetch_function_defaults(func_tuple);
4333 
4334  /* Delete any unused defaults from the list */
4335  ndelete = nargsprovided + list_length(defaults) - pronargs;
4336  if (ndelete < 0)
4337  elog(ERROR, "not enough default arguments");
4338  if (ndelete > 0)
4339  defaults = list_delete_first_n(defaults, ndelete);
4340 
4341  /* And form the combined argument list, not modifying the input list */
4342  return list_concat_copy(args, defaults);
4343 }
4344 
4345 /*
4346  * fetch_function_defaults: get function's default arguments as expression list
4347  */
4348 static List *
4350 {
4351  List *defaults;
4352  Datum proargdefaults;
4353  char *str;
4354 
4355  proargdefaults = SysCacheGetAttrNotNull(PROCOID, func_tuple,
4356  Anum_pg_proc_proargdefaults);
4357  str = TextDatumGetCString(proargdefaults);
4358  defaults = castNode(List, stringToNode(str));
4359  pfree(str);
4360  return defaults;
4361 }
4362 
4363 /*
4364  * recheck_cast_function_args: recheck function args and typecast as needed
4365  * after adding defaults.
4366  *
4367  * It is possible for some of the defaulted arguments to be polymorphic;
4368  * therefore we can't assume that the default expressions have the correct
4369  * data types already. We have to re-resolve polymorphics and do coercion
4370  * just like the parser did.
4371  *
4372  * This should be a no-op if there are no polymorphic arguments,
4373  * but we do it anyway to be sure.
4374  *
4375  * Note: if any casts are needed, the args list is modified in-place;
4376  * caller should have already copied the list structure.
4377  */
4378 static void
4380  Oid *proargtypes, int pronargs,
4381  HeapTuple func_tuple)
4382 {
4383  Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4384  int nargs;
4385  Oid actual_arg_types[FUNC_MAX_ARGS];
4386  Oid declared_arg_types[FUNC_MAX_ARGS];
4387  Oid rettype;
4388  ListCell *lc;
4389 
4391  elog(ERROR, "too many function arguments");
4392  nargs = 0;
4393  foreach(lc, args)
4394  {
4395  actual_arg_types[nargs++] = exprType((Node *) lfirst(lc));
4396  }
4397  Assert(nargs == pronargs);
4398  memcpy(declared_arg_types, proargtypes, pronargs * sizeof(Oid));
4399  rettype = enforce_generic_type_consistency(actual_arg_types,
4400  declared_arg_types,
4401  nargs,
4402  funcform->prorettype,
4403  false);
4404  /* let's just check we got the same answer as the parser did ... */
4405  if (rettype != result_type)
4406  elog(ERROR, "function's resolved result type changed during planning");
4407 
4408  /* perform any necessary typecasting of arguments */
4409  make_fn_arguments(NULL, args, actual_arg_types, declared_arg_types);
4410 }
4411 
4412 /*
4413  * evaluate_function: try to pre-evaluate a function call
4414  *
4415  * We can do this if the function is strict and has any constant-null inputs
4416  * (just return a null constant), or if the function is immutable and has all
4417  * constant inputs (call it and return the result as a Const node). In
4418  * estimation mode we are willing to pre-evaluate stable functions too.
4419  *
4420  * Returns a simplified expression if successful, or NULL if cannot
4421  * simplify the function.
4422  */
4423 static Expr *
4424 evaluate_function(Oid funcid, Oid result_type, int32 result_typmod,
4425  Oid result_collid, Oid input_collid, List *args,
4426  bool funcvariadic,
4427  HeapTuple func_tuple,
4429 {
4430  Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4431  bool has_nonconst_input = false;
4432  bool has_null_input = false;
4433  ListCell *arg;
4434  FuncExpr *newexpr;
4435 
4436  /*
4437  * Can't simplify if it returns a set.
4438  */
4439  if (funcform->proretset)
4440  return NULL;
4441 
4442  /*
4443  * Can't simplify if it returns RECORD. The immediate problem is that it
4444  * will be needing an expected tupdesc which we can't supply here.
4445  *
4446  * In the case where it has OUT parameters, we could build an expected
4447  * tupdesc from those, but there may be other gotchas lurking. In
4448  * particular, if the function were to return NULL, we would produce a
4449  * null constant with no remaining indication of which concrete record
4450  * type it is. For now, seems best to leave the function call unreduced.
4451  */
4452  if (funcform->prorettype == RECORDOID)
4453  return NULL;
4454 
4455  /*
4456  * Check for constant inputs and especially constant-NULL inputs.
4457  */
4458  foreach(arg, args)
4459  {
4460  if (IsA(lfirst(arg), Const))
4461  has_null_input |= ((Const *) lfirst(arg))->constisnull;
4462  else
4463  has_nonconst_input = true;
4464  }
4465 
4466  /*
4467  * If the function is strict and has a constant-NULL input, it will never
4468  * be called at all, so we can replace the call by a NULL constant, even
4469  * if there are other inputs that aren't constant, and even if the
4470  * function is not otherwise immutable.
4471  */
4472  if (funcform->proisstrict && has_null_input)
4473  return (Expr *) makeNullConst(result_type, result_typmod,
4474  result_collid);
4475 
4476  /*
4477  * Otherwise, can simplify only if all inputs are constants. (For a
4478  * non-strict function, constant NULL inputs are treated the same as
4479  * constant non-NULL inputs.)
4480  */
4481  if (has_nonconst_input)
4482  return NULL;
4483 
4484  /*
4485  * Ordinarily we are only allowed to simplify immutable functions. But for
4486  * purposes of estimation, we consider it okay to simplify functions that
4487  * are merely stable; the risk that the result might change from planning
4488  * time to execution time is worth taking in preference to not being able
4489  * to estimate the value at all.
4490  */
4491  if (funcform->provolatile == PROVOLATILE_IMMUTABLE)
4492  /* okay */ ;
4493  else if (context->estimate && funcform->provolatile == PROVOLATILE_STABLE)
4494  /* okay */ ;
4495  else
4496  return NULL;
4497 
4498  /*
4499  * OK, looks like we can simplify this operator/function.
4500  *
4501  * Build a new FuncExpr node containing the already-simplified arguments.
4502  */
4503  newexpr = makeNode(FuncExpr);
4504  newexpr->funcid = funcid;
4505  newexpr->funcresulttype = result_type;
4506  newexpr->funcretset = false;
4507  newexpr->funcvariadic = funcvariadic;
4508  newexpr->funcformat = COERCE_EXPLICIT_CALL; /* doesn't matter */
4509  newexpr->funccollid = result_collid; /* doesn't matter */
4510  newexpr->inputcollid = input_collid;
4511  newexpr->args = args;
4512  newexpr->location = -1;
4513 
4514  return evaluate_expr((Expr *) newexpr, result_type, result_typmod,
4515  result_collid);
4516 }
4517 
4518 /*
4519  * inline_function: try to expand a function call inline
4520  *
4521  * If the function is a sufficiently simple SQL-language function
4522  * (just "SELECT expression"), then we can inline it and avoid the rather
4523  * high per-call overhead of SQL functions. Furthermore, this can expose
4524  * opportunities for constant-folding within the function expression.
4525  *
4526  * We have to beware of some special cases however. A directly or
4527  * indirectly recursive function would cause us to recurse forever,
4528  * so we keep track of which functions we are already expanding and
4529  * do not re-expand them. Also, if a parameter is used more than once
4530  * in the SQL-function body, we require it not to contain any volatile
4531  * functions (volatiles might deliver inconsistent answers) nor to be
4532  * unreasonably expensive to evaluate. The expensiveness check not only
4533  * prevents us from doing multiple evaluations of an expensive parameter
4534  * at runtime, but is a safety value to limit growth of an expression due
4535  * to repeated inlining.
4536  *
4537  * We must also beware of changing the volatility or strictness status of
4538  * functions by inlining them.
4539  *
4540  * Also, at the moment we can't inline functions returning RECORD. This
4541  * doesn't work in the general case because it discards information such
4542  * as OUT-parameter declarations.
4543  *
4544  * Also, context-dependent expression nodes in the argument list are trouble.
4545  *
4546  * Returns a simplified expression if successful, or NULL if cannot
4547  * simplify the function.
4548  */
4549 static Expr *
4550 inline_function(Oid funcid, Oid result_type, Oid result_collid,
4551  Oid input_collid, List *args,
4552  bool funcvariadic,
4553  HeapTuple func_tuple,
4555 {
4556  Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4557  char *src;
4558  Datum tmp;
4559  bool isNull;
4560  MemoryContext oldcxt;
4561  MemoryContext mycxt;
4562  inline_error_callback_arg callback_arg;
4563  ErrorContextCallback sqlerrcontext;
4564  FuncExpr *fexpr;
4566  TupleDesc rettupdesc;
4567  ParseState *pstate;
4568  List *raw_parsetree_list;
4569  List *querytree_list;
4570  Query *querytree;
4571  Node *newexpr;
4572  int *usecounts;
4573  ListCell *arg;
4574  int i;
4575 
4576  /*
4577  * Forget it if the function is not SQL-language or has other showstopper
4578  * properties. (The prokind and nargs checks are just paranoia.)
4579  */
4580  if (funcform->prolang != SQLlanguageId ||
4581  funcform->prokind != PROKIND_FUNCTION ||
4582  funcform->prosecdef ||
4583  funcform->proretset ||
4584  funcform->prorettype == RECORDOID ||
4585  !heap_attisnull(func_tuple, Anum_pg_proc_proconfig, NULL) ||
4586  funcform->pronargs != list_length(args))
4587  return NULL;
4588 
4589  /* Check for recursive function, and give up trying to expand if so */
4590  if (list_member_oid(context->active_fns, funcid))
4591  return NULL;
4592 
4593  /* Check permission to call function (fail later, if not) */
4594  if (object_aclcheck(ProcedureRelationId, funcid, GetUserId(), ACL_EXECUTE) != ACLCHECK_OK)
4595  return NULL;
4596 
4597  /* Check whether a plugin wants to hook function entry/exit */
4598  if (FmgrHookIsNeeded(funcid))
4599  return NULL;
4600 
4601  /*
4602  * Make a temporary memory context, so that we don't leak all the stuff
4603  * that parsing might create.
4604  */
4606  "inline_function",
4608  oldcxt = MemoryContextSwitchTo(mycxt);
4609 
4610  /*
4611  * We need a dummy FuncExpr node containing the already-simplified
4612  * arguments. (In some cases we don't really need it, but building it is
4613  * cheap enough that it's not worth contortions to avoid.)
4614  */
4615  fexpr = makeNode(FuncExpr);
4616  fexpr->funcid = funcid;
4617  fexpr->funcresulttype = result_type;
4618  fexpr->funcretset = false;
4619  fexpr->funcvariadic = funcvariadic;
4620  fexpr->funcformat = COERCE_EXPLICIT_CALL; /* doesn't matter */
4621  fexpr->funccollid = result_collid; /* doesn't matter */
4622  fexpr->inputcollid = input_collid;
4623  fexpr->args = args;
4624  fexpr->location = -1;
4625 
4626  /* Fetch the function body */
4627  tmp = SysCacheGetAttrNotNull(PROCOID, func_tuple, Anum_pg_proc_prosrc);
4628  src = TextDatumGetCString(tmp);
4629 
4630  /*
4631  * Setup error traceback support for ereport(). This is so that we can
4632  * finger the function that bad information came from.
4633  */
4634  callback_arg.proname = NameStr(funcform->proname);
4635  callback_arg.prosrc = src;
4636 
4637  sqlerrcontext.callback = sql_inline_error_callback;
4638  sqlerrcontext.arg = &callback_arg;
4639  sqlerrcontext.previous = error_context_stack;
4640  error_context_stack = &sqlerrcontext;
4641 
4642  /* If we have prosqlbody, pay attention to that not prosrc */
4643  tmp = SysCacheGetAttr(PROCOID,
4644  func_tuple,
4645  Anum_pg_proc_prosqlbody,
4646  &isNull);
4647  if (!isNull)
4648  {
4649  Node *n;
4650  List *query_list;
4651 
4653  if (IsA(n, List))
4654  query_list = linitial_node(List, castNode(List, n));
4655  else
4656  query_list = list_make1(n);
4657  if (list_length(query_list) != 1)
4658  goto fail;
4659  querytree = linitial(query_list);
4660 
4661  /*
4662  * Because we'll insist below that the querytree have an empty rtable
4663  * and no sublinks, it cannot have any relation references that need
4664  * to be locked or rewritten. So we can omit those steps.
4665  */
4666  }
4667  else
4668  {
4669  /* Set up to handle parameters while parsing the function body. */
4670  pinfo = prepare_sql_fn_parse_info(func_tuple,
4671  (Node *) fexpr,
4672  input_collid);
4673 
4674  /*
4675  * We just do parsing and parse analysis, not rewriting, because
4676  * rewriting will not affect table-free-SELECT-only queries, which is
4677  * all that we care about. Also, we can punt as soon as we detect
4678  * more than one command in the function body.
4679  */
4680  raw_parsetree_list = pg_parse_query(src);
4681  if (list_length(raw_parsetree_list) != 1)
4682  goto fail;
4683 
4684  pstate = make_parsestate(NULL);
4685  pstate->p_sourcetext = src;
4686  sql_fn_parser_setup(pstate, pinfo);
4687 
4688  querytree = transformTopLevelStmt(pstate, linitial(raw_parsetree_list));
4689 
4690  free_parsestate(pstate);
4691  }
4692 
4693  /*
4694  * The single command must be a simple "SELECT expression".
4695  *
4696  * Note: if you change the tests involved in this, see also plpgsql's
4697  * exec_simple_check_plan(). That generally needs to have the same idea
4698  * of what's a "simple expression", so that inlining a function that
4699  * previously wasn't inlined won't change plpgsql's conclusion.
4700  */
4701  if (!IsA(querytree, Query) ||
4702  querytree->commandType != CMD_SELECT ||
4703  querytree->hasAggs ||
4704  querytree->hasWindowFuncs ||
4705  querytree->hasTargetSRFs ||
4706  querytree->hasSubLinks ||
4707  querytree->cteList ||
4708  querytree->rtable ||
4709  querytree->jointree->fromlist ||
4710  querytree->jointree->quals ||
4711  querytree->groupClause ||
4712  querytree->groupingSets ||
4713  querytree->havingQual ||
4714  querytree->windowClause ||
4715  querytree->distinctClause ||
4716  querytree->sortClause ||
4717  querytree->limitOffset ||
4718  querytree->limitCount ||
4719  querytree->setOperations ||
4720  list_length(querytree->targetList) != 1)
4721  goto fail;
4722 
4723  /* If the function result is composite, resolve it */
4724  (void) get_expr_result_type((Node *) fexpr,
4725  NULL,
4726  &rettupdesc);
4727 
4728  /*
4729  * Make sure the function (still) returns what it's declared to. This
4730  * will raise an error if wrong, but that's okay since the function would
4731  * fail at runtime anyway. Note that check_sql_fn_retval will also insert
4732  * a coercion if needed to make the tlist expression match the declared
4733  * type of the function.
4734  *
4735  * Note: we do not try this until we have verified that no rewriting was
4736  * needed; that's probably not important, but let's be careful.
4737  */
4738  querytree_list = list_make1(querytree);
4739  if (check_sql_fn_retval(list_make1(querytree_list),
4740  result_type, rettupdesc,
4741  funcform->prokind,
4742  false, NULL))
4743  goto fail; /* reject whole-tuple-result cases */
4744 
4745  /*
4746  * Given the tests above, check_sql_fn_retval shouldn't have decided to
4747  * inject a projection step, but let's just make sure.
4748  */
4749  if (querytree != linitial(querytree_list))
4750  goto fail;
4751 
4752  /* Now we can grab the tlist expression */
4753  newexpr = (Node *) ((TargetEntry *) linitial(querytree->targetList))->expr;
4754 
4755  /*
4756  * If the SQL function returns VOID, we can only inline it if it is a
4757  * SELECT of an expression returning VOID (ie, it's just a redirection to
4758  * another VOID-returning function). In all non-VOID-returning cases,
4759  * check_sql_fn_retval should ensure that newexpr returns the function's
4760  * declared result type, so this test shouldn't fail otherwise; but we may
4761  * as well cope gracefully if it does.
4762  */
4763  if (exprType(newexpr) != result_type)
4764  goto fail;
4765 
4766  /*
4767  * Additional validity checks on the expression. It mustn't be more
4768  * volatile than the surrounding function (this is to avoid breaking hacks
4769  * that involve pretending a function is immutable when it really ain't).
4770  * If the surrounding function is declared strict, then the expression
4771  * must contain only strict constructs and must use all of the function
4772  * parameters (this is overkill, but an exact analysis is hard).
4773  */
4774  if (funcform->provolatile == PROVOLATILE_IMMUTABLE &&
4775  contain_mutable_functions(newexpr))
4776  goto fail;
4777  else if (funcform->provolatile == PROVOLATILE_STABLE &&
4778  contain_volatile_functions(newexpr))
4779  goto fail;
4780 
4781  if (funcform->proisstrict &&
4782  contain_nonstrict_functions(newexpr))
4783  goto fail;
4784 
4785  /*
4786  * If any parameter expression contains a context-dependent node, we can't
4787  * inline, for fear of putting such a node into the wrong context.
4788  */
4790  goto fail;
4791 
4792  /*
4793  * We may be able to do it; there are still checks on parameter usage to
4794  * make, but those are most easily done in combination with the actual
4795  * substitution of the inputs. So start building expression with inputs
4796  * substituted.
4797  */
4798  usecounts = (int *) palloc0(funcform->pronargs * sizeof(int));
4799  newexpr = substitute_actual_parameters(newexpr, funcform->pronargs,
4800  args, usecounts);
4801 
4802  /* Now check for parameter usage */
4803  i = 0;
4804  foreach(arg, args)
4805  {
4806  Node *param = lfirst(arg);
4807 
4808  if (usecounts[i] == 0)
4809  {
4810  /* Param not used at all: uncool if func is strict */
4811  if (funcform->proisstrict)
4812  goto fail;
4813  }
4814  else if (usecounts[i] != 1)
4815  {
4816  /* Param used multiple times: uncool if expensive or volatile */
4817  QualCost eval_cost;
4818 
4819  /*
4820  * We define "expensive" as "contains any subplan or more than 10
4821  * operators". Note that the subplan search has to be done
4822  * explicitly, since cost_qual_eval() will barf on unplanned
4823  * subselects.
4824  */
4825  if (contain_subplans(param))
4826  goto fail;
4827  cost_qual_eval(&eval_cost, list_make1(param), NULL);
4828  if (eval_cost.startup + eval_cost.per_tuple >
4829  10 * cpu_operator_cost)
4830  goto fail;
4831 
4832  /*
4833  * Check volatility last since this is more expensive than the
4834  * above tests
4835  */
4836  if (contain_volatile_functions(param))
4837  goto fail;
4838  }
4839  i++;
4840  }
4841 
4842  /*
4843  * Whew --- we can make the substitution. Copy the modified expression
4844  * out of the temporary memory context, and clean up.
4845  */
4846  MemoryContextSwitchTo(oldcxt);
4847 
4848  newexpr = copyObject(newexpr);
4849 
4850  MemoryContextDelete(mycxt);
4851 
4852  /*
4853  * If the result is of a collatable type, force the result to expose the
4854  * correct collation. In most cases this does not matter, but it's
4855  * possible that the function result is used directly as a sort key or in
4856  * other places where we expect exprCollation() to tell the truth.
4857  */
4858  if (OidIsValid(result_collid))
4859  {
4860  Oid exprcoll = exprCollation(newexpr);
4861 
4862  if (OidIsValid(exprcoll) && exprcoll != result_collid)
4863  {
4864  CollateExpr *newnode = makeNode(CollateExpr);
4865 
4866  newnode->arg = (Expr *) newexpr;
4867  newnode->collOid = result_collid;
4868  newnode->location = -1;
4869 
4870  newexpr = (Node *) newnode;
4871  }
4872  }
4873 
4874  /*
4875  * Since there is now no trace of the function in the plan tree, we must
4876  * explicitly record the plan's dependency on the function.
4877  */
4878  if (context->root)
4880 
4881  /*
4882  * Recursively try to simplify the modified expression. Here we must add
4883  * the current function to the context list of active functions.
4884  */
4885  context->active_fns = lappend_oid(context->active_fns, funcid);
4886  newexpr = eval_const_expressions_mutator(newexpr, context);
4887  context->active_fns = list_delete_last(context->active_fns);
4888 
4889  error_context_stack = sqlerrcontext.previous;
4890 
4891  return (Expr *) newexpr;
4892 
4893  /* Here if func is not inlinable: release temp memory and return NULL */
4894 fail:
4895  MemoryContextSwitchTo(oldcxt);
4896  MemoryContextDelete(mycxt);
4897  error_context_stack = sqlerrcontext.previous;
4898 
4899  return NULL;
4900 }
4901 
4902 /*
4903  * Replace Param nodes by appropriate actual parameters
4904  */
4905 static Node *
4907  int *usecounts)
4908 {
4910 
4911  context.nargs = nargs;
4912  context.args = args;
4913  context.usecounts = usecounts;
4914 
4916 }
4917 
4918 static Node *
4921 {
4922  if (node == NULL)
4923  return NULL;
4924  if (IsA(node, Param))
4925  {
4926  Param *param = (Param *) node;
4927 
4928  if (param->paramkind != PARAM_EXTERN)
4929  elog(ERROR, "unexpected paramkind: %d", (int) param->paramkind);
4930  if (param->paramid <= 0 || param->paramid > context->nargs)
4931  elog(ERROR, "invalid paramid: %d", param->paramid);
4932 
4933  /* Count usage of parameter */
4934  context->usecounts[param->paramid - 1]++;
4935 
4936  /* Select the appropriate actual arg and replace the Param with it */
4937  /* We don't need to copy at this time (it'll get done later) */
4938  return list_nth(context->args, param->paramid - 1);
4939  }
4941 }
4942 
4943 /*
4944  * error context callback to let us supply a call-stack traceback
4945  */
4946 static void
4948 {
4950  int syntaxerrposition;
4951 
4952  /* If it's a syntax error, convert to internal syntax error report */
4953  syntaxerrposition = geterrposition();
4954  if (syntaxerrposition > 0)
4955  {
4956  errposition(0);
4957  internalerrposition(syntaxerrposition);
4958  internalerrquery(callback_arg->prosrc);
4959  }
4960 
4961  errcontext("SQL function \"%s\" during inlining", callback_arg->proname);
4962 }
4963 
4964 /*
4965  * evaluate_expr: pre-evaluate a constant expression
4966  *
4967  * We use the executor's routine ExecEvalExpr() to avoid duplication of
4968  * code and ensure we get the same result as the executor would get.
4969  */
4970 Expr *
4971 evaluate_expr(Expr *expr, Oid result_type, int32 result_typmod,
4972  Oid result_collation)
4973 {
4974  EState *estate;
4975  ExprState *exprstate;
4976  MemoryContext oldcontext;
4977  Datum const_val;
4978  bool const_is_null;
4979  int16 resultTypLen;
4980  bool resultTypByVal;
4981 
4982  /*
4983  * To use the executor, we need an EState.
4984  */
4985  estate = CreateExecutorState();
4986 
4987  /* We can use the estate's working context to avoid memory leaks. */
4988  oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
4989 
4990  /* Make sure any opfuncids are filled in. */
4991  fix_opfuncids((Node *) expr);
4992 
4993  /*
4994  * Prepare expr for execution. (Note: we can't use ExecPrepareExpr
4995  * because it'd result in recursively invoking eval_const_expressions.)
4996  */
4997  exprstate = ExecInitExpr(expr, NULL);
4998 
4999  /*
5000  * And evaluate it.
5001  *
5002  * It is OK to use a default econtext because none of the ExecEvalExpr()
5003  * code used in this situation will use econtext. That might seem
5004  * fortuitous, but it's not so unreasonable --- a constant expression does
5005  * not depend on context, by definition, n'est ce pas?
5006  */
5007  const_val = ExecEvalExprSwitchContext(exprstate,
5008  GetPerTupleExprContext(estate),
5009  &const_is_null);
5010 
5011  /* Get info needed about result datatype */
5012  get_typlenbyval(result_type, &resultTypLen, &resultTypByVal);
5013 
5014  /* Get back to outer memory context */
5015  MemoryContextSwitchTo(oldcontext);
5016 
5017  /*
5018  * Must copy result out of sub-context used by expression eval.
5019  *
5020  * Also, if it's varlena, forcibly detoast it. This protects us against
5021  * storing TOAST pointers into plans that might outlive the referenced
5022  * data. (makeConst would handle detoasting anyway, but it's worth a few
5023  * extra lines here so that we can do the copy and detoast in one step.)
5024  */
5025  if (!const_is_null)
5026  {
5027  if (resultTypLen == -1)
5028  const_val = PointerGetDatum(PG_DETOAST_DATUM_COPY(const_val));
5029  else
5030  const_val = datumCopy(const_val, resultTypByVal, resultTypLen);
5031  }
5032 
5033  /* Release all the junk we just created */
5034  FreeExecutorState(estate);
5035 
5036  /*
5037  * Make the constant result node.
5038  */
5039  return (Expr *) makeConst(result_type, result_typmod, result_collation,
5040  resultTypLen,
5041  const_val, const_is_null,
5042  resultTypByVal);
5043 }
5044 
5045 
5046 /*
5047  * inline_set_returning_function
5048  * Attempt to "inline" a set-returning function in the FROM clause.
5049  *
5050  * "rte" is an RTE_FUNCTION rangetable entry. If it represents a call of a
5051  * set-returning SQL function that can safely be inlined, expand the function
5052  * and return the substitute Query structure. Otherwise, return NULL.
5053  *
5054  * We assume that the RTE's expression has already been put through
5055  * eval_const_expressions(), which among other things will take care of
5056  * default arguments and named-argument notation.
5057  *
5058  * This has a good deal of similarity to inline_function(), but that's
5059  * for the non-set-returning case, and there are enough differences to
5060  * justify separate functions.
5061  */
5062 Query *
5064 {
5065  RangeTblFunction *rtfunc;
5066  FuncExpr *fexpr;
5067  Oid func_oid;
5068  HeapTuple func_tuple;
5069  Form_pg_proc funcform;
5070  char *src;
5071  Datum tmp;
5072  bool isNull;
5073  MemoryContext oldcxt;
5074  MemoryContext mycxt;
5075  inline_error_callback_arg callback_arg;
5076  ErrorContextCallback sqlerrcontext;
5078  TypeFuncClass functypclass;
5079  TupleDesc rettupdesc;
5080  List *raw_parsetree_list;
5081  List *querytree_list;
5082  Query *querytree;
5083 
5084  Assert(rte->rtekind == RTE_FUNCTION);
5085 
5086  /*
5087  * It doesn't make a lot of sense for a SQL SRF to refer to itself in its
5088  * own FROM clause, since that must cause infinite recursion at runtime.
5089  * It will cause this code to recurse too, so check for stack overflow.
5090  * (There's no need to do more.)
5091  */
5093 
5094  /* Fail if the RTE has ORDINALITY - we don't implement that here. */
5095  if (rte->funcordinality)
5096  return NULL;
5097 
5098  /* Fail if RTE isn't a single, simple FuncExpr */
5099  if (list_length(rte->functions) != 1)
5100  return NULL;
5101  rtfunc = (RangeTblFunction *) linitial(rte->functions);
5102 
5103  if (!IsA(rtfunc->funcexpr, FuncExpr))
5104  return NULL;
5105  fexpr = (FuncExpr *) rtfunc->funcexpr;
5106 
5107  func_oid = fexpr->funcid;
5108 
5109  /*
5110  * The function must be declared to return a set, else inlining would
5111  * change the results if the contained SELECT didn't return exactly one
5112  * row.
5113  */
5114  if (!fexpr->funcretset)
5115  return NULL;
5116 
5117  /*
5118  * Refuse to inline if the arguments contain any volatile functions or
5119  * sub-selects. Volatile functions are rejected because inlining may
5120  * result in the arguments being evaluated multiple times, risking a
5121  * change in behavior. Sub-selects are rejected partly for implementation
5122  * reasons (pushing them down another level might change their behavior)
5123  * and partly because they're likely to be expensive and so multiple
5124  * evaluation would be bad.
5125  */
5126  if (contain_volatile_functions((Node *) fexpr->args) ||
5127  contain_subplans((Node *) fexpr->args))
5128  return NULL;
5129 
5130  /* Check permission to call function (fail later, if not) */
5131  if (object_aclcheck(ProcedureRelationId, func_oid, GetUserId(), ACL_EXECUTE) != ACLCHECK_OK)
5132  return NULL;
5133 
5134  /* Check whether a plugin wants to hook function entry/exit */
5135  if (FmgrHookIsNeeded(func_oid))
5136  return NULL;
5137 
5138  /*
5139  * OK, let's take a look at the function's pg_proc entry.
5140  */
5141  func_tuple = SearchSysCache1(PROCOID, ObjectIdGetDatum(func_oid));
5142  if (!HeapTupleIsValid(func_tuple))
5143  elog(ERROR, "cache lookup failed for function %u", func_oid);
5144  funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
5145 
5146  /*
5147  * Forget it if the function is not SQL-language or has other showstopper
5148  * properties. In particular it mustn't be declared STRICT, since we
5149  * couldn't enforce that. It also mustn't be VOLATILE, because that is
5150  * supposed to cause it to be executed with its own snapshot, rather than
5151  * sharing the snapshot of the calling query. We also disallow returning
5152  * SETOF VOID, because inlining would result in exposing the actual result
5153  * of the function's last SELECT, which should not happen in that case.
5154  * (Rechecking prokind, proretset, and pronargs is just paranoia.)
5155  */
5156  if (funcform->prolang != SQLlanguageId ||
5157  funcform->prokind != PROKIND_FUNCTION ||
5158  funcform->proisstrict ||
5159  funcform->provolatile == PROVOLATILE_VOLATILE ||
5160  funcform->prorettype == VOIDOID ||
5161  funcform->prosecdef ||
5162  !funcform->proretset ||
5163  list_length(fexpr->args) != funcform->pronargs ||
5164  !heap_attisnull(func_tuple, Anum_pg_proc_proconfig, NULL))
5165  {
5166  ReleaseSysCache(func_tuple);
5167  return NULL;
5168  }
5169 
5170  /*
5171  * Make a temporary memory context, so that we don't leak all the stuff
5172  * that parsing might create.
5173  */
5175  "inline_set_returning_function",
5177  oldcxt = MemoryContextSwitchTo(mycxt);
5178 
5179  /* Fetch the function body */
5180  tmp = SysCacheGetAttrNotNull(PROCOID, func_tuple, Anum_pg_proc_prosrc);
5181  src = TextDatumGetCString(tmp);
5182 
5183  /*
5184  * Setup error traceback support for ereport(). This is so that we can
5185  * finger the function that bad information came from.
5186  */
5187  callback_arg.proname = NameStr(funcform->proname);
5188  callback_arg.prosrc = src;
5189 
5190  sqlerrcontext.callback = sql_inline_error_callback;
5191  sqlerrcontext.arg = &callback_arg;
5192  sqlerrcontext.previous = error_context_stack;
5193  error_context_stack = &sqlerrcontext;
5194 
5195  /* If we have prosqlbody, pay attention to that not prosrc */
5196  tmp = SysCacheGetAttr(PROCOID,
5197  func_tuple,
5198  Anum_pg_proc_prosqlbody,
5199  &isNull);
5200  if (!isNull)
5201  {
5202  Node *n;
5203 
5205  if (IsA(n, List))
5206  querytree_list = linitial_node(List, castNode(List, n));
5207  else
5208  querytree_list = list_make1(n);
5209  if (list_length(querytree_list) != 1)
5210  goto fail;
5211  querytree = linitial(querytree_list);
5212 
5213  /* Acquire necessary locks, then apply rewriter. */
5214  AcquireRewriteLocks(querytree, true, false);
5215  querytree_list = pg_rewrite_query(querytree);
5216  if (list_length(querytree_list) != 1)
5217  goto fail;
5218  querytree = linitial(querytree_list);
5219  }
5220  else
5221  {
5222  /*
5223  * Set up to handle parameters while parsing the function body. We
5224  * can use the FuncExpr just created as the input for
5225  * prepare_sql_fn_parse_info.
5226  */
5227  pinfo = prepare_sql_fn_parse_info(func_tuple,
5228  (Node *) fexpr,
5229  fexpr->inputcollid);
5230 
5231  /*
5232  * Parse, analyze, and rewrite (unlike inline_function(), we can't
5233  * skip rewriting here). We can fail as soon as we find more than one
5234  * query, though.
5235  */
5236  raw_parsetree_list = pg_parse_query(src);
5237  if (list_length(raw_parsetree_list) != 1)
5238  goto fail;
5239 
5240  querytree_list = pg_analyze_and_rewrite_withcb(linitial(raw_parsetree_list),
5241  src,
5243  pinfo, NULL);
5244  if (list_length(querytree_list) != 1)
5245  goto fail;
5246  querytree = linitial(querytree_list);
5247  }
5248 
5249  /*
5250  * Also resolve the actual function result tupdesc, if composite. If we
5251  * have a coldeflist, believe that; otherwise use get_expr_result_type.
5252  * (This logic should match ExecInitFunctionScan.)
5253  */
5254  if (rtfunc->funccolnames != NIL)
5255  {
5256  functypclass = TYPEFUNC_RECORD;
5257  rettupdesc = BuildDescFromLists(rtfunc->funccolnames,
5258  rtfunc->funccoltypes,
5259  rtfunc->funccoltypmods,
5260  rtfunc->funccolcollations);
5261  }
5262  else
5263  functypclass = get_expr_result_type((Node *) fexpr, NULL, &rettupdesc);
5264 
5265  /*
5266  * The single command must be a plain SELECT.
5267  */
5268  if (!IsA(querytree, Query) ||
5269  querytree->commandType != CMD_SELECT)
5270  goto fail;
5271 
5272  /*
5273  * Make sure the function (still) returns what it's declared to. This
5274  * will raise an error if wrong, but that's okay since the function would
5275  * fail at runtime anyway. Note that check_sql_fn_retval will also insert
5276  * coercions if needed to make the tlist expression(s) match the declared
5277  * type of the function. We also ask it to insert dummy NULL columns for
5278  * any dropped columns in rettupdesc, so that the elements of the modified
5279  * tlist match up to the attribute numbers.
5280  *
5281  * If the function returns a composite type, don't inline unless the check
5282  * shows it's returning a whole tuple result; otherwise what it's
5283  * returning is a single composite column which is not what we need.
5284  */
5285  if (!check_sql_fn_retval(list_make1(querytree_list),
5286  fexpr->funcresulttype, rettupdesc,
5287  funcform->prokind,
5288  true, NULL) &&
5289  (functypclass == TYPEFUNC_COMPOSITE ||
5290  functypclass == TYPEFUNC_COMPOSITE_DOMAIN ||
5291  functypclass == TYPEFUNC_RECORD))
5292  goto fail; /* reject not-whole-tuple-result cases */
5293 
5294  /*
5295  * check_sql_fn_retval might've inserted a projection step, but that's
5296  * fine; just make sure we use the upper Query.
5297  */
5298  querytree = linitial_node(Query, querytree_list);
5299 
5300  /*
5301  * Looks good --- substitute parameters into the query.
5302  */
5304  funcform->pronargs,
5305  fexpr->args);
5306 
5307  /*
5308  * Copy the modified query out of the temporary memory context, and clean
5309  * up.
5310  */
5311  MemoryContextSwitchTo(oldcxt);
5312 
5314 
5315  MemoryContextDelete(mycxt);
5316  error_context_stack = sqlerrcontext.previous;
5317  ReleaseSysCache(func_tuple);
5318 
5319  /*
5320  * We don't have to fix collations here because the upper query is already
5321  * parsed, ie, the collations in the RTE are what count.
5322  */
5323 
5324  /*
5325  * Since there is now no trace of the function in the plan tree, we must
5326  * explicitly record the plan's dependency on the function.
5327  */
5329 
5330  /*
5331  * We must also notice if the inserted query adds a dependency on the
5332  * calling role due to RLS quals.
5333  */
5334  if (querytree->hasRowSecurity)
5335  root->glob->dependsOnRole = true;
5336 
5337  return querytree;
5338 
5339  /* Here if func is not inlinable: release temp memory and return NULL */
5340 fail:
5341  MemoryContextSwitchTo(oldcxt);
5342  MemoryContextDelete(mycxt);
5343  error_context_stack = sqlerrcontext.previous;
5344  ReleaseSysCache(func_tuple);
5345 
5346  return NULL;
5347 }
5348 
5349 /*
5350  * Replace Param nodes by appropriate actual parameters
5351  *
5352  * This is just enough different from substitute_actual_parameters()
5353  * that it needs its own code.
5354  */
5355 static Query *
5357 {
5359 
5360  context.nargs = nargs;
5361  context.args = args;
5362  context.sublevels_up = 1;
5363 
5364  return query_tree_mutator(expr,
5366  &context,
5367  0);
5368 }
5369 
5370 static Node *
5373 {
5374  Node *result;
5375 
5376  if (node == NULL)
5377  return NULL;
5378  if (IsA(node, Query))
5379  {
5380  context->sublevels_up++;
5381  result = (Node *) query_tree_mutator((Query *) node,
5383  context,
5384  0);
5385  context->sublevels_up--;
5386  return result;
5387  }
5388  if (IsA(node, Param))
5389  {
5390  Param *param = (Param *) node;
5391 
5392  if (param->paramkind == PARAM_EXTERN)
5393  {
5394  if (param->paramid <= 0 || param->paramid > context->nargs)
5395  elog(ERROR, "invalid paramid: %d", param->paramid);
5396 
5397  /*
5398  * Since the parameter is being inserted into a subquery, we must
5399  * adjust levels.
5400  */
5401  result = copyObject(list_nth(context->args, param->paramid - 1));
5402  IncrementVarSublevelsUp(result, context->sublevels_up, 0);
5403  return result;
5404  }
5405  }
5406  return expression_tree_mutator(node,
5408  context);
5409 }
5410 
5411 /*
5412  * pull_paramids
5413  * Returns a Bitmapset containing the paramids of all Params in 'expr'.
5414  */
5415 Bitmapset *
5417 {
5418  Bitmapset *result = NULL;
5419 
5420  (void) pull_paramids_walker((Node *) expr, &result);
5421 
5422  return result;
5423 }
5424 
5425 static bool
5427 {
5428  if (node == NULL)
5429  return false;
5430  if (IsA(node, Param))
5431  {
5432  Param *param = (Param *) node;
5433 
5434  *context = bms_add_member(*context, param->paramid);
5435  return false;
5436  }
5438 }
Datum querytree(PG_FUNCTION_ARGS)
Definition: _int_bool.c:665
@ ACLCHECK_OK
Definition: acl.h:183
AclResult object_aclcheck(Oid classid, Oid objectid, Oid roleid, AclMode mode)
Definition: aclchk.c:3810
#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:725
signed short int16
Definition: c.h:481
signed int int32
Definition: c.h:482
#define Assert(condition)
Definition: c.h:837
unsigned int Index
Definition: c.h:593
#define OidIsValid(objectId)
Definition: c.h:754
static bool contain_subplans_walker(Node *node, void *context)
Definition: clauses.c:335
#define CCDN_CASETESTEXPR_OK
Definition: clauses.c:1185
static List * simplify_or_arguments(List *args, eval_const_expressions_context *context, bool *haveNull, bool *forceTrue)
Definition: clauses.c:3789
static bool is_strict_saop(ScalarArrayOpExpr *expr, bool falseOK)
Definition: clauses.c:2025
bool contain_volatile_functions_not_nextval(Node *clause)
Definition: clauses.c:672
static bool contain_leaked_vars_checker(Oid func_id, void *context)
Definition: clauses.c:1268
static bool rowtype_field_matches(Oid rowtypeid, int fieldnum, Oid expectedtype, int32 expectedtypmod, Oid expectedcollation)
Definition: clauses.c:2185
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:1004
static List * add_function_defaults(List *args, int pronargs, HeapTuple func_tuple)
Definition: clauses.c:4325
#define ece_all_arguments_const(node)
Definition: clauses.c:2425
Expr * evaluate_expr(Expr *expr, Oid result_type, int32 result_typmod, Oid result_collation)
Definition: clauses.c:4971
#define ece_evaluate_expr(node)
Definition: clauses.c:2429
static bool max_parallel_hazard_checker(Oid func_id, void *context)
Definition: clauses.c:821
static bool max_parallel_hazard_test(char proparallel, max_parallel_hazard_context *context)
Definition: clauses.c:793
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:5371
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:998
int NumRelids(PlannerInfo *root, Node *clause)
Definition: clauses.c:2129
bool contain_mutable_functions(Node *clause)
Definition: clauses.c:369
bool is_pseudo_constant_clause(Node *clause)
Definition: clauses.c:2087
static bool max_parallel_hazard_walker(Node *node, max_parallel_hazard_context *context)
Definition: clauses.c:828
bool contain_window_function(Node *clause)
Definition: clauses.c:214
#define ece_generic_processing(node)
Definition: clauses.c:2416
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:4424
static Node * substitute_actual_parameters_mutator(Node *node, substitute_actual_parameters_context *context)
Definition: clauses.c:4919
static bool contain_mutable_functions_checker(Oid func_id, void *context)
Definition: clauses.c:375
bool is_pseudo_constant_clause_relids(Node *clause, Relids relids)
Definition: clauses.c:2107
List * expand_function_arguments(List *args, bool include_out_arguments, Oid result_type, HeapTuple func_tuple)
Definition: clauses.c:4174
static Query * substitute_actual_srf_parameters(Query *expr, int nargs, List *args)
Definition: clauses.c:5356
Var * find_forced_null_var(Node *node)
Definition: clauses.c:1976
static bool ece_function_is_safe(Oid funcid, eval_const_expressions_context *context)
Definition: clauses.c:3751
static bool contain_volatile_functions_checker(Oid func_id, void *context)
Definition: clauses.c:543
static List * simplify_and_arguments(List *args, eval_const_expressions_context *context, bool *haveNull, bool *forceFalse)
Definition: clauses.c:3895
List * find_forced_null_vars(Node *node)
Definition: clauses.c:1915
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:4058
bool contain_volatile_functions_after_planning(Expr *expr)
Definition: clauses.c:658
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:4550
static List * reorder_function_arguments(List *args, int pronargs, HeapTuple func_tuple)
Definition: clauses.c:4255
static Node * substitute_actual_parameters(Node *expr, int nargs, List *args, int *usecounts)
Definition: clauses.c:4906
static bool contain_mutable_functions_walker(Node *node, void *context)
Definition: clauses.c:381
Query * inline_set_returning_function(PlannerInfo *root, RangeTblEntry *rte)
Definition: clauses.c:5063
bool contain_mutable_functions_after_planning(Expr *expr)
Definition: clauses.c:489
static bool contain_volatile_functions_walker(Node *node, void *context)
Definition: clauses.c:549
bool contain_leaked_vars(Node *clause)
Definition: clauses.c:1262
static Relids find_nonnullable_rels_walker(Node *node, bool top_level)
Definition: clauses.c:1461
Bitmapset * pull_paramids(Expr *expr)
Definition: clauses.c:5416
void convert_saop_to_hashed_saop(Node *node)
Definition: clauses.c:2286
static void sql_inline_error_callback(void *arg)
Definition: clauses.c:4947
static bool contain_volatile_functions_not_nextval_walker(Node *node, void *context)
Definition: clauses.c:685
static bool contain_leaked_vars_walker(Node *node, void *context)
Definition: clauses.c:1274
static bool contain_non_const_walker(Node *node, void *context)
Definition: clauses.c:3735
static bool contain_context_dependent_node(Node *clause)
Definition: clauses.c:1178
Relids find_nonnullable_rels(Node *clause)
Definition: clauses.c:1455
static void recheck_cast_function_args(List *args, Oid result_type, Oid *proargtypes, int pronargs, HeapTuple func_tuple)
Definition: clauses.c:4379
static bool find_window_functions_walker(Node *node, WindowFuncLists *lists)
Definition: clauses.c:239
List * find_nonnullable_vars(Node *clause)
Definition: clauses.c:1706
char max_parallel_hazard(Query *parse)
Definition: clauses.c:733
bool is_parallel_safe(PlannerInfo *root, Node *node)
Definition: clauses.c:752
bool contain_nonstrict_functions(Node *clause)
Definition: clauses.c:992
static bool contain_volatile_functions_not_nextval_checker(Oid func_id, void *context)
Definition: clauses.c:678
static List * find_nonnullable_vars_walker(Node *node, bool top_level)
Definition: clauses.c:1712
Node * eval_const_expressions(PlannerInfo *root, Node *node)
Definition: clauses.c:2253
bool contain_subplans(Node *clause)
Definition: clauses.c:329
static Node * simplify_boolean_equality(Oid opno, List *args)
Definition: clauses.c:3989
static bool contain_exec_param_walker(Node *node, List *param_ids)
Definition: clauses.c:1142
void CommuteOpExpr(OpExpr *clause)
Definition: clauses.c:2146
static Node * eval_const_expressions_mutator(Node *node, eval_const_expressions_context *context)
Definition: clauses.c:2439
static bool pull_paramids_walker(Node *node, Bitmapset **context)
Definition: clauses.c:5426
static bool convert_saop_to_hashed_saop_walker(Node *node, void *context)
Definition: clauses.c:2292
static List * fetch_function_defaults(HeapTuple func_tuple)
Definition: clauses.c:4349
bool contain_volatile_functions(Node *clause)
Definition: clauses.c:537
double expression_returns_set_rows(PlannerInfo *root, Node *clause)
Definition: clauses.c:288
static bool contain_context_dependent_node_walker(Node *node, int *flags)
Definition: clauses.c:1188
bool contain_exec_param(Node *clause, List *param_ids)
Definition: clauses.c:1136
#define MIN_ARRAY_SIZE_FOR_HASHED_SAOP
Definition: clauses.c:2268
Node * estimate_expression_value(PlannerInfo *root, Node *node)
Definition: clauses.c:2394
double cpu_operator_cost
Definition: costsize.c:134
void cost_qual_eval(QualCost *cost, List *quals, PlannerInfo *root)
Definition: costsize.c:4732
double clamp_row_est(double nrows)
Definition: costsize.c:213
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:1595
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:225
ExprState * ExecInitExpr(Expr *node, PlanState *parent)
Definition: execExpr.c:138
EState * CreateExecutorState(void)
Definition: execUtils.c:88
void FreeExecutorState(EState *estate)
Definition: execUtils.c:191
#define GetPerTupleExprContext(estate)
Definition: executor.h:561
static Datum ExecEvalExprSwitchContext(ExprState *state, ExprContext *econtext, bool *isNull)
Definition: executor.h:359
#define OidFunctionCall1(functionId, arg1)
Definition: fmgr.h:679
#define PG_DETOAST_DATUM_COPY(datum)
Definition: fmgr.h:242
#define FmgrHookIsNeeded(fn_oid)
Definition: fmgr.h:796
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