Age Owner TLA Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * clauses.c
4 : * routines to manipulate qualification clauses
5 : *
6 : * Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/optimizer/util/clauses.c
12 : *
13 : * HISTORY
14 : * AUTHOR DATE MAJOR EVENT
15 : * Andrew Yu Nov 3, 1994 clause.c and clauses.c combined
16 : *
17 : *-------------------------------------------------------------------------
18 : */
19 :
20 : #include "postgres.h"
21 :
22 : #include "access/htup_details.h"
23 : #include "catalog/pg_aggregate.h"
24 : #include "catalog/pg_class.h"
25 : #include "catalog/pg_language.h"
26 : #include "catalog/pg_operator.h"
27 : #include "catalog/pg_proc.h"
28 : #include "catalog/pg_type.h"
29 : #include "executor/executor.h"
30 : #include "executor/functions.h"
31 : #include "funcapi.h"
32 : #include "miscadmin.h"
33 : #include "nodes/makefuncs.h"
34 : #include "nodes/multibitmapset.h"
35 : #include "nodes/nodeFuncs.h"
36 : #include "nodes/subscripting.h"
37 : #include "nodes/supportnodes.h"
38 : #include "optimizer/clauses.h"
39 : #include "optimizer/cost.h"
40 : #include "optimizer/optimizer.h"
41 : #include "optimizer/plancat.h"
42 : #include "optimizer/planmain.h"
43 : #include "parser/analyze.h"
44 : #include "parser/parse_agg.h"
45 : #include "parser/parse_coerce.h"
46 : #include "parser/parse_func.h"
47 : #include "rewrite/rewriteHandler.h"
48 : #include "rewrite/rewriteManip.h"
49 : #include "tcop/tcopprot.h"
50 : #include "utils/acl.h"
51 : #include "utils/builtins.h"
52 : #include "utils/datum.h"
53 : #include "utils/fmgroids.h"
54 : #include "utils/json.h"
55 : #include "utils/jsonb.h"
56 : #include "utils/lsyscache.h"
57 : #include "utils/memutils.h"
58 : #include "utils/syscache.h"
59 : #include "utils/typcache.h"
60 :
61 : typedef struct
62 : {
63 : ParamListInfo boundParams;
64 : PlannerInfo *root;
65 : List *active_fns;
66 : Node *case_val;
67 : bool estimate;
68 : } eval_const_expressions_context;
69 :
70 : typedef struct
71 : {
72 : int nargs;
73 : List *args;
74 : int *usecounts;
75 : } substitute_actual_parameters_context;
76 :
77 : typedef struct
78 : {
79 : int nargs;
80 : List *args;
81 : int sublevels_up;
82 : } substitute_actual_srf_parameters_context;
83 :
84 : typedef struct
85 : {
86 : char *proname;
87 : char *prosrc;
88 : } inline_error_callback_arg;
89 :
90 : typedef struct
91 : {
92 : char max_hazard; /* worst proparallel hazard found so far */
93 : char max_interesting; /* worst proparallel hazard of interest */
94 : List *safe_param_ids; /* PARAM_EXEC Param IDs to treat as safe */
95 : } max_parallel_hazard_context;
96 :
97 : static bool contain_agg_clause_walker(Node *node, void *context);
98 : static bool find_window_functions_walker(Node *node, WindowFuncLists *lists);
99 : static bool contain_subplans_walker(Node *node, void *context);
100 : static bool contain_mutable_functions_walker(Node *node, void *context);
101 : static bool contain_volatile_functions_walker(Node *node, void *context);
102 : static bool contain_volatile_functions_not_nextval_walker(Node *node, void *context);
103 : static bool max_parallel_hazard_walker(Node *node,
104 : max_parallel_hazard_context *context);
105 : static bool contain_nonstrict_functions_walker(Node *node, void *context);
106 : static bool contain_exec_param_walker(Node *node, List *param_ids);
107 : static bool contain_context_dependent_node(Node *clause);
108 : static bool contain_context_dependent_node_walker(Node *node, int *flags);
109 : static bool contain_leaked_vars_walker(Node *node, void *context);
110 : static Relids find_nonnullable_rels_walker(Node *node, bool top_level);
111 : static List *find_nonnullable_vars_walker(Node *node, bool top_level);
112 : static bool is_strict_saop(ScalarArrayOpExpr *expr, bool falseOK);
113 : static bool convert_saop_to_hashed_saop_walker(Node *node, void *context);
114 : static Node *eval_const_expressions_mutator(Node *node,
115 : eval_const_expressions_context *context);
116 : static bool contain_non_const_walker(Node *node, void *context);
117 : static bool ece_function_is_safe(Oid funcid,
118 : eval_const_expressions_context *context);
119 : static List *simplify_or_arguments(List *args,
120 : eval_const_expressions_context *context,
121 : bool *haveNull, bool *forceTrue);
122 : static List *simplify_and_arguments(List *args,
123 : eval_const_expressions_context *context,
124 : bool *haveNull, bool *forceFalse);
125 : static Node *simplify_boolean_equality(Oid opno, List *args);
126 : static Expr *simplify_function(Oid funcid,
127 : Oid result_type, int32 result_typmod,
128 : Oid result_collid, Oid input_collid, List **args_p,
129 : bool funcvariadic, bool process_args, bool allow_non_const,
130 : eval_const_expressions_context *context);
131 : static List *reorder_function_arguments(List *args, int pronargs,
132 : HeapTuple func_tuple);
133 : static List *add_function_defaults(List *args, int pronargs,
134 : HeapTuple func_tuple);
135 : static List *fetch_function_defaults(HeapTuple func_tuple);
136 : static void recheck_cast_function_args(List *args, Oid result_type,
137 : Oid *proargtypes, int pronargs,
138 : HeapTuple func_tuple);
139 : static Expr *evaluate_function(Oid funcid, Oid result_type, int32 result_typmod,
140 : Oid result_collid, Oid input_collid, List *args,
141 : bool funcvariadic,
142 : HeapTuple func_tuple,
143 : eval_const_expressions_context *context);
144 : static Expr *inline_function(Oid funcid, Oid result_type, Oid result_collid,
145 : Oid input_collid, List *args,
146 : bool funcvariadic,
147 : HeapTuple func_tuple,
148 : eval_const_expressions_context *context);
149 : static Node *substitute_actual_parameters(Node *expr, int nargs, List *args,
150 : int *usecounts);
151 : static Node *substitute_actual_parameters_mutator(Node *node,
152 : substitute_actual_parameters_context *context);
153 : static void sql_inline_error_callback(void *arg);
154 : static Query *substitute_actual_srf_parameters(Query *expr,
155 : int nargs, List *args);
156 : static Node *substitute_actual_srf_parameters_mutator(Node *node,
157 : substitute_actual_srf_parameters_context *context);
158 : static bool pull_paramids_walker(Node *node, Bitmapset **context);
159 :
160 :
161 : /*****************************************************************************
162 : * Aggregate-function clause manipulation
163 : *****************************************************************************/
164 :
165 : /*
166 : * contain_agg_clause
167 : * Recursively search for Aggref/GroupingFunc nodes within a clause.
168 : *
169 : * Returns true if any aggregate found.
170 : *
171 : * This does not descend into subqueries, and so should be used only after
172 : * reduction of sublinks to subplans, or in contexts where it's known there
173 : * are no subqueries. There mustn't be outer-aggregate references either.
174 : *
175 : * (If you want something like this but able to deal with subqueries,
176 : * see rewriteManip.c's contain_aggs_of_level().)
177 : */
178 : bool
8518 tgl 179 GIC 884 : contain_agg_clause(Node *clause)
180 : {
181 884 : return contain_agg_clause_walker(clause, NULL);
8518 tgl 182 ECB : }
183 :
184 : static bool
8518 tgl 185 GIC 1271 : contain_agg_clause_walker(Node *node, void *context)
186 : {
187 1271 : if (node == NULL)
8518 tgl 188 CBC 9 : return false;
8518 tgl 189 GIC 1262 : if (IsA(node, Aggref))
7247 tgl 190 ECB : {
7247 tgl 191 CBC 321 : Assert(((Aggref *) node)->agglevelsup == 0);
6385 bruce 192 321 : return true; /* abort the tree traversal and return true */
193 : }
2814 andres 194 941 : if (IsA(node, GroupingFunc))
2814 andres 195 ECB : {
2814 andres 196 GIC 6 : Assert(((GroupingFunc *) node)->agglevelsup == 0);
2814 andres 197 CBC 6 : return true; /* abort the tree traversal and return true */
198 : }
7247 tgl 199 935 : Assert(!IsA(node, SubLink));
8518 200 935 : return expression_tree_walker(node, contain_agg_clause_walker, context);
201 : }
8518 tgl 202 ECB :
5215 203 : /*****************************************************************************
204 : * Window-function clause manipulation
205 : *****************************************************************************/
206 :
207 : /*
208 : * contain_window_function
209 : * Recursively search for WindowFunc nodes within a clause.
210 : *
211 : * Since window functions don't have level fields, but are hard-wired to
212 : * be associated with the current query level, this is just the same as
213 : * rewriteManip.c's function.
214 : */
215 : bool
5215 tgl 216 GIC 492 : contain_window_function(Node *clause)
217 : {
3894 218 492 : return contain_windowfuncs(clause);
5215 tgl 219 ECB : }
220 :
221 : /*
222 : * find_window_functions
223 : * Locate all the WindowFunc nodes in an expression tree, and organize
224 : * them by winref ID number.
225 : *
226 : * Caller must provide an upper bound on the winref IDs expected in the tree.
227 : */
228 : WindowFuncLists *
5215 tgl 229 GIC 1020 : find_window_functions(Node *clause, Index maxWinRef)
230 : {
231 1020 : WindowFuncLists *lists = palloc(sizeof(WindowFuncLists));
5215 tgl 232 ECB :
5215 tgl 233 GIC 1020 : lists->numWindowFuncs = 0;
5215 tgl 234 CBC 1020 : lists->maxWinRef = maxWinRef;
5215 tgl 235 GIC 1020 : lists->windowFuncs = (List **) palloc0((maxWinRef + 1) * sizeof(List *));
5215 tgl 236 CBC 1020 : (void) find_window_functions_walker(clause, lists);
237 1020 : return lists;
5215 tgl 238 ECB : }
239 :
240 : static bool
5215 tgl 241 GIC 9067 : find_window_functions_walker(Node *node, WindowFuncLists *lists)
242 : {
243 9067 : if (node == NULL)
5215 tgl 244 CBC 83 : return false;
5215 tgl 245 GIC 8984 : if (IsA(node, WindowFunc))
5215 tgl 246 ECB : {
5215 tgl 247 CBC 1347 : WindowFunc *wfunc = (WindowFunc *) node;
5215 tgl 248 ECB :
249 : /* winref is unsigned, so one-sided test is OK */
5215 tgl 250 CBC 1347 : if (wfunc->winref > lists->maxWinRef)
5215 tgl 251 UIC 0 : elog(ERROR, "WindowFunc contains out-of-range winref %u",
252 : wfunc->winref);
2587 tgl 253 ECB : /* eliminate duplicates, so that we avoid repeated computation */
2587 tgl 254 GBC 1347 : if (!list_member(lists->windowFuncs[wfunc->winref], wfunc))
255 : {
2587 tgl 256 GIC 2682 : lists->windowFuncs[wfunc->winref] =
2587 tgl 257 CBC 1341 : lappend(lists->windowFuncs[wfunc->winref], wfunc);
2587 tgl 258 GIC 1341 : lists->numWindowFuncs++;
2587 tgl 259 ECB : }
5215 260 :
261 : /*
262 : * We assume that the parser checked that there are no window
263 : * functions in the arguments or filter clause. Hence, we need not
264 : * recurse into them. (If either the parser or the planner screws up
265 : * on this point, the executor will still catch it; see ExecInitExpr.)
266 : */
5215 tgl 267 GIC 1347 : return false;
268 : }
269 7637 : Assert(!IsA(node, SubLink));
5215 tgl 270 CBC 7637 : return expression_tree_walker(node, find_window_functions_walker,
271 : (void *) lists);
5215 tgl 272 ECB : }
273 :
274 :
275 : /*****************************************************************************
276 : * Support for expressions returning sets
277 : *****************************************************************************/
278 :
279 : /*
280 : * expression_returns_set_rows
281 : * Estimate the number of rows returned by a set-returning expression.
282 : * The result is 1 if it's not a set-returning expression.
283 : *
284 : * We should only examine the top-level function or operator; it used to be
285 : * appropriate to recurse, but not anymore. (Even if there are more SRFs in
286 : * the function's inputs, their multipliers are accounted for separately.)
287 : *
288 : * Note: keep this in sync with expression_returns_set() in nodes/nodeFuncs.c.
289 : */
290 : double
1520 tgl 291 GIC 134040 : expression_returns_set_rows(PlannerInfo *root, Node *clause)
292 : {
2272 andres 293 134040 : if (clause == NULL)
2272 andres 294 LBC 0 : return 1.0;
2272 andres 295 GIC 134040 : if (IsA(clause, FuncExpr))
5921 tgl 296 ECB : {
2272 andres 297 GBC 21771 : FuncExpr *expr = (FuncExpr *) clause;
5921 tgl 298 ECB :
5921 tgl 299 GIC 21771 : if (expr->funcretset)
1520 tgl 300 CBC 19382 : return clamp_row_est(get_function_rows(root, expr->funcid, clause));
301 : }
2272 andres 302 114658 : if (IsA(clause, OpExpr))
5921 tgl 303 ECB : {
2272 andres 304 GIC 1411 : OpExpr *expr = (OpExpr *) clause;
5921 tgl 305 ECB :
5921 tgl 306 GIC 1411 : if (expr->opretset)
5921 tgl 307 ECB : {
5921 tgl 308 GIC 3 : set_opfuncid(expr);
1520 tgl 309 CBC 3 : return clamp_row_est(get_function_rows(root, expr->opfuncid, clause));
310 : }
5921 tgl 311 ECB : }
2272 andres 312 CBC 114655 : return 1.0;
313 : }
314 :
5921 tgl 315 ECB :
316 : /*****************************************************************************
317 : * Subplan clause manipulation
318 : *****************************************************************************/
319 :
320 : /*
321 : * contain_subplans
322 : * Recursively search for subplan nodes within a clause.
323 : *
324 : * If we see a SubLink node, we will return true. This is only possible if
325 : * the expression tree hasn't yet been transformed by subselect.c. We do not
326 : * know whether the node will produce a true subplan or just an initplan,
327 : * but we make the conservative assumption that it will be a subplan.
328 : *
329 : * Returns true if any subplan found.
330 : */
331 : bool
8405 tgl 332 GIC 17840 : contain_subplans(Node *clause)
333 : {
334 17840 : return contain_subplans_walker(clause, NULL);
8405 tgl 335 ECB : }
336 :
337 : static bool
8405 tgl 338 GIC 67570 : contain_subplans_walker(Node *node, void *context)
339 : {
340 67570 : if (node == NULL)
8405 tgl 341 CBC 2415 : return false;
7421 tgl 342 GIC 65155 : if (IsA(node, SubPlan) ||
5343 tgl 343 CBC 65113 : IsA(node, AlternativeSubPlan) ||
7423 344 65113 : IsA(node, SubLink))
6385 bruce 345 168 : return true; /* abort the tree traversal and return true */
8405 tgl 346 64987 : return expression_tree_walker(node, contain_subplans_walker, context);
8405 tgl 347 ECB : }
348 :
8631 349 :
350 : /*****************************************************************************
351 : * Check clauses for mutable functions
352 : *****************************************************************************/
353 :
354 : /*
355 : * contain_mutable_functions
356 : * Recursively search for mutable functions within a clause.
357 : *
358 : * Returns true if any mutable function (or operator implemented by a
359 : * mutable function) is found. This test is needed so that we don't
360 : * mistakenly think that something like "WHERE random() < 0.5" can be treated
361 : * as a constant qualification.
362 : *
363 : * We will recursively look into Query nodes (i.e., SubLink sub-selects)
364 : * but not into SubPlans. See comments for contain_volatile_functions().
365 : */
366 : bool
7674 tgl 367 GIC 64633 : contain_mutable_functions(Node *clause)
368 : {
369 64633 : return contain_mutable_functions_walker(clause, NULL);
8274 tgl 370 ECB : }
371 :
2494 372 : static bool
2494 tgl 373 GIC 47613 : contain_mutable_functions_checker(Oid func_id, void *context)
374 : {
375 47613 : return (func_volatile(func_id) != PROVOLATILE_IMMUTABLE);
2494 tgl 376 ECB : }
377 :
8274 378 : static bool
7674 tgl 379 GIC 166041 : contain_mutable_functions_walker(Node *node, void *context)
380 : {
8274 381 166041 : if (node == NULL)
8274 tgl 382 CBC 1037 : return false;
383 : /* Check for mutable functions in node itself */
2494 384 165004 : if (check_functions_in_node(node, contain_mutable_functions_checker,
2494 tgl 385 ECB : context))
2494 tgl 386 GIC 4015 : return true;
5857 tgl 387 ECB :
11 alvherre 388 GNC 160989 : if (IsA(node, JsonConstructorExpr))
11 alvherre 389 ECB : {
11 alvherre 390 UNC 0 : const JsonConstructorExpr *ctor = (JsonConstructorExpr *) node;
391 : ListCell *lc;
392 : bool is_jsonb;
393 :
394 0 : is_jsonb = ctor->returning->format->format_type == JS_FORMAT_JSONB;
395 :
396 : /*
397 : * Check argument_type => json[b] conversions specifically. We still
398 : * recurse to check 'args' below, but here we want to specifically
399 : * check whether or not the emitted clause would fail to be immutable
400 : * because of TimeZone, for example.
401 : */
402 0 : foreach(lc, ctor->args)
403 : {
404 0 : Oid typid = exprType(lfirst(lc));
405 :
406 0 : if (is_jsonb ?
407 0 : !to_jsonb_is_immutable(typid) :
408 0 : !to_json_is_immutable(typid))
409 0 : return true;
410 : }
411 :
412 : /* Check all subnodes */
413 : }
11 alvherre 414 EUB :
2095 tgl 415 GIC 160989 : if (IsA(node, NextValueExpr))
416 : {
417 : /* NextValueExpr is volatile */
2095 tgl 418 UBC 0 : return true;
419 : }
420 :
421 : /*
422 : * It should be safe to treat MinMaxExpr as immutable, because it will
423 : * depend on a non-cross-type btree comparison function, and those should
424 : * always be immutable. Treating XmlExpr as immutable is more dubious,
425 : * and treating CoerceToDomain as immutable is outright dangerous. But we
2494 tgl 426 EUB : * have done so historically, and changing this would probably cause more
427 : * problems than it would fix. In practice, if you have a non-immutable
428 : * domain constraint you are in for pain anyhow.
429 : */
7394 430 :
2494 431 : /* Recurse to check arguments */
2494 tgl 432 GBC 160989 : if (IsA(node, Query))
3439 tgl 433 EUB : {
434 : /* Recurse into subselects */
3439 tgl 435 UIC 0 : return query_tree_walker((Query *) node,
436 : contain_mutable_functions_walker,
437 : context, 0);
438 : }
7674 tgl 439 CBC 160989 : return expression_tree_walker(node, contain_mutable_functions_walker,
440 : context);
441 : }
7674 tgl 442 EUB :
443 :
444 : /*****************************************************************************
445 : * Check clauses for volatile functions
446 : *****************************************************************************/
447 :
448 : /*
449 : * contain_volatile_functions
450 : * Recursively search for volatile functions within a clause.
451 : *
452 : * Returns true if any volatile function (or operator implemented by a
453 : * volatile function) is found. This test prevents, for example,
454 : * invalid conversions of volatile expressions into indexscan quals.
455 : *
3439 tgl 456 ECB : * We will recursively look into Query nodes (i.e., SubLink sub-selects)
457 : * but not into SubPlans. This is a bit odd, but intentional. If we are
458 : * looking at a SubLink, we are probably deciding whether a query tree
3439 tgl 459 EUB : * transformation is safe, and a contained sub-select should affect that;
460 : * for example, duplicating a sub-select containing a volatile function
461 : * would be bad. However, once we've got to the stage of having SubPlans,
462 : * subsequent planning need not consider volatility within those, since
3439 tgl 463 ECB : * the executor won't change its evaluation rules for a SubPlan based on
464 : * volatility.
465 : *
466 : * For some node types, for example, RestrictInfo and PathTarget, we cache
467 : * whether we found any volatile functions or not and reuse that value in any
468 : * future checks for that node. All of the logic for determining if the
469 : * cached value should be set to VOLATILITY_NOVOLATILE or VOLATILITY_VOLATILE
470 : * belongs in this function. Any code which makes changes to these nodes
471 : * which could change the outcome this function must set the cached value back
472 : * to VOLATILITY_UNKNOWN. That allows this function to redetermine the
473 : * correct value during the next call, should we need to redetermine if the
474 : * node contains any volatile functions again in the future.
475 : */
476 : bool
7674 tgl 477 GIC 959137 : contain_volatile_functions(Node *clause)
478 : {
479 959137 : return contain_volatile_functions_walker(clause, NULL);
480 : }
481 :
482 : static bool
2494 483 301405 : contain_volatile_functions_checker(Oid func_id, void *context)
484 : {
485 301405 : return (func_volatile(func_id) == PROVOLATILE_VOLATILE);
486 : }
487 :
488 : static bool
7674 489 2501027 : contain_volatile_functions_walker(Node *node, void *context)
490 : {
491 2501027 : if (node == NULL)
492 135878 : return false;
493 : /* Check for volatile functions in node itself */
2494 494 2365149 : if (check_functions_in_node(node, contain_volatile_functions_checker,
495 : context))
496 712 : return true;
497 :
2095 498 2364437 : if (IsA(node, NextValueExpr))
499 : {
500 : /* NextValueExpr is volatile */
2095 tgl 501 LBC 0 : return true;
502 : }
2095 tgl 503 ECB :
741 drowley 504 GIC 2364437 : if (IsA(node, RestrictInfo))
505 : {
506 294694 : RestrictInfo *rinfo = (RestrictInfo *) node;
741 drowley 507 ECB :
508 : /*
509 : * For RestrictInfo, check if we've checked the volatility of it
510 : * before. If so, we can just use the cached value and not bother
511 : * checking it again. Otherwise, check it and cache if whether we
512 : * found any volatile functions.
513 : */
741 drowley 514 GIC 294694 : if (rinfo->has_volatile == VOLATILITY_NOVOLATILE)
741 drowley 515 CBC 125584 : return false;
516 169110 : else if (rinfo->has_volatile == VOLATILITY_VOLATILE)
741 drowley 517 GIC 4 : return true;
741 drowley 518 ECB : else
519 : {
520 : bool hasvolatile;
521 :
741 drowley 522 CBC 169106 : hasvolatile = contain_volatile_functions_walker((Node *) rinfo->clause,
523 : context);
741 drowley 524 GIC 169106 : if (hasvolatile)
741 drowley 525 GBC 13 : rinfo->has_volatile = VOLATILITY_VOLATILE;
526 : else
741 drowley 527 GIC 169093 : rinfo->has_volatile = VOLATILITY_NOVOLATILE;
741 drowley 528 ECB :
741 drowley 529 GIC 169106 : return hasvolatile;
741 drowley 530 ECB : }
531 : }
532 :
741 drowley 533 GIC 2069743 : if (IsA(node, PathTarget))
534 : {
535 130961 : PathTarget *target = (PathTarget *) node;
536 :
537 : /*
741 drowley 538 ECB : * We also do caching for PathTarget the same as we do above for
539 : * RestrictInfos.
540 : */
741 drowley 541 CBC 130961 : if (target->has_volatile_expr == VOLATILITY_NOVOLATILE)
741 drowley 542 GIC 109318 : return false;
543 21643 : else if (target->has_volatile_expr == VOLATILITY_VOLATILE)
741 drowley 544 UIC 0 : return true;
545 : else
741 drowley 546 ECB : {
547 : bool hasvolatile;
548 :
741 drowley 549 CBC 21643 : hasvolatile = contain_volatile_functions_walker((Node *) target->exprs,
550 : context);
741 drowley 551 ECB :
741 drowley 552 GIC 21643 : if (hasvolatile)
741 drowley 553 LBC 0 : target->has_volatile_expr = VOLATILITY_VOLATILE;
554 : else
741 drowley 555 GIC 21643 : target->has_volatile_expr = VOLATILITY_NOVOLATILE;
556 :
741 drowley 557 CBC 21643 : return hasvolatile;
558 : }
741 drowley 559 ECB : }
560 :
561 : /*
562 : * See notes in contain_mutable_functions_walker about why we treat
563 : * MinMaxExpr, XmlExpr, and CoerceToDomain as immutable. Hence, none of
564 : * them are of interest here.
2494 tgl 565 : */
7394 566 :
2494 567 : /* Recurse to check arguments */
2494 tgl 568 GBC 1938782 : if (IsA(node, Query))
569 : {
570 : /* Recurse into subselects */
3439 tgl 571 GIC 7772 : return query_tree_walker((Query *) node,
572 : contain_volatile_functions_walker,
3439 tgl 573 ECB : context, 0);
574 : }
7674 tgl 575 GIC 1931010 : return expression_tree_walker(node, contain_volatile_functions_walker,
8274 tgl 576 ECB : context);
8274 tgl 577 EUB : }
578 :
3366 simon 579 ECB : /*
580 : * Special purpose version of contain_volatile_functions() for use in COPY:
2494 tgl 581 : * ignore nextval(), but treat all other functions normally.
582 : */
583 : bool
2494 tgl 584 GIC 206 : contain_volatile_functions_not_nextval(Node *clause)
585 : {
586 206 : return contain_volatile_functions_not_nextval_walker(clause, NULL);
587 : }
588 :
589 : static bool
590 90 : contain_volatile_functions_not_nextval_checker(Oid func_id, void *context)
591 : {
888 tgl 592 CBC 138 : return (func_id != F_NEXTVAL &&
2494 tgl 593 GIC 48 : func_volatile(func_id) == PROVOLATILE_VOLATILE);
594 : }
2494 tgl 595 ECB :
596 : static bool
3366 simon 597 GIC 296 : contain_volatile_functions_not_nextval_walker(Node *node, void *context)
598 : {
3366 simon 599 CBC 296 : if (node == NULL)
3366 simon 600 UIC 0 : return false;
601 : /* Check for volatile functions in node itself */
2494 tgl 602 GIC 296 : if (check_functions_in_node(node,
603 : contain_volatile_functions_not_nextval_checker,
604 : context))
2494 tgl 605 UIC 0 : return true;
606 :
607 : /*
2494 tgl 608 ECB : * See notes in contain_mutable_functions_walker about why we treat
609 : * MinMaxExpr, XmlExpr, and CoerceToDomain as immutable. Hence, none of
610 : * them are of interest here. Also, since we're intentionally ignoring
611 : * nextval(), presumably we should ignore NextValueExpr.
612 : */
3366 simon 613 :
614 : /* Recurse to check arguments */
2494 tgl 615 CBC 296 : if (IsA(node, Query))
3366 simon 616 ECB : {
617 : /* Recurse into subselects */
2494 tgl 618 UIC 0 : return query_tree_walker((Query *) node,
619 : contain_volatile_functions_not_nextval_walker,
2494 tgl 620 ECB : context, 0);
621 : }
2494 tgl 622 CBC 296 : return expression_tree_walker(node,
2118 tgl 623 EUB : contain_volatile_functions_not_nextval_walker,
624 : context);
3366 simon 625 ECB : }
626 :
627 :
2762 rhaas 628 EUB : /*****************************************************************************
629 : * Check queries for parallel unsafe and/or restricted constructs
630 : *****************************************************************************/
631 :
632 : /*
633 : * max_parallel_hazard
634 : * Find the worst parallel-hazard level in the given query
635 : *
636 : * Returns the worst function hazard property (the earliest in this list:
637 : * PROPARALLEL_UNSAFE, PROPARALLEL_RESTRICTED, PROPARALLEL_SAFE) that can
2424 tgl 638 ECB : * be found in the given parsetree. We use this to find out whether the query
639 : * can be parallelized at all. The caller will also save the result in
640 : * PlannerGlobal so as to short-circuit checks of portions of the querytree
2424 tgl 641 EUB : * later, in the common case where everything is SAFE.
642 : */
643 : char
746 akapila 644 GIC 135745 : max_parallel_hazard(Query *parse)
2424 tgl 645 ECB : {
646 : max_parallel_hazard_context context;
647 :
2424 tgl 648 GIC 135745 : context.max_hazard = PROPARALLEL_SAFE;
649 135745 : context.max_interesting = PROPARALLEL_UNSAFE;
2182 650 135745 : context.safe_param_ids = NIL;
2424 651 135745 : (void) max_parallel_hazard_walker((Node *) parse, &context);
652 135745 : return context.max_hazard;
653 : }
654 :
655 : /*
656 : * is_parallel_safe
657 : * Detect whether the given expr contains only parallel-safe functions
658 : *
659 : * root->glob->maxParallelHazard must previously have been set to the
660 : * result of max_parallel_hazard() on the whole query.
661 : */
662 : bool
663 856012 : is_parallel_safe(PlannerInfo *root, Node *node)
664 : {
665 : max_parallel_hazard_context context;
666 : PlannerInfo *proot;
1970 rhaas 667 ECB : ListCell *l;
668 :
669 : /*
670 : * Even if the original querytree contained nothing unsafe, we need to
2330 tgl 671 : * search the expression if we have generated any PARAM_EXEC Params while
672 : * planning, because those are parallel-restricted and there might be one
673 : * in this expression. But otherwise we don't need to look.
674 : */
2330 tgl 675 CBC 856012 : if (root->glob->maxParallelHazard == PROPARALLEL_SAFE &&
1973 rhaas 676 GIC 522064 : root->glob->paramExecTypes == NIL)
2424 tgl 677 509818 : return true;
678 : /* Else use max_parallel_hazard's search logic, but stop on RESTRICTED */
679 346194 : context.max_hazard = PROPARALLEL_SAFE;
680 346194 : context.max_interesting = PROPARALLEL_RESTRICTED;
2182 681 346194 : context.safe_param_ids = NIL;
682 :
683 : /*
684 : * The params that refer to the same or parent query level are considered
685 : * parallel-safe. The idea is that we compute such params at Gather or
1970 rhaas 686 ECB : * Gather Merge node and pass their value to workers.
687 : */
1970 rhaas 688 GIC 821374 : for (proot = root; proot != NULL; proot = proot->parent_root)
689 : {
690 501805 : foreach(l, proot->init_plans)
691 : {
692 26625 : SubPlan *initsubplan = (SubPlan *) lfirst(l);
693 :
1362 tgl 694 26625 : context.safe_param_ids = list_concat(context.safe_param_ids,
695 26625 : initsubplan->setParam);
696 : }
697 : }
1970 rhaas 698 ECB :
2424 tgl 699 CBC 346194 : return !max_parallel_hazard_walker(node, &context);
2762 rhaas 700 ECB : }
701 :
2424 tgl 702 : /* core logic for all parallel-hazard checks */
2494 703 : static bool
2424 tgl 704 CBC 578616 : max_parallel_hazard_test(char proparallel, max_parallel_hazard_context *context)
705 : {
2424 tgl 706 GIC 578616 : switch (proparallel)
707 : {
708 456825 : case PROPARALLEL_SAFE:
709 : /* nothing to see here, move along */
710 456825 : break;
2424 tgl 711 CBC 85396 : case PROPARALLEL_RESTRICTED:
712 : /* increase max_hazard to RESTRICTED */
713 85396 : Assert(context->max_hazard != PROPARALLEL_UNSAFE);
2424 tgl 714 GIC 85396 : context->max_hazard = proparallel;
2424 tgl 715 ECB : /* done if we are not expecting any unsafe functions */
2424 tgl 716 GIC 85396 : if (context->max_interesting == proparallel)
2424 tgl 717 CBC 51031 : return true;
718 34365 : break;
2424 tgl 719 GIC 36395 : case PROPARALLEL_UNSAFE:
720 36395 : context->max_hazard = proparallel;
721 : /* we're always done at the first unsafe construct */
2424 tgl 722 CBC 36395 : return true;
2424 tgl 723 UIC 0 : default:
724 0 : elog(ERROR, "unrecognized proparallel value \"%c\"", proparallel);
725 : break;
726 : }
2424 tgl 727 CBC 491190 : return false;
728 : }
2494 tgl 729 ECB :
730 : /* check_functions_in_node callback */
2424 731 : static bool
2424 tgl 732 GIC 517136 : max_parallel_hazard_checker(Oid func_id, void *context)
2424 tgl 733 ECB : {
2424 tgl 734 CBC 517136 : return max_parallel_hazard_test(func_parallel(func_id),
735 : (max_parallel_hazard_context *) context);
2494 tgl 736 ECB : }
737 :
738 : static bool
2424 tgl 739 CBC 7219870 : max_parallel_hazard_walker(Node *node, max_parallel_hazard_context *context)
2762 rhaas 740 ECB : {
2762 rhaas 741 CBC 7219870 : if (node == NULL)
742 1866709 : return false;
2706 rhaas 743 ECB :
744 : /* Check for hazardous functions in node itself */
2424 tgl 745 CBC 5353161 : if (check_functions_in_node(node, max_parallel_hazard_checker,
2494 tgl 746 EUB : context))
2494 tgl 747 GBC 47197 : return true;
748 :
749 : /*
2494 tgl 750 ECB : * It should be OK to treat MinMaxExpr as parallel-safe, since btree
751 : * opclass support functions are generally parallel-safe. XmlExpr is a
752 : * bit more dubious but we can probably get away with it. We err on the
753 : * side of caution by treating CoerceToDomain as parallel-restricted.
754 : * (Note: in principle that's wrong because a domain constraint could
746 akapila 755 : * contain a parallel-unsafe function; but useful constraints probably
756 : * never would have such, and assuming they do would cripple use of
757 : * parallel query in the presence of domain types.) NextValueExpr is
758 : * parallel-unsafe.
759 : */
2494 tgl 760 GIC 5305964 : if (IsA(node, CoerceToDomain))
761 : {
2424 tgl 762 CBC 21603 : if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context))
2706 rhaas 763 GIC 15186 : return true;
2706 rhaas 764 ECB : }
765 :
1678 akapila 766 GIC 5284361 : else if (IsA(node, NextValueExpr))
767 : {
2095 tgl 768 CBC 128 : if (max_parallel_hazard_test(PROPARALLEL_UNSAFE, context))
2095 tgl 769 GIC 128 : return true;
2095 tgl 770 ECB : }
771 :
772 : /*
773 : * Treat window functions as parallel-restricted because we aren't sure
774 : * whether the input row ordering is fully deterministic, and the output
775 : * of window functions might vary across workers if not. (In some cases,
776 : * like where the window frame orders by a primary key, we could relax
777 : * this restriction. But it doesn't currently seem worth expending extra
778 : * effort to do so.)
779 : */
1678 akapila 780 GIC 5284233 : else if (IsA(node, WindowFunc))
781 : {
782 2322 : if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context))
1678 akapila 783 CBC 1047 : return true;
784 : }
1678 akapila 785 ECB :
2494 tgl 786 : /*
787 : * As a notational convenience for callers, look through RestrictInfo.
788 : */
2494 tgl 789 CBC 5281911 : else if (IsA(node, RestrictInfo))
790 : {
2706 rhaas 791 83097 : RestrictInfo *rinfo = (RestrictInfo *) node;
2495 rhaas 792 ECB :
2424 tgl 793 GIC 83097 : return max_parallel_hazard_walker((Node *) rinfo->clause, context);
794 : }
795 :
796 : /*
797 : * Really we should not see SubLink during a max_interesting == restricted
798 : * scan, but if we do, return true.
799 : */
2245 rhaas 800 5198814 : else if (IsA(node, SubLink))
801 : {
2424 tgl 802 12881 : if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context))
2762 rhaas 803 LBC 0 : return true;
804 : }
2762 rhaas 805 ECB :
2182 tgl 806 : /*
807 : * Only parallel-safe SubPlans can be sent to workers. Within the
808 : * testexpr of the SubPlan, Params representing the output columns of the
809 : * subplan can be treated as parallel-safe, so temporarily add their IDs
810 : * to the safe_param_ids list while examining the testexpr.
811 : */
2245 rhaas 812 CBC 5185933 : else if (IsA(node, SubPlan))
813 : {
2182 tgl 814 11050 : SubPlan *subplan = (SubPlan *) node;
815 : List *save_safe_param_ids;
2182 tgl 816 ECB :
2182 tgl 817 GIC 21968 : if (!subplan->parallel_safe &&
818 10918 : max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context))
819 10918 : return true;
820 132 : save_safe_param_ids = context->safe_param_ids;
1336 821 264 : context->safe_param_ids = list_concat_copy(context->safe_param_ids,
822 132 : subplan->paramIds);
2182 tgl 823 CBC 132 : if (max_parallel_hazard_walker(subplan->testexpr, context))
2182 tgl 824 GIC 3 : return true; /* no need to restore safe_param_ids */
1362 tgl 825 CBC 129 : list_free(context->safe_param_ids);
2182 tgl 826 GBC 129 : context->safe_param_ids = save_safe_param_ids;
827 : /* we must also check args, but no special Param treatment there */
2182 tgl 828 GIC 129 : if (max_parallel_hazard_walker((Node *) subplan->args, context))
2182 tgl 829 UIC 0 : return true;
830 : /* don't want to recurse normally, so we're done */
2182 tgl 831 GIC 129 : return false;
832 : }
833 :
834 : /*
2494 tgl 835 ECB : * We can't pass Params to workers at the moment either, so they are also
836 : * parallel-restricted, unless they are PARAM_EXTERN Params or are
1990 rhaas 837 : * PARAM_EXEC Params listed in safe_param_ids, meaning they could be
838 : * either generated within workers or can be computed by the leader and
839 : * then their value can be passed to workers.
2494 tgl 840 : */
2494 tgl 841 CBC 5174883 : else if (IsA(node, Param))
2762 rhaas 842 ECB : {
2182 tgl 843 CBC 48994 : Param *param = (Param *) node;
2182 tgl 844 ECB :
1990 rhaas 845 CBC 48994 : if (param->paramkind == PARAM_EXTERN)
846 28632 : return false;
1990 rhaas 847 ECB :
2182 tgl 848 CBC 20362 : if (param->paramkind != PARAM_EXEC ||
849 19651 : !list_member_int(context->safe_param_ids, param->paramid))
850 : {
851 13628 : if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context))
2182 tgl 852 GBC 12950 : return true;
853 : }
2182 tgl 854 CBC 7412 : return false; /* nothing to recurse to */
855 : }
856 :
857 : /*
858 : * When we're first invoked on a completely unplanned tree, we must
859 : * recurse into subqueries so to as to locate parallel-unsafe constructs
860 : * anywhere in the tree.
861 : */
2494 tgl 862 GIC 5125889 : else if (IsA(node, Query))
863 : {
2494 tgl 864 CBC 166416 : Query *query = (Query *) node;
865 :
2494 tgl 866 ECB : /* SELECT FOR UPDATE/SHARE must be treated as unsafe */
2494 tgl 867 GIC 166416 : if (query->rowMarks != NULL)
2424 tgl 868 ECB : {
2424 tgl 869 CBC 863 : context->max_hazard = PROPARALLEL_UNSAFE;
2762 rhaas 870 GIC 863 : return true;
2424 tgl 871 ECB : }
2495 rhaas 872 :
873 : /* Recurse into subselects */
2494 tgl 874 CBC 165553 : return query_tree_walker(query,
2424 tgl 875 ECB : max_parallel_hazard_walker,
876 : context, 0);
2762 rhaas 877 : }
878 :
879 : /* Recurse to check arguments */
2762 rhaas 880 GIC 4980046 : return expression_tree_walker(node,
881 : max_parallel_hazard_walker,
882 : context);
883 : }
884 :
2424 tgl 885 ECB :
886 : /*****************************************************************************
7434 887 : * Check clauses for nonstrict functions
888 : *****************************************************************************/
889 :
890 : /*
891 : * contain_nonstrict_functions
892 : * Recursively search for nonstrict functions within a clause.
893 : *
894 : * Returns true if any nonstrict construct is found --- ie, anything that
895 : * could produce non-NULL output with a NULL input.
896 : *
7029 897 : * The idea here is that the caller has verified that the expression contains
898 : * one or more Var or Param nodes (as appropriate for the caller's need), and
899 : * now wishes to prove that the expression result will be NULL if any of these
900 : * inputs is NULL. If we return false, then the proof succeeded.
901 : */
902 : bool
7434 tgl 903 CBC 688 : contain_nonstrict_functions(Node *clause)
904 : {
7434 tgl 905 GIC 688 : return contain_nonstrict_functions_walker(clause, NULL);
906 : }
907 :
908 : static bool
2494 909 953 : contain_nonstrict_functions_checker(Oid func_id, void *context)
910 : {
911 953 : return !func_strict(func_id);
912 : }
913 :
914 : static bool
7434 915 2997 : contain_nonstrict_functions_walker(Node *node, void *context)
916 : {
917 2997 : if (node == NULL)
7434 tgl 918 UIC 0 : return false;
7029 tgl 919 GIC 2997 : if (IsA(node, Aggref))
920 : {
921 : /* an aggregate could return non-null with null input */
7029 tgl 922 UIC 0 : return true;
923 : }
2494 tgl 924 GIC 2997 : if (IsA(node, GroupingFunc))
925 : {
2494 tgl 926 ECB : /*
927 : * A GroupingFunc doesn't evaluate its arguments, and therefore must
928 : * be treated as nonstrict.
929 : */
2494 tgl 930 UIC 0 : return true;
931 : }
5215 tgl 932 CBC 2997 : if (IsA(node, WindowFunc))
933 : {
5215 tgl 934 ECB : /* a window function could return non-null with null input */
5215 tgl 935 UIC 0 : return true;
936 : }
1528 alvherre 937 GIC 2997 : if (IsA(node, SubscriptingRef))
6878 tgl 938 ECB : {
851 tgl 939 UIC 0 : SubscriptingRef *sbsref = (SubscriptingRef *) node;
851 tgl 940 ECB : const SubscriptRoutines *sbsroutines;
1528 alvherre 941 EUB :
851 tgl 942 ECB : /* Subscripting assignment is always presumed nonstrict */
851 tgl 943 UIC 0 : if (sbsref->refassgnexpr != NULL)
944 0 : return true;
851 tgl 945 EUB : /* Otherwise we must look up the subscripting support methods */
851 tgl 946 UIC 0 : sbsroutines = getSubscriptingRoutines(sbsref->refcontainertype, NULL);
849 tgl 947 LBC 0 : if (!(sbsroutines && sbsroutines->fetch_strict))
851 tgl 948 UIC 0 : return true;
949 : /* else fall through to check args */
950 : }
7423 tgl 951 GIC 2997 : if (IsA(node, DistinctExpr))
952 : {
7423 tgl 953 EUB : /* IS DISTINCT FROM is inherently non-strict */
7423 tgl 954 UIC 0 : return true;
7423 tgl 955 ECB : }
4404 tgl 956 GIC 2997 : if (IsA(node, NullIfExpr))
957 : {
2494 tgl 958 EUB : /* NULLIF is inherently non-strict */
2494 tgl 959 UIC 0 : return true;
7224 tgl 960 ECB : }
7423 tgl 961 GIC 2997 : if (IsA(node, BoolExpr))
7434 tgl 962 EUB : {
7423 tgl 963 UIC 0 : BoolExpr *expr = (BoolExpr *) node;
964 :
965 0 : switch (expr->boolop)
7434 tgl 966 EUB : {
7434 tgl 967 UBC 0 : case AND_EXPR:
968 : case OR_EXPR:
7029 tgl 969 EUB : /* AND, OR are inherently non-strict */
7434 tgl 970 UBC 0 : return true;
971 0 : default:
7434 tgl 972 UIC 0 : break;
973 : }
7434 tgl 974 ECB : }
7029 tgl 975 GIC 2997 : if (IsA(node, SubLink))
976 : {
7029 tgl 977 EUB : /* In some cases a sublink might be strict, but in general not */
7029 tgl 978 UIC 0 : return true;
7029 tgl 979 ECB : }
7029 tgl 980 GIC 2997 : if (IsA(node, SubPlan))
7029 tgl 981 UIC 0 : return true;
5343 tgl 982 GBC 2997 : if (IsA(node, AlternativeSubPlan))
5343 tgl 983 UIC 0 : return true;
6878 tgl 984 CBC 2997 : if (IsA(node, FieldStore))
6878 tgl 985 UIC 0 : return true;
1509 tgl 986 GBC 2997 : if (IsA(node, CoerceViaIO))
987 : {
1509 tgl 988 EUB : /*
989 : * CoerceViaIO is strict regardless of whether the I/O functions are,
990 : * so just go look at its argument; asking check_functions_in_node is
991 : * useless expense and could deliver the wrong answer.
992 : */
1509 tgl 993 GBC 417 : return contain_nonstrict_functions_walker((Node *) ((CoerceViaIO *) node)->arg,
1509 tgl 994 EUB : context);
995 : }
2017 tgl 996 GIC 2580 : if (IsA(node, ArrayCoerceExpr))
997 : {
2017 tgl 998 ECB : /*
999 : * ArrayCoerceExpr is strict at the array level, regardless of what
1000 : * the per-element expression is; so we should ignore elemexpr and
2017 tgl 1001 EUB : * recurse only into the arg.
1002 : */
1509 tgl 1003 LBC 0 : return contain_nonstrict_functions_walker((Node *) ((ArrayCoerceExpr *) node)->arg,
1509 tgl 1004 EUB : context);
2017 tgl 1005 ECB : }
7434 tgl 1006 GBC 2580 : if (IsA(node, CaseExpr))
7434 tgl 1007 CBC 68 : return true;
6352 tgl 1008 GBC 2512 : if (IsA(node, ArrayExpr))
6352 tgl 1009 LBC 0 : return true;
6908 tgl 1010 GIC 2512 : if (IsA(node, RowExpr))
6908 tgl 1011 UIC 0 : return true;
6311 tgl 1012 GIC 2512 : if (IsA(node, RowCompareExpr))
6311 tgl 1013 UIC 0 : return true;
7357 tgl 1014 GIC 2512 : if (IsA(node, CoalesceExpr))
7357 tgl 1015 UIC 0 : return true;
6496 tgl 1016 CBC 2512 : if (IsA(node, MinMaxExpr))
6496 tgl 1017 UIC 0 : return true;
5950 tgl 1018 GIC 2512 : if (IsA(node, XmlExpr))
5950 tgl 1019 LBC 0 : return true;
7434 tgl 1020 GIC 2512 : if (IsA(node, NullTest))
7434 tgl 1021 UIC 0 : return true;
7434 tgl 1022 GIC 2512 : if (IsA(node, BooleanTest))
7434 tgl 1023 UIC 0 : return true;
1024 :
1025 : /* Check other function-containing nodes */
2017 tgl 1026 GBC 2512 : if (check_functions_in_node(node, contain_nonstrict_functions_checker,
1027 : context))
2494 tgl 1028 GIC 3 : return true;
2017 tgl 1029 ECB :
7434 tgl 1030 CBC 2509 : return expression_tree_walker(node, contain_nonstrict_functions_walker,
7434 tgl 1031 ECB : context);
7434 tgl 1032 EUB : }
7434 tgl 1033 ECB :
968 tgl 1034 EUB : /*****************************************************************************
968 tgl 1035 ECB : * Check clauses for Params
968 tgl 1036 EUB : *****************************************************************************/
968 tgl 1037 ECB :
968 tgl 1038 EUB : /*
968 tgl 1039 ECB : * contain_exec_param
968 tgl 1040 EUB : * Recursively search for PARAM_EXEC Params within a clause.
968 tgl 1041 ECB : *
968 tgl 1042 EUB : * Returns true if the clause contains any PARAM_EXEC Param with a paramid
968 tgl 1043 ECB : * appearing in the given list of Param IDs. Does not descend into
968 tgl 1044 EUB : * subqueries!
968 tgl 1045 ECB : */
968 tgl 1046 EUB : bool
968 tgl 1047 GIC 1322 : contain_exec_param(Node *clause, List *param_ids)
1048 : {
968 tgl 1049 CBC 1322 : return contain_exec_param_walker(clause, param_ids);
1050 : }
968 tgl 1051 ECB :
1052 : static bool
968 tgl 1053 CBC 1424 : contain_exec_param_walker(Node *node, List *param_ids)
1054 : {
968 tgl 1055 GIC 1424 : if (node == NULL)
1056 9 : return false;
1057 1415 : if (IsA(node, Param))
1058 : {
1059 6 : Param *p = (Param *) node;
1060 :
1061 12 : if (p->paramkind == PARAM_EXEC &&
1062 6 : list_member_int(param_ids, p->paramid))
1063 6 : return true;
1064 : }
1065 1409 : return expression_tree_walker(node, contain_exec_param_walker, param_ids);
1066 : }
1067 :
1068 : /*****************************************************************************
1069 : * Check clauses for context-dependent nodes
2435 tgl 1070 ECB : *****************************************************************************/
1071 :
1072 : /*
1073 : * contain_context_dependent_node
1074 : * Recursively search for context-dependent nodes within a clause.
1075 : *
1076 : * CaseTestExpr nodes must appear directly within the corresponding CaseExpr,
1077 : * not nested within another one, or they'll see the wrong test value. If one
1078 : * appears "bare" in the arguments of a SQL function, then we can't inline the
1622 1079 : * SQL function for fear of creating such a situation. The same applies for
1080 : * CaseTestExpr used within the elemexpr of an ArrayCoerceExpr.
1081 : *
2435 1082 : * CoerceToDomainValue would have the same issue if domain CHECK expressions
1083 : * could get inlined into larger expressions, but presently that's impossible.
1084 : * Still, it might be allowed in future, or other node types with similar
1085 : * issues might get invented. So give this function a generic name, and set
1086 : * up the recursion state to allow multiple flag bits.
1087 : */
1088 : static bool
2435 tgl 1089 GIC 5564 : contain_context_dependent_node(Node *clause)
1090 : {
1091 5564 : int flags = 0;
1092 :
1093 5564 : return contain_context_dependent_node_walker(clause, &flags);
1094 : }
1095 :
1096 : #define CCDN_CASETESTEXPR_OK 0x0001 /* CaseTestExpr okay here? */
1097 :
1098 : static bool
1099 8267 : contain_context_dependent_node_walker(Node *node, int *flags)
1100 : {
1101 8267 : if (node == NULL)
1102 4344 : return false;
1103 3923 : if (IsA(node, CaseTestExpr))
1622 1104 3 : return !(*flags & CCDN_CASETESTEXPR_OK);
1105 3920 : else if (IsA(node, CaseExpr))
1106 : {
2435 tgl 1107 UIC 0 : CaseExpr *caseexpr = (CaseExpr *) node;
1108 :
1109 : /*
1110 : * If this CASE doesn't have a test expression, then it doesn't create
1111 : * a context in which CaseTestExprs should appear, so just fall
2435 tgl 1112 ECB : * through and treat it as a generic expression node.
1113 : */
2435 tgl 1114 LBC 0 : if (caseexpr->arg)
1115 : {
1116 0 : int save_flags = *flags;
1117 : bool res;
1118 :
1119 : /*
1120 : * Note: in principle, we could distinguish the various sub-parts
1121 : * of a CASE construct and set the flag bit only for some of them,
2435 tgl 1122 ECB : * since we are only expecting CaseTestExprs to appear in the
1123 : * "expr" subtree of the CaseWhen nodes. But it doesn't really
1124 : * seem worth any extra code. If there are any bare CaseTestExprs
1125 : * elsewhere in the CASE, something's wrong already.
1126 : */
1622 tgl 1127 LBC 0 : *flags |= CCDN_CASETESTEXPR_OK;
2435 1128 0 : res = expression_tree_walker(node,
1129 : contain_context_dependent_node_walker,
2435 tgl 1130 EUB : (void *) flags);
2435 tgl 1131 UIC 0 : *flags = save_flags;
1132 0 : return res;
1133 : }
1134 : }
1622 tgl 1135 GIC 3920 : else if (IsA(node, ArrayCoerceExpr))
1136 : {
1622 tgl 1137 UBC 0 : ArrayCoerceExpr *ac = (ArrayCoerceExpr *) node;
1138 : int save_flags;
1622 tgl 1139 EUB : bool res;
1140 :
1141 : /* Check the array expression */
1622 tgl 1142 UIC 0 : if (contain_context_dependent_node_walker((Node *) ac->arg, flags))
1143 0 : return true;
1144 :
1145 : /* Check the elemexpr, which is allowed to contain CaseTestExpr */
1146 0 : save_flags = *flags;
1147 0 : *flags |= CCDN_CASETESTEXPR_OK;
1148 0 : res = contain_context_dependent_node_walker((Node *) ac->elemexpr,
1149 : flags);
1622 tgl 1150 UBC 0 : *flags = save_flags;
1151 0 : return res;
1152 : }
2435 tgl 1153 GIC 3920 : return expression_tree_walker(node, contain_context_dependent_node_walker,
2435 tgl 1154 EUB : (void *) flags);
1155 : }
1156 :
1157 : /*****************************************************************************
2904 sfrost 1158 ECB : * Check clauses for Vars passed to non-leakproof functions
1159 : *****************************************************************************/
4073 rhaas 1160 EUB :
1161 : /*
1162 : * contain_leaked_vars
1163 : * Recursively scan a clause to discover whether it contains any Var
1164 : * nodes (of the current query level) that are passed as arguments to
2904 sfrost 1165 : * leaky functions.
4073 rhaas 1166 : *
1167 : * Returns true if the clause contains any non-leakproof functions that are
1168 : * passed Var nodes of the current query level, and which might therefore leak
2272 tgl 1169 : * data. Such clauses must be applied after any lower-level security barrier
1170 : * clauses.
4073 rhaas 1171 : */
1172 : bool
2904 sfrost 1173 GBC 2164 : contain_leaked_vars(Node *clause)
4073 rhaas 1174 EUB : {
2904 sfrost 1175 GIC 2164 : return contain_leaked_vars_walker(clause, NULL);
4073 rhaas 1176 ECB : }
1177 :
1178 : static bool
2494 tgl 1179 GIC 2237 : contain_leaked_vars_checker(Oid func_id, void *context)
1180 : {
1181 2237 : return !get_func_leakproof(func_id);
1182 : }
1183 :
1184 : static bool
2904 sfrost 1185 4313 : contain_leaked_vars_walker(Node *node, void *context)
1186 : {
4073 rhaas 1187 4313 : if (node == NULL)
4073 rhaas 1188 UIC 0 : return false;
1189 :
4073 rhaas 1190 GIC 4313 : switch (nodeTag(node))
1191 : {
1192 2049 : case T_Var:
1193 : case T_Const:
1194 : case T_Param:
1195 : case T_ArrayExpr:
2494 tgl 1196 ECB : case T_FieldSelect:
1197 : case T_FieldStore:
4073 rhaas 1198 : case T_NamedArgExpr:
1199 : case T_BoolExpr:
1200 : case T_RelabelType:
1201 : case T_CollateExpr:
1202 : case T_CaseExpr:
1203 : case T_CaseTestExpr:
1204 : case T_RowExpr:
1205 : case T_NullTest:
1206 : case T_BooleanTest:
2095 tgl 1207 : case T_NextValueExpr:
1208 : case T_List:
3955 bruce 1209 :
4073 rhaas 1210 EUB : /*
1211 : * We know these node types don't contain function calls; but
4073 rhaas 1212 ECB : * something further down in the node tree might.
1213 : */
4073 rhaas 1214 CBC 2049 : break;
1215 :
4073 rhaas 1216 GIC 2237 : case T_FuncExpr:
1217 : case T_OpExpr:
1218 : case T_DistinctExpr:
1219 : case T_NullIfExpr:
1220 : case T_ScalarArrayOpExpr:
1221 : case T_CoerceViaIO:
1222 : case T_ArrayCoerceExpr:
1223 :
1224 : /*
1225 : * If node contains a leaky function call, and there's any Var
1226 : * underneath it, reject.
1227 : */
2494 tgl 1228 2237 : if (check_functions_in_node(node, contain_leaked_vars_checker,
1229 1105 : context) &&
1230 1105 : contain_var_clause(node))
1231 1074 : return true;
4073 rhaas 1232 1163 : break;
1233 :
852 tgl 1234 UIC 0 : case T_SubscriptingRef:
1235 : {
852 tgl 1236 LBC 0 : SubscriptingRef *sbsref = (SubscriptingRef *) node;
1237 : const SubscriptRoutines *sbsroutines;
851 tgl 1238 ECB :
1239 : /* Consult the subscripting support method info */
851 tgl 1240 UIC 0 : sbsroutines = getSubscriptingRoutines(sbsref->refcontainertype,
1241 : NULL);
849 1242 0 : if (!sbsroutines ||
1243 0 : !(sbsref->refassgnexpr != NULL ?
851 1244 0 : sbsroutines->store_leakproof :
1245 0 : sbsroutines->fetch_leakproof))
1246 : {
1247 : /* Node is leaky, so reject if it contains Vars */
852 1248 0 : if (contain_var_clause(node))
1249 0 : return true;
852 tgl 1250 ECB : }
1251 : }
852 tgl 1252 LBC 0 : break;
852 tgl 1253 ECB :
4073 rhaas 1254 LBC 0 : case T_RowCompareExpr:
1255 : {
2494 tgl 1256 EUB : /*
1257 : * It's worth special-casing this because a leaky comparison
1258 : * function only compromises one pair of row elements, which
1259 : * might not contain Vars while others do.
1260 : */
4073 rhaas 1261 UIC 0 : RowCompareExpr *rcexpr = (RowCompareExpr *) node;
4073 rhaas 1262 EUB : ListCell *opid;
1263 : ListCell *larg;
2904 sfrost 1264 : ListCell *rarg;
4073 rhaas 1265 :
2904 sfrost 1266 UBC 0 : forthree(opid, rcexpr->opnos,
2904 sfrost 1267 EUB : larg, rcexpr->largs,
1268 : rarg, rcexpr->rargs)
1269 : {
3955 bruce 1270 UBC 0 : Oid funcid = get_opcode(lfirst_oid(opid));
4073 rhaas 1271 EUB :
2904 sfrost 1272 UIC 0 : if (!get_func_leakproof(funcid) &&
1273 0 : (contain_var_clause((Node *) lfirst(larg)) ||
2904 sfrost 1274 UBC 0 : contain_var_clause((Node *) lfirst(rarg))))
4073 rhaas 1275 UIC 0 : return true;
4073 rhaas 1276 EUB : }
1277 : }
4073 rhaas 1278 UIC 0 : break;
1279 :
1558 tgl 1280 0 : case T_MinMaxExpr:
1281 : {
1282 : /*
1558 tgl 1283 EUB : * MinMaxExpr is leakproof if the comparison function it calls
1284 : * is leakproof.
1285 : */
1558 tgl 1286 UIC 0 : MinMaxExpr *minmaxexpr = (MinMaxExpr *) node;
1287 : TypeCacheEntry *typentry;
1558 tgl 1288 EUB : bool leakproof;
1289 :
1290 : /* Look up the btree comparison function for the datatype */
1558 tgl 1291 UIC 0 : typentry = lookup_type_cache(minmaxexpr->minmaxtype,
1558 tgl 1292 EUB : TYPECACHE_CMP_PROC);
1558 tgl 1293 UIC 0 : if (OidIsValid(typentry->cmp_proc))
1558 tgl 1294 UBC 0 : leakproof = get_func_leakproof(typentry->cmp_proc);
1558 tgl 1295 EUB : else
1296 : {
1297 : /*
1298 : * The executor will throw an error, but here we just
1299 : * treat the missing function as leaky.
1300 : */
1558 tgl 1301 UIC 0 : leakproof = false;
1558 tgl 1302 EUB : }
1303 :
1558 tgl 1304 UIC 0 : if (!leakproof &&
1305 0 : contain_var_clause((Node *) minmaxexpr->args))
1306 0 : return true;
1307 : }
1558 tgl 1308 UBC 0 : break;
1309 :
2816 mail 1310 GIC 15 : case T_CurrentOfExpr:
1311 :
1312 : /*
2272 tgl 1313 EUB : * WHERE CURRENT OF doesn't contain leaky function calls.
1314 : * Moreover, it is essential that this is considered non-leaky,
1315 : * since the planner must always generate a TID scan when CURRENT
1914 peter_e 1316 : * OF is present -- cf. cost_tidscan.
1317 : */
2816 mail 1318 GIC 15 : return false;
1319 :
4073 rhaas 1320 12 : default:
1321 :
1322 : /*
4073 rhaas 1323 EUB : * If we don't recognize the node tag, assume it might be leaky.
1324 : * This prevents an unexpected security hole if someone adds a new
1325 : * node type that can call a function.
1326 : */
4073 rhaas 1327 GBC 12 : return true;
4073 rhaas 1328 EUB : }
2904 sfrost 1329 GIC 3212 : return expression_tree_walker(node, contain_leaked_vars_walker,
4073 rhaas 1330 EUB : context);
1331 : }
7434 tgl 1332 ECB :
1333 : /*
1334 : * find_nonnullable_rels
1335 : * Determine which base rels are forced nonnullable by given clause.
1336 : *
1337 : * Returns the set of all Relids that are referenced in the clause in such
1338 : * a way that the clause cannot possibly return TRUE if any of these Relids
1339 : * is an all-NULL row. (It is OK to err on the side of conservatism; hence
6319 1340 : * the analysis here is simplistic.)
1341 : *
1342 : * The semantics here are subtly different from contain_nonstrict_functions:
1343 : * that function is concerned with NULL results from arbitrary expressions,
1344 : * but here we assume that the input is a Boolean expression, and wish to
1345 : * see if NULL inputs will provably cause a FALSE-or-NULL result. We expect
1346 : * the expression to have been AND/OR flattened and converted to implicit-AND
1347 : * format.
1348 : *
5351 1349 : * Note: this function is largely duplicative of find_nonnullable_vars().
1350 : * The reason not to simplify this function into a thin wrapper around
1351 : * find_nonnullable_vars() is that the tested conditions really are different:
1352 : * a clause like "t1.v1 IS NOT NULL OR t1.v2 IS NOT NULL" does not prove
1353 : * that either v1 or v2 can't be NULL, but it does prove that the t1 row
1354 : * as a whole can't be all-NULL. Also, the behavior for PHVs is different.
1355 : *
1356 : * top_level is true while scanning top-level AND/OR structure; here, showing
1357 : * the result is either FALSE or NULL is good enough. top_level is false when
1358 : * we have descended below a NOT or a strict function: now we must be able to
1359 : * prove that the subexpression goes to NULL.
1360 : *
1361 : * We don't use expression_tree_walker here because we don't want to descend
1362 : * through very many kinds of nodes; only the ones we can be sure are strict.
1363 : */
1364 : Relids
6319 tgl 1365 GIC 39221 : find_nonnullable_rels(Node *clause)
1366 : {
1367 39221 : return find_nonnullable_rels_walker(clause, true);
1368 : }
1369 :
1370 : static Relids
1371 249650 : find_nonnullable_rels_walker(Node *node, bool top_level)
1372 : {
1373 249650 : Relids result = NULL;
1374 : ListCell *l;
1375 :
1376 249650 : if (node == NULL)
1377 2389 : return NULL;
1378 247261 : if (IsA(node, Var))
1379 : {
1380 81540 : Var *var = (Var *) node;
1381 :
1382 81540 : if (var->varlevelsup == 0)
1383 81540 : result = bms_make_singleton(var->varno);
1384 : }
1385 165721 : else if (IsA(node, List))
1386 : {
5896 tgl 1387 ECB : /*
1388 : * At top level, we are examining an implicit-AND list: if any of the
5624 bruce 1389 : * arms produces FALSE-or-NULL then the result is FALSE-or-NULL. If
1390 : * not at top level, we are examining the arguments of a strict
1391 : * function: if any of them produce NULL then the result of the
1392 : * function must be NULL. So in both cases, the set of nonnullable
1393 : * rels is the union of those found in the arms, and we pass down the
1394 : * top_level flag unmodified.
5896 tgl 1395 : */
6319 tgl 1396 GIC 242694 : foreach(l, (List *) node)
1397 : {
6319 tgl 1398 CBC 153755 : result = bms_join(result,
1399 153755 : find_nonnullable_rels_walker(lfirst(l),
6319 tgl 1400 ECB : top_level));
1401 : }
1402 : }
6319 tgl 1403 GIC 76782 : else if (IsA(node, FuncExpr))
6319 tgl 1404 ECB : {
6319 tgl 1405 CBC 2863 : FuncExpr *expr = (FuncExpr *) node;
1406 :
1407 2863 : if (func_strict(expr->funcid))
6319 tgl 1408 GIC 2785 : result = find_nonnullable_rels_walker((Node *) expr->args, false);
1409 : }
1410 73919 : else if (IsA(node, OpExpr))
1411 : {
1412 45353 : OpExpr *expr = (OpExpr *) node;
1413 :
5617 1414 45353 : set_opfuncid(expr);
1415 45353 : if (func_strict(expr->opfuncid))
6319 1416 45353 : result = find_nonnullable_rels_walker((Node *) expr->args, false);
1417 : }
6319 tgl 1418 CBC 28566 : else if (IsA(node, ScalarArrayOpExpr))
1419 : {
1420 3059 : ScalarArrayOpExpr *expr = (ScalarArrayOpExpr *) node;
6319 tgl 1421 ECB :
6271 tgl 1422 GIC 3059 : if (is_strict_saop(expr, true))
6319 1423 3059 : result = find_nonnullable_rels_walker((Node *) expr->args, false);
1424 : }
6319 tgl 1425 CBC 25507 : else if (IsA(node, BoolExpr))
1426 : {
1427 2186 : BoolExpr *expr = (BoolExpr *) node;
1428 :
5896 1429 2186 : switch (expr->boolop)
5896 tgl 1430 ECB : {
5896 tgl 1431 GIC 160 : case AND_EXPR:
5896 tgl 1432 ECB : /* At top level we can just recurse (to the List case) */
5896 tgl 1433 GIC 160 : if (top_level)
5896 tgl 1434 ECB : {
5896 tgl 1435 GIC 160 : result = find_nonnullable_rels_walker((Node *) expr->args,
5896 tgl 1436 ECB : top_level);
5896 tgl 1437 CBC 160 : break;
5896 tgl 1438 ECB : }
1439 :
1440 : /*
1441 : * Below top level, even if one arm produces NULL, the result
1442 : * could be FALSE (hence not NULL). However, if *all* the
1443 : * arms produce NULL then the result is NULL, so we can take
5624 bruce 1444 : * the intersection of the sets of nonnullable rels, just as
3260 1445 : * for OR. Fall through to share code.
1446 : */
1061 alvherre 1447 : /* FALL THRU */
1448 : case OR_EXPR:
5624 bruce 1449 :
1450 : /*
5896 tgl 1451 : * OR is strict if all of its arms are, so we can take the
1452 : * intersection of the sets of nonnullable rels for each arm.
1453 : * This works for both values of top_level.
1454 : */
5896 tgl 1455 CBC 2819 : foreach(l, expr->args)
1456 : {
5896 tgl 1457 ECB : Relids subresult;
1458 :
5896 tgl 1459 CBC 2534 : subresult = find_nonnullable_rels_walker(lfirst(l),
1460 : top_level);
5624 bruce 1461 GIC 2534 : if (result == NULL) /* first subresult? */
5896 tgl 1462 1276 : result = subresult;
1463 : else
1464 1258 : result = bms_int_members(result, subresult);
1465 :
1466 : /*
1467 : * If the intersection is empty, we can stop looking. This
1468 : * also justifies the test for first-subresult above.
1469 : */
1470 2534 : if (bms_is_empty(result))
1471 991 : break;
1472 : }
1473 1276 : break;
1474 750 : case NOT_EXPR:
1475 : /* NOT will return null if its arg is null */
1476 750 : result = find_nonnullable_rels_walker((Node *) expr->args,
5896 tgl 1477 ECB : false);
5896 tgl 1478 GIC 750 : break;
5896 tgl 1479 UIC 0 : default:
1480 0 : elog(ERROR, "unrecognized boolop: %d", (int) expr->boolop);
5896 tgl 1481 ECB : break;
1482 : }
6319 1483 : }
6319 tgl 1484 CBC 23321 : else if (IsA(node, RelabelType))
1485 : {
1486 680 : RelabelType *expr = (RelabelType *) node;
1487 :
6319 tgl 1488 GIC 680 : result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
1489 : }
5787 1490 22641 : else if (IsA(node, CoerceViaIO))
1491 : {
5787 tgl 1492 ECB : /* not clear this is useful, but it can't hurt */
5787 tgl 1493 CBC 33 : CoerceViaIO *expr = (CoerceViaIO *) node;
1494 :
1495 33 : result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
5787 tgl 1496 ECB : }
5857 tgl 1497 GIC 22608 : else if (IsA(node, ArrayCoerceExpr))
5857 tgl 1498 ECB : {
1499 : /* ArrayCoerceExpr is strict at the array level; ignore elemexpr */
5857 tgl 1500 LBC 0 : ArrayCoerceExpr *expr = (ArrayCoerceExpr *) node;
5857 tgl 1501 EUB :
5857 tgl 1502 UBC 0 : result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
1503 : }
6319 tgl 1504 GIC 22608 : else if (IsA(node, ConvertRowtypeExpr))
1505 : {
6319 tgl 1506 ECB : /* not clear this is useful, but it can't hurt */
6319 tgl 1507 UIC 0 : ConvertRowtypeExpr *expr = (ConvertRowtypeExpr *) node;
6319 tgl 1508 ECB :
6319 tgl 1509 UIC 0 : result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
6319 tgl 1510 ECB : }
4412 tgl 1511 GIC 22608 : else if (IsA(node, CollateExpr))
4412 tgl 1512 ECB : {
4412 tgl 1513 UIC 0 : CollateExpr *expr = (CollateExpr *) node;
1514 :
4412 tgl 1515 LBC 0 : result = find_nonnullable_rels_walker((Node *) expr->arg, top_level);
1516 : }
6319 tgl 1517 CBC 22608 : else if (IsA(node, NullTest))
1518 : {
5896 tgl 1519 ECB : /* IS NOT NULL can be considered strict, but only at top level */
6319 tgl 1520 GIC 1852 : NullTest *expr = (NullTest *) node;
1521 :
4846 tgl 1522 GBC 1852 : if (top_level && expr->nulltesttype == IS_NOT_NULL && !expr->argisrow)
6319 tgl 1523 GIC 1131 : result = find_nonnullable_rels_walker((Node *) expr->arg, false);
6319 tgl 1524 EUB : }
6319 tgl 1525 GIC 20756 : else if (IsA(node, BooleanTest))
6319 tgl 1526 ECB : {
1527 : /* Boolean tests that reject NULL are strict at top level */
6319 tgl 1528 GIC 3 : BooleanTest *expr = (BooleanTest *) node;
6319 tgl 1529 EUB :
6319 tgl 1530 GIC 3 : if (top_level &&
6319 tgl 1531 GBC 3 : (expr->booltesttype == IS_TRUE ||
6319 tgl 1532 GIC 3 : expr->booltesttype == IS_FALSE ||
6319 tgl 1533 CBC 3 : expr->booltesttype == IS_NOT_UNKNOWN))
6319 tgl 1534 UIC 0 : result = find_nonnullable_rels_walker((Node *) expr->arg, false);
6319 tgl 1535 EUB : }
155 tgl 1536 GNC 20753 : else if (IsA(node, SubPlan))
1537 : {
1538 34 : SubPlan *splan = (SubPlan *) node;
1539 :
1540 : /*
1541 : * For some types of SubPlan, we can infer strictness from Vars in the
1542 : * testexpr (the LHS of the original SubLink).
1543 : *
1544 : * For ANY_SUBLINK, if the subquery produces zero rows, the result is
1545 : * always FALSE. If the subquery produces more than one row, the
1546 : * per-row results of the testexpr are combined using OR semantics.
1547 : * Hence ANY_SUBLINK can be strict only at top level, but there it's
1548 : * as strict as the testexpr is.
1549 : *
1550 : * For ROWCOMPARE_SUBLINK, if the subquery produces zero rows, the
1551 : * result is always NULL. Otherwise, the result is as strict as the
1552 : * testexpr is. So we can check regardless of top_level.
1553 : *
1554 : * We can't prove anything for other sublink types (in particular,
1555 : * note that ALL_SUBLINK will return TRUE if the subquery is empty).
1556 : */
1557 34 : if ((top_level && splan->subLinkType == ANY_SUBLINK) ||
1558 28 : splan->subLinkType == ROWCOMPARE_SUBLINK)
1559 6 : result = find_nonnullable_rels_walker(splan->testexpr, top_level);
1560 : }
5283 tgl 1561 GIC 20719 : else if (IsA(node, PlaceHolderVar))
5283 tgl 1562 EUB : {
5283 tgl 1563 GIC 183 : PlaceHolderVar *phv = (PlaceHolderVar *) node;
5283 tgl 1564 ECB :
1565 : /*
1566 : * If the contained expression forces any rels non-nullable, so does
1532 1567 : * the PHV.
1568 : */
5283 tgl 1569 CBC 183 : result = find_nonnullable_rels_walker((Node *) phv->phexpr, top_level);
1532 tgl 1570 ECB :
1571 : /*
1572 : * If the PHV's syntactic scope is exactly one rel, it will be forced
1573 : * to be evaluated at that rel, and so it will behave like a Var of
1574 : * that rel: if the rel's entire output goes to null, so will the PHV.
1575 : * (If the syntactic scope is a join, we know that the PHV will go to
1576 : * null if the whole join does; but that is AND semantics while we
1577 : * need OR semantics for find_nonnullable_rels' result, so we can't do
1578 : * anything with the knowledge.)
1579 : */
1532 tgl 1580 CBC 366 : if (phv->phlevelsup == 0 &&
1532 tgl 1581 GBC 183 : bms_membership(phv->phrels) == BMS_SINGLETON)
1532 tgl 1582 GIC 105 : result = bms_add_members(result, phv->phrels);
5283 tgl 1583 ECB : }
5351 tgl 1584 GIC 247261 : return result;
5351 tgl 1585 ECB : }
1586 :
1587 : /*
1588 : * find_nonnullable_vars
1589 : * Determine which Vars are forced nonnullable by given clause.
1590 : *
1591 : * Returns the set of all level-zero Vars that are referenced in the clause in
1592 : * such a way that the clause cannot possibly return TRUE if any of these Vars
1593 : * is NULL. (It is OK to err on the side of conservatism; hence the analysis
1594 : * here is simplistic.)
1595 : *
1596 : * The semantics here are subtly different from contain_nonstrict_functions:
1597 : * that function is concerned with NULL results from arbitrary expressions,
1598 : * but here we assume that the input is a Boolean expression, and wish to
1599 : * see if NULL inputs will provably cause a FALSE-or-NULL result. We expect
1600 : * the expression to have been AND/OR flattened and converted to implicit-AND
1601 : * format.
1602 : *
1603 : * Attnos of the identified Vars are returned in a multibitmapset (a List of
1604 : * Bitmapsets). List indexes correspond to relids (varnos), while the per-rel
1605 : * Bitmapsets hold varattnos offset by FirstLowInvalidHeapAttributeNumber.
1606 : *
2062 peter_e 1607 : * top_level is true while scanning top-level AND/OR structure; here, showing
1608 : * the result is either FALSE or NULL is good enough. top_level is false when
5351 tgl 1609 : * we have descended below a NOT or a strict function: now we must be able to
1610 : * prove that the subexpression goes to NULL.
1611 : *
1612 : * We don't use expression_tree_walker here because we don't want to descend
1613 : * through very many kinds of nodes; only the ones we can be sure are strict.
1614 : */
1615 : List *
5351 tgl 1616 GIC 17447 : find_nonnullable_vars(Node *clause)
5351 tgl 1617 ECB : {
5351 tgl 1618 GIC 17447 : return find_nonnullable_vars_walker(clause, true);
1619 : }
1620 :
1621 : static List *
1622 111920 : find_nonnullable_vars_walker(Node *node, bool top_level)
1623 : {
1624 111920 : List *result = NIL;
1625 : ListCell *l;
1626 :
1627 111920 : if (node == NULL)
5351 tgl 1628 CBC 133 : return NIL;
1629 111787 : if (IsA(node, Var))
5351 tgl 1630 ECB : {
5351 tgl 1631 GIC 42561 : Var *var = (Var *) node;
5351 tgl 1632 ECB :
5351 tgl 1633 GIC 42561 : if (var->varlevelsup == 0)
144 tgl 1634 GNC 42561 : result = mbms_add_member(result,
1635 : var->varno,
1636 42561 : var->varattno - FirstLowInvalidHeapAttributeNumber);
1637 : }
5351 tgl 1638 GIC 69226 : else if (IsA(node, List))
1639 : {
1640 : /*
1641 : * At top level, we are examining an implicit-AND list: if any of the
1642 : * arms produces FALSE-or-NULL then the result is FALSE-or-NULL. If
1643 : * not at top level, we are examining the arguments of a strict
1644 : * function: if any of them produce NULL then the result of the
1645 : * function must be NULL. So in both cases, the set of nonnullable
1646 : * vars is the union of those found in the arms, and we pass down the
1647 : * top_level flag unmodified.
1648 : */
1649 111234 : foreach(l, (List *) node)
1650 : {
144 tgl 1651 GNC 70573 : result = mbms_add_members(result,
1652 70573 : find_nonnullable_vars_walker(lfirst(l),
1653 : top_level));
1654 : }
1655 : }
5351 tgl 1656 GIC 28565 : else if (IsA(node, FuncExpr))
1657 : {
1658 169 : FuncExpr *expr = (FuncExpr *) node;
1659 :
1660 169 : if (func_strict(expr->funcid))
1661 169 : result = find_nonnullable_vars_walker((Node *) expr->args, false);
1662 : }
1663 28396 : else if (IsA(node, OpExpr))
1664 : {
1665 22550 : OpExpr *expr = (OpExpr *) node;
5351 tgl 1666 ECB :
5351 tgl 1667 GIC 22550 : set_opfuncid(expr);
5351 tgl 1668 CBC 22550 : if (func_strict(expr->opfuncid))
5351 tgl 1669 GIC 22550 : result = find_nonnullable_vars_walker((Node *) expr->args, false);
1670 : }
1671 5846 : else if (IsA(node, ScalarArrayOpExpr))
5351 tgl 1672 ECB : {
5351 tgl 1673 GIC 607 : ScalarArrayOpExpr *expr = (ScalarArrayOpExpr *) node;
5351 tgl 1674 ECB :
5351 tgl 1675 GIC 607 : if (is_strict_saop(expr, true))
1676 607 : result = find_nonnullable_vars_walker((Node *) expr->args, false);
5351 tgl 1677 ECB : }
5351 tgl 1678 CBC 5239 : else if (IsA(node, BoolExpr))
5351 tgl 1679 ECB : {
5351 tgl 1680 GIC 132 : BoolExpr *expr = (BoolExpr *) node;
5351 tgl 1681 ECB :
5351 tgl 1682 GIC 132 : switch (expr->boolop)
5351 tgl 1683 ECB : {
5351 tgl 1684 LBC 0 : case AND_EXPR:
1685 :
1686 : /*
1687 : * At top level we can just recurse (to the List case), since
1688 : * the result should be the union of what we can prove in each
1689 : * arm.
1690 : */
1691 0 : if (top_level)
1692 : {
1693 0 : result = find_nonnullable_vars_walker((Node *) expr->args,
1694 : top_level);
5351 tgl 1695 UIC 0 : break;
1696 : }
1697 :
1698 : /*
1699 : * Below top level, even if one arm produces NULL, the result
1700 : * could be FALSE (hence not NULL). However, if *all* the
1701 : * arms produce NULL then the result is NULL, so we can take
1702 : * the intersection of the sets of nonnullable vars, just as
1703 : * for OR. Fall through to share code.
5351 tgl 1704 ECB : */
1705 : /* FALL THRU */
1706 : case OR_EXPR:
1707 :
1708 : /*
1709 : * OR is strict if all of its arms are, so we can take the
1710 : * intersection of the sets of nonnullable vars for each arm.
1711 : * This works for both values of top_level.
1712 : */
5351 tgl 1713 CBC 293 : foreach(l, expr->args)
1714 : {
5351 tgl 1715 ECB : List *subresult;
1716 :
5351 tgl 1717 GIC 236 : subresult = find_nonnullable_vars_walker(lfirst(l),
5351 tgl 1718 ECB : top_level);
5351 tgl 1719 GIC 236 : if (result == NIL) /* first subresult? */
5351 tgl 1720 CBC 111 : result = subresult;
1721 : else
144 tgl 1722 GNC 125 : result = mbms_int_members(result, subresult);
5351 tgl 1723 ECB :
1724 : /*
1725 : * If the intersection is empty, we can stop looking. This
1726 : * also justifies the test for first-subresult above.
1727 : */
5351 tgl 1728 CBC 236 : if (result == NIL)
5351 tgl 1729 GIC 54 : break;
5351 tgl 1730 ECB : }
5351 tgl 1731 CBC 111 : break;
5351 tgl 1732 GIC 21 : case NOT_EXPR:
5351 tgl 1733 ECB : /* NOT will return null if its arg is null */
5351 tgl 1734 GIC 21 : result = find_nonnullable_vars_walker((Node *) expr->args,
5351 tgl 1735 ECB : false);
5351 tgl 1736 GIC 21 : break;
5351 tgl 1737 LBC 0 : default:
5351 tgl 1738 UIC 0 : elog(ERROR, "unrecognized boolop: %d", (int) expr->boolop);
5351 tgl 1739 EUB : break;
1740 : }
1741 : }
5351 tgl 1742 GIC 5107 : else if (IsA(node, RelabelType))
1743 : {
1744 251 : RelabelType *expr = (RelabelType *) node;
1745 :
5351 tgl 1746 GBC 251 : result = find_nonnullable_vars_walker((Node *) expr->arg, top_level);
1747 : }
1748 4856 : else if (IsA(node, CoerceViaIO))
1749 : {
5351 tgl 1750 EUB : /* not clear this is useful, but it can't hurt */
5351 tgl 1751 GIC 6 : CoerceViaIO *expr = (CoerceViaIO *) node;
1752 :
1753 6 : result = find_nonnullable_vars_walker((Node *) expr->arg, false);
1754 : }
1755 4850 : else if (IsA(node, ArrayCoerceExpr))
1756 : {
1757 : /* ArrayCoerceExpr is strict at the array level; ignore elemexpr */
5351 tgl 1758 UIC 0 : ArrayCoerceExpr *expr = (ArrayCoerceExpr *) node;
1759 :
1760 0 : result = find_nonnullable_vars_walker((Node *) expr->arg, top_level);
1761 : }
5351 tgl 1762 GIC 4850 : else if (IsA(node, ConvertRowtypeExpr))
1763 : {
1764 : /* not clear this is useful, but it can't hurt */
5351 tgl 1765 UIC 0 : ConvertRowtypeExpr *expr = (ConvertRowtypeExpr *) node;
1766 :
1767 0 : result = find_nonnullable_vars_walker((Node *) expr->arg, top_level);
5351 tgl 1768 ECB : }
4412 tgl 1769 GIC 4850 : else if (IsA(node, CollateExpr))
1770 : {
4412 tgl 1771 UIC 0 : CollateExpr *expr = (CollateExpr *) node;
4412 tgl 1772 ECB :
4412 tgl 1773 UIC 0 : result = find_nonnullable_vars_walker((Node *) expr->arg, top_level);
4412 tgl 1774 ECB : }
5351 tgl 1775 CBC 4850 : else if (IsA(node, NullTest))
1776 : {
5351 tgl 1777 ECB : /* IS NOT NULL can be considered strict, but only at top level */
5351 tgl 1778 GIC 87 : NullTest *expr = (NullTest *) node;
1779 :
4846 1780 87 : if (top_level && expr->nulltesttype == IS_NOT_NULL && !expr->argisrow)
5351 1781 33 : result = find_nonnullable_vars_walker((Node *) expr->arg, false);
1782 : }
5351 tgl 1783 CBC 4763 : else if (IsA(node, BooleanTest))
5351 tgl 1784 ECB : {
1785 : /* Boolean tests that reject NULL are strict at top level */
5351 tgl 1786 LBC 0 : BooleanTest *expr = (BooleanTest *) node;
5351 tgl 1787 ECB :
5351 tgl 1788 UIC 0 : if (top_level &&
5351 tgl 1789 LBC 0 : (expr->booltesttype == IS_TRUE ||
5351 tgl 1790 UIC 0 : expr->booltesttype == IS_FALSE ||
5351 tgl 1791 LBC 0 : expr->booltesttype == IS_NOT_UNKNOWN))
5351 tgl 1792 UBC 0 : result = find_nonnullable_vars_walker((Node *) expr->arg, false);
5351 tgl 1793 EUB : }
155 tgl 1794 GNC 4763 : else if (IsA(node, SubPlan))
1795 : {
155 tgl 1796 UNC 0 : SubPlan *splan = (SubPlan *) node;
1797 :
1798 : /* See analysis in find_nonnullable_rels_walker */
1799 0 : if ((top_level && splan->subLinkType == ANY_SUBLINK) ||
1800 0 : splan->subLinkType == ROWCOMPARE_SUBLINK)
1801 0 : result = find_nonnullable_vars_walker(splan->testexpr, top_level);
1802 : }
5283 tgl 1803 GIC 4763 : else if (IsA(node, PlaceHolderVar))
1804 : {
1805 27 : PlaceHolderVar *phv = (PlaceHolderVar *) node;
5283 tgl 1806 ECB :
5283 tgl 1807 GIC 27 : result = find_nonnullable_vars_walker((Node *) phv->phexpr, top_level);
5283 tgl 1808 ECB : }
6319 tgl 1809 GIC 111787 : return result;
6319 tgl 1810 ECB : }
1811 :
5351 1812 : /*
1813 : * find_forced_null_vars
1814 : * Determine which Vars must be NULL for the given clause to return TRUE.
1815 : *
1816 : * This is the complement of find_nonnullable_vars: find the level-zero Vars
1817 : * that must be NULL for the clause to return TRUE. (It is OK to err on the
1818 : * side of conservatism; hence the analysis here is simplistic. In fact,
1819 : * we only detect simple "var IS NULL" tests at the top level.)
1820 : *
1821 : * As with find_nonnullable_vars, we return the varattnos of the identified
1822 : * Vars in a multibitmapset.
1823 : */
5351 tgl 1824 EUB : List *
5351 tgl 1825 GIC 44485 : find_forced_null_vars(Node *node)
5351 tgl 1826 ECB : {
5351 tgl 1827 GIC 44485 : List *result = NIL;
1828 : Var *var;
5351 tgl 1829 EUB : ListCell *l;
1830 :
5351 tgl 1831 GBC 44485 : if (node == NULL)
5351 tgl 1832 GIC 1937 : return NIL;
5351 tgl 1833 ECB : /* Check single-clause cases using subroutine */
5351 tgl 1834 GIC 42548 : var = find_forced_null_var(node);
5351 tgl 1835 GBC 42548 : if (var)
1836 : {
144 tgl 1837 GNC 541 : result = mbms_add_member(result,
1838 : var->varno,
1839 541 : var->varattno - FirstLowInvalidHeapAttributeNumber);
1840 : }
5351 tgl 1841 ECB : /* Otherwise, handle AND-conditions */
5351 tgl 1842 GIC 42007 : else if (IsA(node, List))
1843 : {
5351 tgl 1844 ECB : /*
1845 : * At top level, we are examining an implicit-AND list: if any of the
1846 : * arms produces FALSE-or-NULL then the result is FALSE-or-NULL.
1847 : */
5351 tgl 1848 GIC 42548 : foreach(l, (List *) node)
5351 tgl 1849 ECB : {
144 tgl 1850 GNC 25011 : result = mbms_add_members(result,
1851 25011 : find_forced_null_vars((Node *) lfirst(l)));
5351 tgl 1852 EUB : }
1853 : }
5351 tgl 1854 GBC 24470 : else if (IsA(node, BoolExpr))
5351 tgl 1855 EUB : {
5351 tgl 1856 GBC 1698 : BoolExpr *expr = (BoolExpr *) node;
5351 tgl 1857 EUB :
1858 : /*
1859 : * We don't bother considering the OR case, because it's fairly
5050 bruce 1860 ECB : * unlikely anyone would write "v1 IS NULL OR v1 IS NULL". Likewise,
1861 : * the NOT case isn't worth expending code on.
5351 tgl 1862 EUB : */
5351 tgl 1863 GIC 1698 : if (expr->boolop == AND_EXPR)
1864 : {
5351 tgl 1865 EUB : /* At top level we can just recurse (to the List case) */
5351 tgl 1866 UBC 0 : result = find_forced_null_vars((Node *) expr->args);
5351 tgl 1867 EUB : }
1868 : }
5351 tgl 1869 CBC 42548 : return result;
1870 : }
5351 tgl 1871 ECB :
1872 : /*
1873 : * find_forced_null_var
1874 : * Return the Var forced null by the given clause, or NULL if it's
1875 : * not an IS NULL-type clause. For success, the clause must enforce
1876 : * *only* nullness of the particular Var, not any other conditions.
1877 : *
1878 : * This is just the single-clause case of find_forced_null_vars(), without
1879 : * any allowance for AND conditions. It's used by initsplan.c on individual
1880 : * qual clauses. The reason for not just applying find_forced_null_vars()
1881 : * is that if an AND of an IS NULL clause with something else were to somehow
1882 : * survive AND/OR flattening, initsplan.c might get fooled into discarding
1883 : * the whole clause when only the IS NULL part of it had been proved redundant.
1884 : */
1885 : Var *
5351 tgl 1886 GIC 216346 : find_forced_null_var(Node *node)
1887 : {
1888 216346 : if (node == NULL)
5351 tgl 1889 UIC 0 : return NULL;
5351 tgl 1890 GIC 216346 : if (IsA(node, NullTest))
5351 tgl 1891 ECB : {
1892 : /* check for var IS NULL */
5351 tgl 1893 CBC 7129 : NullTest *expr = (NullTest *) node;
1894 :
4846 tgl 1895 GIC 7129 : if (expr->nulltesttype == IS_NULL && !expr->argisrow)
1896 : {
5050 bruce 1897 CBC 1787 : Var *var = (Var *) expr->arg;
5351 tgl 1898 ECB :
5351 tgl 1899 GIC 1787 : if (var && IsA(var, Var) &&
5351 tgl 1900 CBC 1742 : var->varlevelsup == 0)
1901 1742 : return var;
1902 : }
5351 tgl 1903 ECB : }
5351 tgl 1904 GIC 209217 : else if (IsA(node, BooleanTest))
5351 tgl 1905 ECB : {
1906 : /* var IS UNKNOWN is equivalent to var IS NULL */
5351 tgl 1907 GIC 87 : BooleanTest *expr = (BooleanTest *) node;
5351 tgl 1908 ECB :
5351 tgl 1909 GIC 87 : if (expr->booltesttype == IS_UNKNOWN)
1910 : {
5050 bruce 1911 3 : Var *var = (Var *) expr->arg;
1912 :
5351 tgl 1913 3 : if (var && IsA(var, Var) &&
5351 tgl 1914 CBC 3 : var->varlevelsup == 0)
5351 tgl 1915 GIC 3 : return var;
5351 tgl 1916 ECB : }
1917 : }
5351 tgl 1918 GIC 214601 : return NULL;
1919 : }
5351 tgl 1920 ECB :
1921 : /*
6271 1922 : * Can we treat a ScalarArrayOpExpr as strict?
1923 : *
1924 : * If "falseOK" is true, then a "false" result can be considered strict,
1925 : * else we need to guarantee an actual NULL result for NULL input.
1926 : *
1927 : * "foo op ALL array" is strict if the op is strict *and* we can prove
1928 : * that the array input isn't an empty array. We can check that
1929 : * for the cases of an array constant and an ARRAY[] construct.
1930 : *
1931 : * "foo op ANY array" is strict in the falseOK sense if the op is strict.
6271 tgl 1932 EUB : * If not falseOK, the test is the same as for "foo op ALL array".
1933 : */
1934 : static bool
6271 tgl 1935 CBC 3666 : is_strict_saop(ScalarArrayOpExpr *expr, bool falseOK)
1936 : {
1937 : Node *rightop;
1938 :
1939 : /* The contained operator must be strict. */
5617 tgl 1940 GIC 3666 : set_sa_opfuncid(expr);
1941 3666 : if (!func_strict(expr->opfuncid))
6271 tgl 1942 UIC 0 : return false;
1943 : /* If ANY and falseOK, that's all we need to check. */
6271 tgl 1944 GIC 3666 : if (expr->useOr && falseOK)
1945 3650 : return true;
1946 : /* Else, we have to see if the array is provably non-empty. */
1947 16 : Assert(list_length(expr->args) == 2);
1948 16 : rightop = (Node *) lsecond(expr->args);
1949 16 : if (rightop && IsA(rightop, Const))
6271 tgl 1950 UIC 0 : {
6271 tgl 1951 GIC 16 : Datum arraydatum = ((Const *) rightop)->constvalue;
6271 tgl 1952 CBC 16 : bool arrayisnull = ((Const *) rightop)->constisnull;
1953 : ArrayType *arrayval;
6271 tgl 1954 ECB : int nitems;
6271 tgl 1955 EUB :
6271 tgl 1956 CBC 16 : if (arrayisnull)
6271 tgl 1957 UIC 0 : return false;
6271 tgl 1958 GIC 16 : arrayval = DatumGetArrayTypeP(arraydatum);
6271 tgl 1959 CBC 16 : nitems = ArrayGetNItems(ARR_NDIM(arrayval), ARR_DIMS(arrayval));
6271 tgl 1960 GIC 16 : if (nitems > 0)
6271 tgl 1961 CBC 16 : return true;
1962 : }
6271 tgl 1963 LBC 0 : else if (rightop && IsA(rightop, ArrayExpr))
1964 : {
1965 0 : ArrayExpr *arrayexpr = (ArrayExpr *) rightop;
6271 tgl 1966 ECB :
6271 tgl 1967 LBC 0 : if (arrayexpr->elements != NIL && !arrayexpr->multidims)
6271 tgl 1968 UIC 0 : return true;
1969 : }
6271 tgl 1970 LBC 0 : return false;
1971 : }
1972 :
6319 tgl 1973 ECB :
1974 : /*****************************************************************************
8274 1975 : * Check for "pseudo-constant" clauses
1976 : *****************************************************************************/
8405 1977 :
1978 : /*
8274 1979 : * is_pseudo_constant_clause
6126 1980 : * Detect whether an expression is "pseudo constant", ie, it contains no
1981 : * variables of the current query level and no uses of volatile functions.
1982 : * Such an expr is not necessarily a true constant: it can still contain
1983 : * Params and outer-level Vars, not to mention functions whose results
3260 bruce 1984 : * may vary from one statement to the next. However, the expr's value
1985 : * will be constant over any one scan of the current query, so it can be
1986 : * used as, eg, an indexscan key. (Actually, the condition for indexscan
1987 : * keys is weaker than this; see is_pseudo_constant_for_index().)
1988 : *
1989 : * CAUTION: this function omits to test for one very important class of
1990 : * not-constant expressions, namely aggregates (Aggrefs). In current usage
1991 : * this is only applied to WHERE clauses and so a check for Aggrefs would be
1992 : * a waste of cycles; but be sure to also check contain_agg_clause() if you
1993 : * want to know about pseudo-constness in other contexts. The same goes
1994 : * for window functions (WindowFuncs).
1995 : */
1996 : bool
8274 tgl 1997 GIC 2403 : is_pseudo_constant_clause(Node *clause)
1998 : {
1999 : /*
2000 : * We could implement this check in one recursive scan. But since the
6385 bruce 2001 ECB : * check for volatile functions is both moderately expensive and unlikely
2002 : * to fail, it seems better to look for Vars first and only check for
2003 : * volatile functions if we find no Vars.
2004 : */
8274 tgl 2005 GIC 2403 : if (!contain_var_clause(clause) &&
7674 tgl 2006 CBC 2403 : !contain_volatile_functions(clause))
8274 2007 2403 : return true;
8274 tgl 2008 UBC 0 : return false;
2009 : }
8274 tgl 2010 ECB :
7040 2011 : /*
2012 : * is_pseudo_constant_clause_relids
2013 : * Same as above, except caller already has available the var membership
6126 2014 : * of the expression; this lets us avoid the contain_var_clause() scan.
7040 2015 : */
7040 tgl 2016 EUB : bool
7040 tgl 2017 CBC 140429 : is_pseudo_constant_clause_relids(Node *clause, Relids relids)
7040 tgl 2018 ECB : {
7040 tgl 2019 GIC 140429 : if (bms_is_empty(relids) &&
2020 138041 : !contain_volatile_functions(clause))
2021 138041 : return true;
7040 tgl 2022 CBC 2388 : return false;
7040 tgl 2023 EUB : }
7040 tgl 2024 ECB :
8405 2025 :
8274 2026 : /*****************************************************************************
2027 : * *
2028 : * General clause-manipulating routines *
8274 tgl 2029 EUB : * *
2030 : *****************************************************************************/
2031 :
2032 : /*
8821 bruce 2033 : * NumRelids
2034 : * (formerly clause_relids)
2035 : *
2036 : * Returns the number of different base relations referenced in 'clause'.
2037 : */
2038 : int
808 tgl 2039 GIC 816 : NumRelids(PlannerInfo *root, Node *clause)
2040 : {
2041 : int result;
2042 816 : Relids varnos = pull_varnos(root, clause);
2043 :
69 tgl 2044 GNC 816 : varnos = bms_del_members(varnos, root->outer_join_rels);
2045 816 : result = bms_num_members(varnos);
7365 tgl 2046 GIC 816 : bms_free(varnos);
8644 2047 816 : return result;
2048 : }
2049 :
2050 : /*
2051 : * CommuteOpExpr: commute a binary operator clause
2052 : *
2053 : * XXX the clause is destructively modified!
2054 : */
2055 : void
6283 2056 7721 : CommuteOpExpr(OpExpr *clause)
2057 : {
2058 : Oid opoid;
2059 : Node *temp;
2060 :
2061 : /* Sanity checks: caller is at fault if these fail */
7423 2062 15442 : if (!is_opclause(clause) ||
6888 neilc 2063 7721 : list_length(clause->args) != 2)
7198 tgl 2064 UIC 0 : elog(ERROR, "cannot commute non-binary-operator clause");
9770 scrappy 2065 ECB :
7423 tgl 2066 GIC 7721 : opoid = get_commutator(clause->opno);
2067 :
2068 7721 : if (!OidIsValid(opoid))
7198 tgl 2069 UIC 0 : elog(ERROR, "could not find commutator for operator %u",
2070 : clause->opno);
2071 :
2072 : /*
7423 tgl 2073 ECB : * modify the clause in-place!
9345 bruce 2074 : */
7423 tgl 2075 CBC 7721 : clause->opno = opoid;
7423 tgl 2076 GBC 7721 : clause->opfuncid = InvalidOid;
2077 : /* opresulttype, opretset, opcollid, inputcollid need not change */
2078 :
6892 neilc 2079 GIC 7721 : temp = linitial(clause->args);
2080 7721 : linitial(clause->args) = lsecond(clause->args);
8641 tgl 2081 7721 : lsecond(clause->args) = temp;
9345 bruce 2082 7721 : }
2083 :
2084 : /*
6809 tgl 2085 ECB : * Helper for eval_const_expressions: check that datatype of an attribute
2086 : * is still what it was when the expression was parsed. This is needed to
2087 : * guard against improper simplification after ALTER COLUMN TYPE. (XXX we
2088 : * may well need to make similar checks elsewhere?)
1991 2089 : *
2090 : * rowtypeid may come from a whole-row Var, and therefore it can be a domain
2091 : * over composite, but for this purpose we only care about checking the type
2092 : * of a contained field.
2093 : */
2094 : static bool
6809 tgl 2095 GIC 260 : rowtype_field_matches(Oid rowtypeid, int fieldnum,
2096 : Oid expectedtype, int32 expectedtypmod,
2097 : Oid expectedcollation)
2098 : {
2099 : TupleDesc tupdesc;
2100 : Form_pg_attribute attr;
2101 :
2102 : /* No issue for RECORD, since there is no way to ALTER such a type */
2103 260 : if (rowtypeid == RECORDOID)
2104 21 : return true;
1991 2105 239 : tupdesc = lookup_rowtype_tupdesc_domain(rowtypeid, -1, false);
6809 2106 239 : if (fieldnum <= 0 || fieldnum > tupdesc->natts)
6141 tgl 2107 ECB : {
6141 tgl 2108 UIC 0 : ReleaseTupleDesc(tupdesc);
6809 2109 0 : return false;
6141 tgl 2110 ECB : }
2058 andres 2111 GIC 239 : attr = TupleDescAttr(tupdesc, fieldnum - 1);
6809 tgl 2112 CBC 239 : if (attr->attisdropped ||
2113 239 : attr->atttypid != expectedtype ||
4443 peter_e 2114 239 : attr->atttypmod != expectedtypmod ||
2115 239 : attr->attcollation != expectedcollation)
2116 : {
6141 tgl 2117 UIC 0 : ReleaseTupleDesc(tupdesc);
6809 2118 0 : return false;
2119 : }
6141 tgl 2120 GIC 239 : ReleaseTupleDesc(tupdesc);
6809 2121 239 : return true;
2122 : }
2123 :
8695 tgl 2124 ECB :
2125 : /*--------------------
2126 : * eval_const_expressions
2127 : *
2128 : * Reduce any recognizably constant subexpressions of the given
2129 : * expression tree, for example "2 + 2" => "4". More interestingly,
8596 2130 : * we can reduce certain boolean expressions even when they contain
2131 : * non-constant subexpressions: "x OR true" => "true" no matter what
3260 bruce 2132 EUB : * the subexpression x is. (XXX We assume that no such subexpression
2133 : * will have important side-effects, which is not necessarily a good
8596 tgl 2134 ECB : * assumption in the presence of user-defined functions; do we need a
2135 : * pg_proc flag that prevents discarding the execution of a function?)
2136 : *
8596 tgl 2137 EUB : * We do understand that certain functions may deliver non-constant
2138 : * results even with constant inputs, "nextval()" being the classic
2139 : * example. Functions that are not marked "immutable" in pg_proc
2140 : * will not be pre-evaluated here, although we will reduce their
2141 : * arguments as far as possible.
2142 : *
4549 tgl 2143 ECB : * Whenever a function is eliminated from the expression by means of
2144 : * constant-expression evaluation or inlining, we add the function to
2145 : * root->glob->invalItems. This ensures the plan is known to depend on
2146 : * such functions, even though they aren't referenced anymore.
2147 : *
8596 2148 : * We assume that the tree has already been type-checked and contains
2149 : * only operators and functions that are reasonable to try to execute.
6586 2150 : *
2151 : * NOTE: "root" can be passed as NULL if the caller never wants to do any
2152 : * Param substitutions nor receive info about inlined functions.
2153 : *
2154 : * NOTE: the planner assumes that this will always flatten nested AND and
2155 : * OR clauses into N-argument form. See comments in prepqual.c.
2156 : *
2157 : * NOTE: another critical effect is that any function calls that require
2158 : * default arguments will be expanded, and named-argument calls will be
2159 : * converted to positional notation. The executor won't handle either.
2160 : *--------------------
2161 : */
2162 : Node *
5486 tgl 2163 CBC 551275 : eval_const_expressions(PlannerInfo *root, Node *node)
2164 : {
2165 : eval_const_expressions_context context;
2166 :
5486 tgl 2167 GIC 551275 : if (root)
2168 394545 : context.boundParams = root->glob->boundParams; /* bound Params */
2169 : else
2170 156730 : context.boundParams = NULL;
4236 tgl 2171 CBC 551275 : context.root = root; /* for inlined-function dependencies */
6876 2172 551275 : context.active_fns = NIL; /* nothing being recursively simplified */
6640 2173 551275 : context.case_val = NULL; /* no CASE being examined */
6876 2174 551275 : context.estimate = false; /* safe transformations only */
6876 tgl 2175 GIC 551275 : return eval_const_expressions_mutator(node, &context);
6876 tgl 2176 EUB : }
2177 :
2178 : #define MIN_ARRAY_SIZE_FOR_HASHED_SAOP 9
731 drowley 2179 ECB : /*--------------------
2180 : * convert_saop_to_hashed_saop
2181 : *
2182 : * Recursively search 'node' for ScalarArrayOpExprs and fill in the hash
2183 : * function for any ScalarArrayOpExpr that looks like it would be useful to
2184 : * evaluate using a hash table rather than a linear search.
731 drowley 2185 EUB : *
2186 : * We'll use a hash table if all of the following conditions are met:
2187 : * 1. The 2nd argument of the array contain only Consts.
206 drowley 2188 ECB : * 2. useOr is true or there is a valid negator operator for the
2189 : * ScalarArrayOpExpr's opno.
2190 : * 3. There's valid hash function for both left and righthand operands and
2191 : * these hash functions are the same.
2192 : * 4. If the array contains enough elements for us to consider it to be
2193 : * worthwhile using a hash table rather than a linear search.
2194 : */
2195 : void
731 drowley 2196 GIC 349805 : convert_saop_to_hashed_saop(Node *node)
2197 : {
2198 349805 : (void) convert_saop_to_hashed_saop_walker(node, NULL);
2199 349805 : }
2200 :
2201 : static bool
2202 2383352 : convert_saop_to_hashed_saop_walker(Node *node, void *context)
2203 : {
2204 2383352 : if (node == NULL)
2205 50386 : return false;
2206 :
2207 2332966 : if (IsA(node, ScalarArrayOpExpr))
2208 : {
2209 11671 : ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) node;
2210 11671 : Expr *arrayarg = (Expr *) lsecond(saop->args);
2211 : Oid lefthashfunc;
2212 : Oid righthashfunc;
2213 :
641 2214 11671 : if (arrayarg && IsA(arrayarg, Const) &&
2215 6122 : !((Const *) arrayarg)->constisnull)
2216 : {
2217 6113 : if (saop->useOr)
2218 : {
2219 5496 : if (get_op_hash_functions(saop->opno, &lefthashfunc, &righthashfunc) &&
2220 5352 : lefthashfunc == righthashfunc)
2221 : {
2222 5352 : Datum arrdatum = ((Const *) arrayarg)->constvalue;
2223 5352 : ArrayType *arr = (ArrayType *) DatumGetPointer(arrdatum);
2224 : int nitems;
2225 :
2226 : /*
2227 : * Only fill in the hash functions if the array looks
2228 : * large enough for it to be worth hashing instead of
2229 : * doing a linear search.
2230 : */
641 drowley 2231 CBC 5352 : nitems = ArrayGetNItems(ARR_NDIM(arr), ARR_DIMS(arr));
2232 :
641 drowley 2233 GIC 5352 : if (nitems >= MIN_ARRAY_SIZE_FOR_HASHED_SAOP)
2234 : {
641 drowley 2235 ECB : /* Looks good. Fill in the hash functions */
641 drowley 2236 CBC 98 : saop->hashfuncid = lefthashfunc;
2237 : }
2238 5920 : return true;
641 drowley 2239 ECB : }
2240 : }
2241 : else /* !saop->useOr */
731 2242 : {
641 drowley 2243 CBC 617 : Oid negator = get_negator(saop->opno);
2244 :
2245 : /*
2246 : * Check if this is a NOT IN using an operator whose negator
2247 : * is hashable. If so we can still build a hash table and
2248 : * just ensure the lookup items are not in the hash table.
2249 : */
641 drowley 2250 GIC 1234 : if (OidIsValid(negator) &&
2251 617 : get_op_hash_functions(negator, &lefthashfunc, &righthashfunc) &&
2252 568 : lefthashfunc == righthashfunc)
2253 : {
2254 568 : Datum arrdatum = ((Const *) arrayarg)->constvalue;
2255 568 : ArrayType *arr = (ArrayType *) DatumGetPointer(arrdatum);
2256 : int nitems;
2257 :
2258 : /*
2259 : * Only fill in the hash functions if the array looks
2260 : * large enough for it to be worth hashing instead of
2261 : * doing a linear search.
2262 : */
2263 568 : nitems = ArrayGetNItems(ARR_NDIM(arr), ARR_DIMS(arr));
641 drowley 2264 ECB :
641 drowley 2265 GIC 568 : if (nitems >= MIN_ARRAY_SIZE_FOR_HASHED_SAOP)
641 drowley 2266 ECB : {
2267 : /* Looks good. Fill in the hash functions */
641 drowley 2268 GIC 35 : saop->hashfuncid = lefthashfunc;
2269 :
641 drowley 2270 ECB : /*
2271 : * Also set the negfuncid. The executor will need
2272 : * that to perform hashtable lookups.
2273 : */
641 drowley 2274 GIC 35 : saop->negfuncid = get_opcode(negator);
641 drowley 2275 ECB : }
641 drowley 2276 GIC 568 : return true;
641 drowley 2277 ECB : }
731 2278 : }
2279 : }
2280 : }
2281 :
731 drowley 2282 CBC 2327046 : return expression_tree_walker(node, convert_saop_to_hashed_saop_walker, NULL);
731 drowley 2283 ECB : }
2284 :
2285 :
2286 : /*--------------------
6876 tgl 2287 : * estimate_expression_value
2288 : *
2289 : * This function attempts to estimate the value of an expression for
2290 : * planning purposes. It is in essence a more aggressive version of
2291 : * eval_const_expressions(): we will perform constant reductions that are
2292 : * not necessarily 100% safe, but are reasonable for estimation purposes.
2293 : *
2294 : * Currently the extra steps that are taken in this mode are:
2295 : * 1. Substitute values for Params, where a bound Param value has been made
2296 : * available by the caller of planner(), even if the Param isn't marked
2297 : * constant. This effectively means that we plan using the first supplied
2298 : * value of the Param.
6725 2299 : * 2. Fold stable, as well as immutable, functions to constants.
2300 : * 3. Reduce PlaceHolderVar nodes to their contained expressions.
2301 : *--------------------
2302 : */
2303 : Node *
5893 tgl 2304 CBC 302994 : estimate_expression_value(PlannerInfo *root, Node *node)
2305 : {
6876 tgl 2306 ECB : eval_const_expressions_context context;
2307 :
2118 tgl 2308 GIC 302994 : context.boundParams = root->glob->boundParams; /* bound Params */
2309 : /* we do not need to mark the plan as depending on inlined functions */
4236 2310 302994 : context.root = NULL;
6876 tgl 2311 CBC 302994 : context.active_fns = NIL; /* nothing being recursively simplified */
6640 tgl 2312 GIC 302994 : context.case_val = NULL; /* no CASE being examined */
6876 2313 302994 : context.estimate = true; /* unsafe transformations OK */
2314 302994 : return eval_const_expressions_mutator(node, &context);
2315 : }
2316 :
2317 : /*
1922 tgl 2318 ECB : * The generic case in eval_const_expressions_mutator is to recurse using
2319 : * expression_tree_mutator, which will copy the given node unchanged but
2320 : * const-simplify its arguments (if any) as far as possible. If the node
2321 : * itself does immutable processing, and each of its arguments were reduced
2322 : * to a Const, we can then reduce it to a Const using evaluate_expr. (Some
2323 : * node types need more complicated logic; for example, a CASE expression
2324 : * might be reducible to a constant even if not all its subtrees are.)
2325 : */
2326 : #define ece_generic_processing(node) \
2327 : expression_tree_mutator((Node *) (node), eval_const_expressions_mutator, \
2328 : (void *) context)
2329 :
2330 : /*
2331 : * Check whether all arguments of the given node were reduced to Consts.
2332 : * By going directly to expression_tree_walker, contain_non_const_walker
2333 : * is not applied to the node itself, only to its children.
2334 : */
2335 : #define ece_all_arguments_const(node) \
2336 : (!expression_tree_walker((Node *) (node), contain_non_const_walker, NULL))
2337 :
2338 : /* Generic macro for applying evaluate_expr */
2339 : #define ece_evaluate_expr(node) \
2340 : ((Node *) evaluate_expr((Expr *) (node), \
2341 : exprType((Node *) (node)), \
2342 : exprTypmod((Node *) (node)), \
2343 : exprCollation((Node *) (node))))
2344 :
2345 : /*
2346 : * Recursive guts of eval_const_expressions/estimate_expression_value
2347 : */
2348 : static Node *
6876 tgl 2349 GIC 3726978 : eval_const_expressions_mutator(Node *node,
6876 tgl 2350 ECB : eval_const_expressions_context *context)
2351 : {
8596 tgl 2352 GIC 3726978 : if (node == NULL)
2353 188350 : return NULL;
4150 2354 3538628 : switch (nodeTag(node))
2355 : {
2356 133359 : case T_Param:
2357 : {
bruce 2358 133359 : Param *param = (Param *) node;
1935 tgl 2359 133359 : ParamListInfo paramLI = context->boundParams;
2360 :
2361 : /* Look to see if we've been given a value for this Param */
4150 bruce 2362 133359 : if (param->paramkind == PARAM_EXTERN &&
1935 tgl 2363 85398 : paramLI != NULL &&
4150 bruce 2364 85398 : param->paramid > 0 &&
1935 tgl 2365 85398 : param->paramid <= paramLI->numParams)
2366 : {
2367 : ParamExternData *prm;
2368 : ParamExternData prmdata;
2369 :
2370 : /*
2371 : * Give hook a chance in case parameter is dynamic. Tell
1935 tgl 2372 ECB : * it that this fetch is speculative, so it should avoid
2373 : * erroring out if parameter is unavailable.
2374 : */
1935 tgl 2375 GIC 85398 : if (paramLI->paramFetch != NULL)
1935 tgl 2376 CBC 3200 : prm = paramLI->paramFetch(paramLI, param->paramid,
2377 : true, &prmdata);
1935 tgl 2378 ECB : else
1935 tgl 2379 CBC 82198 : prm = ¶mLI->params[param->paramid - 1];
4150 bruce 2380 ECB :
1718 tgl 2381 : /*
2382 : * We don't just check OidIsValid, but insist that the
2383 : * fetched type match the Param, just in case the hook did
2384 : * something unexpected. No need to throw an error here
2385 : * though; leave that for runtime.
2386 : */
1718 tgl 2387 GIC 85398 : if (OidIsValid(prm->ptype) &&
2388 85398 : prm->ptype == param->paramtype)
2389 : {
2390 : /* OK to substitute parameter value? */
4150 2391 85397 : if (context->estimate ||
2392 85394 : (prm->pflags & PARAM_FLAG_CONST))
2393 : {
2394 : /*
2395 : * Return a Const representing the param value.
2396 : * Must copy pass-by-ref datatypes, since the
2397 : * Param might be in a memory context
2398 : * shorter-lived than our output plan should be.
2399 : */
2400 : int16 typLen;
2401 : bool typByVal;
2402 : Datum pval;
2403 : Const *con;
2404 :
2405 85394 : get_typlenbyval(param->paramtype,
2406 : &typLen, &typByVal);
bruce 2407 85394 : if (prm->isnull || typByVal)
2408 80582 : pval = prm->value;
2409 : else
2410 4812 : pval = datumCopy(prm->value, typByVal, typLen);
634 tgl 2411 85394 : con = makeConst(param->paramtype,
2412 : param->paramtypmod,
2413 : param->paramcollid,
2414 : (int) typLen,
2415 : pval,
2416 85394 : prm->isnull,
634 tgl 2417 ECB : typByVal);
634 tgl 2418 GIC 85394 : con->location = param->location;
2419 85394 : return (Node *) con;
4150 bruce 2420 ECB : }
2421 : }
6196 tgl 2422 : }
2423 :
4150 bruce 2424 : /*
2425 : * Not replaceable, so just copy the Param (no need to
2426 : * recurse)
2427 : */
4150 bruce 2428 GIC 47965 : return (Node *) copyObject(param);
2429 : }
3441 tgl 2430 CBC 1428 : case T_WindowFunc:
3441 tgl 2431 ECB : {
3441 tgl 2432 CBC 1428 : WindowFunc *expr = (WindowFunc *) node;
2433 1428 : Oid funcid = expr->winfnoid;
2434 : List *args;
2435 : Expr *aggfilter;
2436 : HeapTuple func_tuple;
2437 : WindowFunc *newexpr;
2438 :
2439 : /*
2440 : * We can't really simplify a WindowFunc node, but we mustn't
2441 : * just fall through to the default processing, because we
2442 : * have to apply expand_function_arguments to its argument
3441 tgl 2443 ECB : * list. That takes care of inserting default arguments and
2444 : * expanding named-argument notation.
2445 : */
3441 tgl 2446 GIC 1428 : func_tuple = SearchSysCache1(PROCOID, ObjectIdGetDatum(funcid));
3441 tgl 2447 CBC 1428 : if (!HeapTupleIsValid(func_tuple))
3441 tgl 2448 UIC 0 : elog(ERROR, "cache lookup failed for function %u", funcid);
2449 :
668 tgl 2450 GIC 1428 : args = expand_function_arguments(expr->args,
2451 : false, expr->wintype,
2452 : func_tuple);
2453 :
3441 2454 1428 : ReleaseSysCache(func_tuple);
3441 tgl 2455 ECB :
2456 : /* Now, recursively simplify the args (which are a List) */
2457 : args = (List *)
3441 tgl 2458 GIC 1428 : expression_tree_mutator((Node *) args,
3441 tgl 2459 ECB : eval_const_expressions_mutator,
2460 : (void *) context);
2461 : /* ... and the filter expression, which isn't */
2462 : aggfilter = (Expr *)
3441 tgl 2463 GIC 1428 : eval_const_expressions_mutator((Node *) expr->aggfilter,
2464 : context);
2465 :
2466 : /* And build the replacement WindowFunc node */
2467 1428 : newexpr = makeNode(WindowFunc);
2468 1428 : newexpr->winfnoid = expr->winfnoid;
2469 1428 : newexpr->wintype = expr->wintype;
2470 1428 : newexpr->wincollid = expr->wincollid;
2471 1428 : newexpr->inputcollid = expr->inputcollid;
2472 1428 : newexpr->args = args;
3441 tgl 2473 CBC 1428 : newexpr->aggfilter = aggfilter;
3441 tgl 2474 GIC 1428 : newexpr->winref = expr->winref;
3441 tgl 2475 CBC 1428 : newexpr->winstar = expr->winstar;
2476 1428 : newexpr->winagg = expr->winagg;
3441 tgl 2477 GIC 1428 : newexpr->location = expr->location;
3441 tgl 2478 ECB :
3441 tgl 2479 CBC 1428 : return (Node *) newexpr;
2480 : }
4150 tgl 2481 GIC 245058 : case T_FuncExpr:
2482 : {
bruce 2483 245058 : FuncExpr *expr = (FuncExpr *) node;
4034 tgl 2484 CBC 245058 : List *args = expr->args;
2485 : Expr *simple;
4150 bruce 2486 ECB : FuncExpr *newexpr;
7188 2487 :
2488 : /*
2489 : * Code for op/func reduction is pretty bulky, so split it out
2490 : * as a separate function. Note: exprTypmod normally returns
2491 : * -1 for a FuncExpr, but not when the node is recognizably a
2492 : * length coercion; we want to preserve the typmod in the
2493 : * eventual Const if so.
2494 : */
4034 tgl 2495 GIC 245058 : simple = simplify_function(expr->funcid,
4150 tgl 2496 ECB : expr->funcresulttype,
2497 : exprTypmod(node),
bruce 2498 : expr->funccollid,
2499 : expr->inputcollid,
2500 : &args,
3730 tgl 2501 CBC 245058 : expr->funcvariadic,
2502 : true,
2503 : true,
2504 : context);
4150 bruce 2505 GIC 243885 : if (simple) /* successfully simplified it */
2506 80393 : return (Node *) simple;
2507 :
2508 : /*
2509 : * The expression cannot be simplified any further, so build
2510 : * and return a replacement FuncExpr node using the
2511 : * possibly-simplified arguments. Note that we have also
2512 : * converted the argument list to positional notation.
2513 : */
4150 bruce 2514 CBC 163492 : newexpr = makeNode(FuncExpr);
2515 163492 : newexpr->funcid = expr->funcid;
4150 bruce 2516 GBC 163492 : newexpr->funcresulttype = expr->funcresulttype;
4150 bruce 2517 GIC 163492 : newexpr->funcretset = expr->funcretset;
3730 tgl 2518 CBC 163492 : newexpr->funcvariadic = expr->funcvariadic;
4150 bruce 2519 GIC 163492 : newexpr->funcformat = expr->funcformat;
2520 163492 : newexpr->funccollid = expr->funccollid;
2521 163492 : newexpr->inputcollid = expr->inputcollid;
4150 bruce 2522 CBC 163492 : newexpr->args = args;
4150 bruce 2523 GIC 163492 : newexpr->location = expr->location;
2524 163492 : return (Node *) newexpr;
2525 : }
4150 tgl 2526 CBC 249803 : case T_OpExpr:
2527 : {
4150 bruce 2528 GIC 249803 : OpExpr *expr = (OpExpr *) node;
4034 tgl 2529 249803 : List *args = expr->args;
2530 : Expr *simple;
4150 bruce 2531 ECB : OpExpr *newexpr;
2532 :
2533 : /*
2534 : * Need to get OID of underlying function. Okay to scribble
2535 : * on input to this extent.
2536 : */
4150 bruce 2537 CBC 249803 : set_opfuncid(expr);
7188 bruce 2538 ECB :
4150 2539 : /*
2540 : * Code for op/func reduction is pretty bulky, so split it out
2541 : * as a separate function.
2542 : */
4034 tgl 2543 CBC 249803 : simple = simplify_function(expr->opfuncid,
4150 bruce 2544 ECB : expr->opresulttype, -1,
2545 : expr->opcollid,
2546 : expr->inputcollid,
2547 : &args,
2548 : false,
4034 tgl 2549 : true,
2550 : true,
2551 : context);
4150 bruce 2552 CBC 249419 : if (simple) /* successfully simplified it */
4150 bruce 2553 GIC 8277 : return (Node *) simple;
2554 :
2555 : /*
2556 : * If the operator is boolean equality or inequality, we know
2557 : * how to simplify cases involving one constant and one
2558 : * non-constant argument.
2559 : */
2560 241142 : if (expr->opno == BooleanEqualOperator ||
2561 240998 : expr->opno == BooleanNotEqualOperator)
2562 : {
4150 tgl 2563 CBC 225 : simple = (Expr *) simplify_boolean_equality(expr->opno,
2564 : args);
4150 bruce 2565 GIC 225 : if (simple) /* successfully simplified it */
2566 150 : return (Node *) simple;
2567 : }
2568 :
4150 bruce 2569 ECB : /*
2570 : * The expression cannot be simplified any further, so build
2571 : * and return a replacement OpExpr node using the
2572 : * possibly-simplified arguments.
2573 : */
4150 bruce 2574 CBC 240992 : newexpr = makeNode(OpExpr);
4150 bruce 2575 GIC 240992 : newexpr->opno = expr->opno;
2576 240992 : newexpr->opfuncid = expr->opfuncid;
2577 240992 : newexpr->opresulttype = expr->opresulttype;
2578 240992 : newexpr->opretset = expr->opretset;
2579 240992 : newexpr->opcollid = expr->opcollid;
2580 240992 : newexpr->inputcollid = expr->inputcollid;
2581 240992 : newexpr->args = args;
4150 bruce 2582 CBC 240992 : newexpr->location = expr->location;
2583 240992 : return (Node *) newexpr;
4150 bruce 2584 ECB : }
4150 tgl 2585 CBC 362 : case T_DistinctExpr:
4150 bruce 2586 ECB : {
4150 bruce 2587 CBC 362 : DistinctExpr *expr = (DistinctExpr *) node;
4150 bruce 2588 ECB : List *args;
2589 : ListCell *arg;
4150 bruce 2590 CBC 362 : bool has_null_input = false;
2591 362 : bool all_null_input = true;
2592 362 : bool has_nonconst_input = false;
2593 : Expr *simple;
4150 bruce 2594 ECB : DistinctExpr *newexpr;
2595 :
2596 : /*
2597 : * Reduce constants in the DistinctExpr's arguments. We know
2598 : * args is either NIL or a List node, so we can call
2599 : * expression_tree_mutator directly rather than recursing to
2600 : * self.
2601 : */
4150 bruce 2602 GIC 362 : args = (List *) expression_tree_mutator((Node *) expr->args,
2603 : eval_const_expressions_mutator,
2604 : (void *) context);
8596 tgl 2605 ECB :
2606 : /*
2607 : * We must do our own check for NULLs because DistinctExpr has
2608 : * different results for NULL input than the underlying
2609 : * operator does.
2610 : */
4150 bruce 2611 CBC 1086 : foreach(arg, args)
2612 : {
4150 bruce 2613 GIC 724 : if (IsA(lfirst(arg), Const))
2614 : {
2615 45 : has_null_input |= ((Const *) lfirst(arg))->constisnull;
2616 45 : all_null_input &= ((Const *) lfirst(arg))->constisnull;
2617 : }
2618 : else
2619 679 : has_nonconst_input = true;
4150 bruce 2620 ECB : }
8397 2621 :
2622 : /* all constants? then can optimize this out */
4150 bruce 2623 GIC 362 : if (!has_nonconst_input)
2624 : {
2625 : /* all nulls? then not distinct */
2626 12 : if (all_null_input)
4150 bruce 2627 UIC 0 : return makeBoolConst(false, false);
8397 bruce 2628 ECB :
4150 2629 : /* one null? then distinct */
4150 bruce 2630 GIC 12 : if (has_null_input)
4150 bruce 2631 LBC 0 : return makeBoolConst(true, false);
2632 :
4150 bruce 2633 ECB : /* otherwise try to evaluate the '=' operator */
2634 : /* (NOT okay to try to inline it, though!) */
2635 :
2636 : /*
2637 : * Need to get OID of underlying function. Okay to
2638 : * scribble on input to this extent.
2639 : */
2118 tgl 2640 GIC 12 : set_opfuncid((OpExpr *) expr); /* rely on struct
2641 : * equivalence */
4150 bruce 2642 ECB :
2643 : /*
2644 : * Code for op/func reduction is pretty bulky, so split it
2645 : * out as a separate function.
2646 : */
4034 tgl 2647 CBC 12 : simple = simplify_function(expr->opfuncid,
4150 bruce 2648 ECB : expr->opresulttype, -1,
2649 : expr->opcollid,
2650 : expr->inputcollid,
2651 : &args,
2652 : false,
4034 tgl 2653 : false,
2654 : false,
2655 : context);
4150 bruce 2656 GIC 12 : if (simple) /* successfully simplified it */
2657 : {
4150 bruce 2658 ECB : /*
2659 : * Since the underlying operator is "=", must negate
2660 : * its result
2661 : */
2238 peter_e 2662 GIC 12 : Const *csimple = castNode(Const, simple);
2663 :
4150 bruce 2664 12 : csimple->constvalue =
2665 12 : BoolGetDatum(!DatumGetBool(csimple->constvalue));
2666 12 : return (Node *) csimple;
2667 : }
2668 : }
2669 :
7422 tgl 2670 ECB : /*
2671 : * The expression cannot be simplified any further, so build
2672 : * and return a replacement DistinctExpr node using the
2673 : * possibly-simplified arguments.
2674 : */
4150 bruce 2675 GIC 350 : newexpr = makeNode(DistinctExpr);
2676 350 : newexpr->opno = expr->opno;
2677 350 : newexpr->opfuncid = expr->opfuncid;
2678 350 : newexpr->opresulttype = expr->opresulttype;
4150 bruce 2679 CBC 350 : newexpr->opretset = expr->opretset;
4150 bruce 2680 GIC 350 : newexpr->opcollid = expr->opcollid;
4150 bruce 2681 CBC 350 : newexpr->inputcollid = expr->inputcollid;
4150 bruce 2682 GIC 350 : newexpr->args = args;
4150 bruce 2683 CBC 350 : newexpr->location = expr->location;
2684 350 : return (Node *) newexpr;
2685 : }
737 peter 2686 GIC 101 : case T_NullIfExpr:
737 peter 2687 ECB : {
2688 : NullIfExpr *expr;
2689 : ListCell *arg;
697 tgl 2690 GIC 101 : bool has_nonconst_input = false;
737 peter 2691 ECB :
2692 : /* Copy the node and const-simplify its arguments */
737 peter 2693 GIC 101 : expr = (NullIfExpr *) ece_generic_processing(node);
737 peter 2694 ECB :
737 peter 2695 EUB : /* If either argument is NULL they can't be equal */
737 peter 2696 GIC 300 : foreach(arg, expr->args)
2697 : {
737 peter 2698 CBC 202 : if (!IsA(lfirst(arg), Const))
737 peter 2699 GBC 85 : has_nonconst_input = true;
737 peter 2700 GIC 117 : else if (((Const *) lfirst(arg))->constisnull)
2701 3 : return (Node *) linitial(expr->args);
2702 : }
2703 :
2704 : /*
2705 : * Need to get OID of underlying function before checking if
2706 : * the function is OK to evaluate.
2707 : */
737 peter 2708 CBC 98 : set_opfuncid((OpExpr *) expr);
2709 :
737 peter 2710 GIC 117 : if (!has_nonconst_input &&
2711 19 : ece_function_is_safe(expr->opfuncid, context))
2712 19 : return ece_evaluate_expr(expr);
2713 :
2714 79 : return (Node *) expr;
737 peter 2715 ECB : }
1922 tgl 2716 GIC 14774 : case T_ScalarArrayOpExpr:
2717 : {
2718 : ScalarArrayOpExpr *saop;
2719 :
2720 : /* Copy the node and const-simplify its arguments */
2721 14774 : saop = (ScalarArrayOpExpr *) ece_generic_processing(node);
2722 :
2723 : /* Make sure we know underlying function */
1922 tgl 2724 CBC 14774 : set_sa_opfuncid(saop);
2725 :
2726 : /*
2727 : * If all arguments are Consts, and it's a safe function, we
2728 : * can fold to a constant
2729 : */
2730 14870 : if (ece_all_arguments_const(saop) &&
1922 tgl 2731 GIC 96 : ece_function_is_safe(saop->opfuncid, context))
1922 tgl 2732 CBC 96 : return ece_evaluate_expr(saop);
2733 14678 : return (Node *) saop;
1922 tgl 2734 ECB : }
4150 tgl 2735 GIC 59193 : case T_BoolExpr:
2736 : {
bruce 2737 59193 : BoolExpr *expr = (BoolExpr *) node;
2738 :
2739 59193 : switch (expr->boolop)
2740 : {
2741 4837 : case OR_EXPR:
2742 : {
4150 bruce 2743 ECB : List *newargs;
4150 bruce 2744 CBC 4837 : bool haveNull = false;
2745 4837 : bool forceTrue = false;
4150 bruce 2746 ECB :
4150 tgl 2747 CBC 4837 : newargs = simplify_or_arguments(expr->args,
4150 tgl 2748 ECB : context,
2749 : &haveNull,
2750 : &forceTrue);
4150 bruce 2751 CBC 4837 : if (forceTrue)
2752 68 : return makeBoolConst(true, false);
4150 bruce 2753 GIC 4769 : if (haveNull)
4150 tgl 2754 CBC 15 : newargs = lappend(newargs,
4150 tgl 2755 GIC 15 : makeBoolConst(false, true));
2756 : /* If all the inputs are FALSE, result is FALSE */
bruce 2757 4769 : if (newargs == NIL)
4150 bruce 2758 CBC 3 : return makeBoolConst(false, false);
2759 :
2760 : /*
4150 bruce 2761 ECB : * If only one nonconst-or-NULL input, it's the
2762 : * result
2763 : */
4150 bruce 2764 CBC 4766 : if (list_length(newargs) == 1)
4150 bruce 2765 GIC 48 : return (Node *) linitial(newargs);
4150 bruce 2766 ECB : /* Else we still need an OR node */
4150 bruce 2767 CBC 4718 : return (Node *) make_orclause(newargs);
4150 bruce 2768 ECB : }
4150 bruce 2769 CBC 49463 : case AND_EXPR:
2770 : {
2771 : List *newargs;
4150 bruce 2772 GIC 49463 : bool haveNull = false;
2773 49463 : bool forceFalse = false;
2774 :
tgl 2775 49463 : newargs = simplify_and_arguments(expr->args,
4150 tgl 2776 ECB : context,
2777 : &haveNull,
2778 : &forceFalse);
4150 bruce 2779 CBC 49463 : if (forceFalse)
2780 703 : return makeBoolConst(false, false);
4150 bruce 2781 GIC 48760 : if (haveNull)
4150 tgl 2782 CBC 3 : newargs = lappend(newargs,
4150 tgl 2783 GIC 3 : makeBoolConst(false, true));
4150 bruce 2784 ECB : /* If all the inputs are TRUE, result is TRUE */
4150 bruce 2785 GIC 48760 : if (newargs == NIL)
2786 167 : return makeBoolConst(true, false);
2787 :
2788 : /*
4150 bruce 2789 ECB : * If only one nonconst-or-NULL input, it's the
2790 : * result
2791 : */
4150 bruce 2792 CBC 48593 : if (list_length(newargs) == 1)
4150 bruce 2793 GIC 13 : return (Node *) linitial(newargs);
2794 : /* Else we still need an AND node */
2795 48580 : return (Node *) make_andclause(newargs);
2796 : }
2797 4893 : case NOT_EXPR:
4150 bruce 2798 ECB : {
2799 : Node *arg;
2800 :
4150 bruce 2801 CBC 4893 : Assert(list_length(expr->args) == 1);
4150 bruce 2802 GIC 4893 : arg = eval_const_expressions_mutator(linitial(expr->args),
4150 bruce 2803 ECB : context);
2804 :
2805 : /*
2806 : * Use negate_clause() to see if we can simplify
2807 : * away the NOT.
2808 : */
4150 bruce 2809 CBC 4893 : return negate_clause(arg);
2810 : }
4150 bruce 2811 UIC 0 : default:
4150 bruce 2812 LBC 0 : elog(ERROR, "unrecognized boolop: %d",
4150 bruce 2813 ECB : (int) expr->boolop);
2814 : break;
7027 tgl 2815 : }
2816 : break;
2817 : }
2818 :
11 alvherre 2819 GNC 69 : case T_JsonValueExpr:
2820 : {
2821 69 : JsonValueExpr *jve = (JsonValueExpr *) node;
2822 : Node *raw;
2823 :
2824 69 : raw = eval_const_expressions_mutator((Node *) jve->raw_expr,
2825 : context);
2826 69 : if (raw && IsA(raw, Const))
2827 : {
2828 : Node *formatted;
2829 21 : Node *save_case_val = context->case_val;
2830 :
2831 21 : context->case_val = raw;
2832 :
2833 21 : formatted = eval_const_expressions_mutator((Node *) jve->formatted_expr,
2834 : context);
2835 :
2836 21 : context->case_val = save_case_val;
2837 :
2838 21 : if (formatted && IsA(formatted, Const))
2839 18 : return formatted;
2840 : }
2841 51 : break;
2842 : }
2843 :
4150 tgl 2844 GIC 255 : case T_SubPlan:
4150 tgl 2845 ECB : case T_AlternativeSubPlan:
bruce 2846 :
2847 : /*
2848 : * Return a SubPlan unchanged --- too late to do anything with it.
2849 : *
2850 : * XXX should we ereport() here instead? Probably this routine
2851 : * should never be invoked after SubPlan creation.
2852 : */
4150 bruce 2853 GIC 255 : return node;
tgl 2854 55942 : case T_RelabelType:
2855 : {
bruce 2856 55942 : RelabelType *relabel = (RelabelType *) node;
2857 : Node *arg;
8449 tgl 2858 ECB :
1138 2859 : /* Simplify the input ... */
4150 bruce 2860 GIC 55942 : arg = eval_const_expressions_mutator((Node *) relabel->arg,
4150 bruce 2861 ECB : context);
2862 : /* ... and attach a new RelabelType node, if needed */
963 tgl 2863 CBC 55942 : return applyRelabelType(arg,
2864 : relabel->resulttype,
2865 : relabel->resulttypmod,
963 tgl 2866 ECB : relabel->resultcollid,
2867 : relabel->relabelformat,
2868 : relabel->location,
2869 : true);
2870 : }
4150 tgl 2871 GIC 9330 : case T_CoerceViaIO:
2872 : {
4150 bruce 2873 CBC 9330 : CoerceViaIO *expr = (CoerceViaIO *) node;
4150 bruce 2874 ECB : List *args;
2875 : Oid outfunc;
2876 : bool outtypisvarlena;
2877 : Oid infunc;
2878 : Oid intypioparam;
2879 : Expr *simple;
2880 : CoerceViaIO *newexpr;
2881 :
2882 : /* Make a List so we can use simplify_function */
4034 tgl 2883 GIC 9330 : args = list_make1(expr->arg);
2884 :
2885 : /*
4150 bruce 2886 ECB : * CoerceViaIO represents calling the source type's output
2887 : * function then the result type's input function. So, try to
2888 : * simplify it as though it were a stack of two such function
2889 : * calls. First we need to know what the functions are.
2890 : *
2891 : * Note that the coercion functions are assumed not to care
2892 : * about input collation, so we just pass InvalidOid for that.
2893 : */
4034 tgl 2894 GIC 9330 : getTypeOutputInfo(exprType((Node *) expr->arg),
4150 tgl 2895 ECB : &outfunc, &outtypisvarlena);
4150 tgl 2896 CBC 9330 : getTypeInputInfo(expr->resulttype,
2897 : &infunc, &intypioparam);
2898 :
4034 tgl 2899 GIC 9330 : simple = simplify_function(outfunc,
2900 : CSTRINGOID, -1,
2901 : InvalidOid,
2902 : InvalidOid,
4150 bruce 2903 ECB : &args,
2904 : false,
4034 tgl 2905 EUB : true,
2906 : true,
2907 : context);
4150 bruce 2908 GIC 9330 : if (simple) /* successfully simplified output fn */
2909 : {
2910 : /*
2911 : * Input functions may want 1 to 3 arguments. We always
2912 : * supply all three, trusting that nothing downstream will
4150 bruce 2913 ECB : * complain.
2914 : */
4150 bruce 2915 CBC 653 : args = list_make3(simple,
2916 : makeConst(OIDOID,
2917 : -1,
4150 tgl 2918 ECB : InvalidOid,
2919 : sizeof(Oid),
2118 2920 : ObjectIdGetDatum(intypioparam),
2921 : false,
2922 : true),
4150 2923 : makeConst(INT4OID,
2924 : -1,
2925 : InvalidOid,
2926 : sizeof(int32),
2927 : Int32GetDatum(-1),
2928 : false,
2929 : true));
5442 2930 :
4034 tgl 2931 GIC 653 : simple = simplify_function(infunc,
4150 bruce 2932 ECB : expr->resulttype, -1,
2933 : expr->resultcollid,
2934 : InvalidOid,
2935 : &args,
2936 : false,
2937 : false,
4034 tgl 2938 : true,
2939 : context);
4150 bruce 2940 GIC 630 : if (simple) /* successfully simplified input fn */
2941 607 : return (Node *) simple;
2942 : }
2943 :
2944 : /*
2945 : * The expression cannot be simplified any further, so build
2946 : * and return a replacement CoerceViaIO node using the
4150 bruce 2947 ECB : * possibly-simplified argument.
2948 : */
4150 bruce 2949 GIC 8700 : newexpr = makeNode(CoerceViaIO);
4034 tgl 2950 CBC 8700 : newexpr->arg = (Expr *) linitial(args);
4150 bruce 2951 GIC 8700 : newexpr->resulttype = expr->resulttype;
2952 8700 : newexpr->resultcollid = expr->resultcollid;
2953 8700 : newexpr->coerceformat = expr->coerceformat;
4150 bruce 2954 CBC 8700 : newexpr->location = expr->location;
4150 bruce 2955 GIC 8700 : return (Node *) newexpr;
2956 : }
4150 tgl 2957 CBC 4097 : case T_ArrayCoerceExpr:
2958 : {
1622 tgl 2959 GIC 4097 : ArrayCoerceExpr *ac = makeNode(ArrayCoerceExpr);
2960 : Node *save_case_val;
2961 :
2962 : /*
2963 : * Copy the node and const-simplify its arguments. We can't
2964 : * use ece_generic_processing() here because we need to mess
1622 tgl 2965 ECB : * with case_val only while processing the elemexpr.
2966 : */
1622 tgl 2967 CBC 4097 : memcpy(ac, node, sizeof(ArrayCoerceExpr));
1622 tgl 2968 GIC 4097 : ac->arg = (Expr *)
2969 4097 : eval_const_expressions_mutator((Node *) ac->arg,
2970 : context);
2971 :
2972 : /*
2973 : * Set up for the CaseTestExpr node contained in the elemexpr.
2974 : * We must prevent it from absorbing any outer CASE value.
2975 : */
2976 4097 : save_case_val = context->case_val;
1622 tgl 2977 CBC 4097 : context->case_val = NULL;
2978 :
1622 tgl 2979 GIC 4097 : ac->elemexpr = (Expr *)
2980 4097 : eval_const_expressions_mutator((Node *) ac->elemexpr,
2981 : context);
2982 :
2983 4097 : context->case_val = save_case_val;
2984 :
2985 : /*
2986 : * If constant argument and the per-element expression is
2987 : * immutable, we can simplify the whole thing to a constant.
2017 tgl 2988 ECB : * Exception: although contain_mutable_functions considers
2989 : * CoerceToDomain immutable for historical reasons, let's not
2990 : * do so here; this ensures coercion to an array-over-domain
2991 : * does not apply the domain's constraints until runtime.
2992 : */
1922 tgl 2993 CBC 4097 : if (ac->arg && IsA(ac->arg, Const) &&
1922 tgl 2994 GIC 733 : ac->elemexpr && !IsA(ac->elemexpr, CoerceToDomain) &&
2995 721 : !contain_mutable_functions((Node *) ac->elemexpr))
2996 721 : return ece_evaluate_expr(ac);
2997 :
2998 3376 : return (Node *) ac;
2999 : }
4150 bruce 3000 3575 : case T_CollateExpr:
3001 : {
4150 bruce 3002 ECB : /*
3003 : * We replace CollateExpr with RelabelType, so as to improve
3004 : * uniformity of expression representation and thus simplify
3005 : * comparison of expressions. Hence this looks very nearly
3006 : * the same as the RelabelType case, and we can apply the same
3007 : * optimizations to avoid unnecessary RelabelTypes.
3008 : */
4150 bruce 3009 CBC 3575 : CollateExpr *collate = (CollateExpr *) node;
3010 : Node *arg;
3011 :
3012 : /* Simplify the input ... */
4150 bruce 3013 GIC 3575 : arg = eval_const_expressions_mutator((Node *) collate->arg,
3014 : context);
3015 : /* ... and attach a new RelabelType node, if needed */
963 tgl 3016 3575 : return applyRelabelType(arg,
3017 : exprType(arg),
3018 : exprTypmod(arg),
3019 : collate->collOid,
3020 : COERCE_IMPLICIT_CAST,
3021 : collate->location,
3022 : true);
3023 : }
4150 bruce 3024 12151 : case T_CaseExpr:
4150 bruce 3025 ECB : {
3026 : /*----------
3027 : * CASE expressions can be simplified if there are constant
3028 : * condition clauses:
3029 : * FALSE (or NULL): drop the alternative
3030 : * TRUE: drop all remaining alternatives
3031 : * If the first non-FALSE alternative is a constant TRUE,
3032 : * we can simplify the entire CASE to that alternative's
3033 : * expression. If there are no non-FALSE alternatives,
tgl 3034 : * we simplify the entire CASE to the default result (ELSE).
bruce 3035 : *
3036 : * If we have a simple-form CASE with constant test
3037 : * expression, we substitute the constant value for contained
3038 : * CaseTestExpr placeholder nodes, so that we have the
3039 : * opportunity to reduce constant test conditions. For
3040 : * example this allows
3041 : * CASE 0 WHEN 0 THEN 1 ELSE 1/0 END
3042 : * to reduce to 1 rather than drawing a divide-by-0 error.
tgl 3043 : * Note that when the test expression is constant, we don't
3044 : * have to include it in the resulting CASE; for example
bruce 3045 : * CASE 0 WHEN x THEN y ELSE z END
3046 : * is transformed by the parser to
3047 : * CASE 0 WHEN CaseTestExpr = x THEN y ELSE z END
3048 : * which we can simplify to
3049 : * CASE WHEN 0 = x THEN y ELSE z END
3050 : * It is not necessary for the executor to evaluate the "arg"
3051 : * expression when executing the CASE, since any contained
3052 : * CaseTestExprs that might have referred to it will have been
3053 : * replaced by the constant.
3054 : *----------
3055 : */
4150 bruce 3056 GIC 12151 : CaseExpr *caseexpr = (CaseExpr *) node;
3057 : CaseExpr *newcase;
3058 : Node *save_case_val;
3059 : Node *newarg;
3060 : List *newargs;
4150 bruce 3061 ECB : bool const_true_cond;
4150 bruce 3062 CBC 12151 : Node *defresult = NULL;
4150 bruce 3063 ECB : ListCell *arg;
3064 :
3065 : /* Simplify the test expression, if any */
4150 bruce 3066 GIC 12151 : newarg = eval_const_expressions_mutator((Node *) caseexpr->arg,
3067 : context);
3068 :
3069 : /* Set up for contained CaseTestExpr nodes */
4150 bruce 3070 CBC 12151 : save_case_val = context->case_val;
3071 12151 : if (newarg && IsA(newarg, Const))
3072 : {
3073 9 : context->case_val = newarg;
2118 tgl 3074 9 : newarg = NULL; /* not needed anymore, see above */
3075 : }
3076 : else
4150 bruce 3077 12142 : context->case_val = NULL;
3078 :
3079 : /* Simplify the WHEN clauses */
4150 bruce 3080 GIC 12151 : newargs = NIL;
3081 12151 : const_true_cond = false;
3082 30824 : foreach(arg, caseexpr->args)
3083 : {
2190 tgl 3084 18830 : CaseWhen *oldcasewhen = lfirst_node(CaseWhen, arg);
3085 : Node *casecond;
3086 : Node *caseresult;
6640 tgl 3087 ECB :
4150 bruce 3088 : /* Simplify this alternative's test condition */
4150 tgl 3089 CBC 18830 : casecond = eval_const_expressions_mutator((Node *) oldcasewhen->expr,
4150 tgl 3090 ECB : context);
3091 :
bruce 3092 : /*
3093 : * If the test condition is constant FALSE (or NULL), then
3094 : * drop this WHEN clause completely, without processing
3095 : * the result.
3096 : */
4150 bruce 3097 GIC 18830 : if (casecond && IsA(casecond, Const))
3098 : {
3099 457 : Const *const_input = (Const *) casecond;
3100 :
3101 457 : if (const_input->constisnull ||
3102 457 : !DatumGetBool(const_input->constvalue))
4150 tgl 3103 CBC 303 : continue; /* drop alternative with FALSE cond */
3104 : /* Else it's constant TRUE */
4150 bruce 3105 GIC 154 : const_true_cond = true;
3106 : }
4150 bruce 3107 ECB :
3108 : /* Simplify this alternative's result value */
4150 tgl 3109 GIC 18527 : caseresult = eval_const_expressions_mutator((Node *) oldcasewhen->result,
4150 tgl 3110 ECB : context);
3111 :
3112 : /* If non-constant test condition, emit a new WHEN node */
4150 bruce 3113 GIC 18524 : if (!const_true_cond)
3114 18370 : {
3115 18370 : CaseWhen *newcasewhen = makeNode(CaseWhen);
3116 :
3117 18370 : newcasewhen->expr = (Expr *) casecond;
4150 bruce 3118 CBC 18370 : newcasewhen->result = (Expr *) caseresult;
4150 bruce 3119 GIC 18370 : newcasewhen->location = oldcasewhen->location;
3120 18370 : newargs = lappend(newargs, newcasewhen);
3121 18370 : continue;
3122 : }
3123 :
3124 : /*
3125 : * Found a TRUE condition, so none of the remaining
3126 : * alternatives can be reached. We treat the result as
3127 : * the default result.
3128 : */
3129 154 : defresult = caseresult;
3130 154 : break;
3131 : }
3132 :
3133 : /* Simplify the default result, unless we replaced it above */
3134 12148 : if (!const_true_cond)
tgl 3135 11994 : defresult = eval_const_expressions_mutator((Node *) caseexpr->defresult,
3136 : context);
3137 :
bruce 3138 12148 : context->case_val = save_case_val;
3139 :
3140 : /*
3141 : * If no non-FALSE alternatives, CASE reduces to the default
3142 : * result
3143 : */
3144 12148 : if (newargs == NIL)
3145 256 : return defresult;
3146 : /* Otherwise we need a new CASE node */
3147 11892 : newcase = makeNode(CaseExpr);
3148 11892 : newcase->casetype = caseexpr->casetype;
3149 11892 : newcase->casecollid = caseexpr->casecollid;
4150 bruce 3150 CBC 11892 : newcase->arg = (Expr *) newarg;
4150 bruce 3151 GIC 11892 : newcase->args = newargs;
3152 11892 : newcase->defresult = (Expr *) defresult;
3153 11892 : newcase->location = caseexpr->location;
3154 11892 : return (Node *) newcase;
3155 : }
4150 tgl 3156 CBC 12296 : case T_CaseTestExpr:
3157 : {
3158 : /*
3159 : * If we know a constant test value for the current CASE
4150 bruce 3160 ECB : * construct, substitute it for the placeholder. Else just
3161 : * return the placeholder as-is.
3162 : */
4150 bruce 3163 GIC 12296 : if (context->case_val)
4150 bruce 3164 CBC 33 : return copyObject(context->case_val);
4150 bruce 3165 ECB : else
4150 bruce 3166 GIC 12263 : return copyObject(node);
4150 bruce 3167 ECB : }
1528 alvherre 3168 CBC 21872 : case T_SubscriptingRef:
3169 : case T_ArrayExpr:
3170 : case T_RowExpr:
1561 tgl 3171 ECB : case T_MinMaxExpr:
3172 : {
3173 : /*
1922 3174 : * Generic handling for node types whose own processing is
3175 : * known to be immutable, and for which we need no smarts
3176 : * beyond "simplify if all inputs are constants".
3177 : *
851 3178 : * Treating SubscriptingRef this way assumes that subscripting
3179 : * fetch and assignment are both immutable. This constrains
3180 : * type-specific subscripting implementations; maybe we should
3181 : * relax it someday.
3182 : *
1561 3183 : * Treating MinMaxExpr this way amounts to assuming that the
3184 : * btree comparison function it calls is immutable; see the
3185 : * reasoning in contain_mutable_functions_walker.
3186 : */
3187 :
3188 : /* Copy the node and const-simplify its arguments */
1922 tgl 3189 GIC 21872 : node = ece_generic_processing(node);
3190 : /* If all arguments are Consts, we can fold to a constant */
1922 tgl 3191 CBC 21872 : if (ece_all_arguments_const(node))
1922 tgl 3192 GIC 12795 : return ece_evaluate_expr(node);
1922 tgl 3193 CBC 9077 : return node;
3194 : }
4150 3195 1165 : case T_CoalesceExpr:
4150 bruce 3196 ECB : {
4150 bruce 3197 CBC 1165 : CoalesceExpr *coalesceexpr = (CoalesceExpr *) node;
3198 : CoalesceExpr *newcoalesce;
4150 bruce 3199 ECB : List *newargs;
3200 : ListCell *arg;
3201 :
4150 bruce 3202 GIC 1165 : newargs = NIL;
4150 bruce 3203 CBC 2825 : foreach(arg, coalesceexpr->args)
3204 : {
3205 : Node *e;
3206 :
3207 2318 : e = eval_const_expressions_mutator((Node *) lfirst(arg),
4150 bruce 3208 ECB : context);
7188 3209 :
3210 : /*
4150 3211 : * We can remove null constants from the list. For a
3212 : * non-null constant, if it has not been preceded by any
3213 : * other non-null-constant expressions then it is the
3214 : * result. Otherwise, it's the next argument, but we can
3215 : * drop following arguments since they will never be
3216 : * reached.
3217 : */
4150 bruce 3218 GIC 2318 : if (IsA(e, Const))
3219 : {
3220 677 : if (((Const *) e)->constisnull)
3221 19 : continue; /* drop null constant */
3222 658 : if (newargs == NIL)
4150 bruce 3223 CBC 25 : return e; /* first expr */
3224 633 : newargs = lappend(newargs, e);
4150 bruce 3225 GIC 633 : break;
3226 : }
3227 1641 : newargs = lappend(newargs, e);
4150 bruce 3228 ECB : }
3229 :
3230 : /*
3231 : * If all the arguments were constant null, the result is just
3232 : * null
3233 : */
6886 tgl 3234 GIC 1140 : if (newargs == NIL)
4150 bruce 3235 UIC 0 : return (Node *) makeNullConst(coalesceexpr->coalescetype,
3236 : -1,
3237 : coalesceexpr->coalescecollid);
4150 bruce 3238 ECB :
4150 bruce 3239 CBC 1140 : newcoalesce = makeNode(CoalesceExpr);
4150 bruce 3240 GIC 1140 : newcoalesce->coalescetype = coalesceexpr->coalescetype;
4150 bruce 3241 CBC 1140 : newcoalesce->coalescecollid = coalesceexpr->coalescecollid;
3242 1140 : newcoalesce->args = newargs;
3243 1140 : newcoalesce->location = coalesceexpr->location;
3244 1140 : return (Node *) newcoalesce;
7357 tgl 3245 ECB : }
4150 tgl 3246 GIC 2072 : case T_FieldSelect:
3247 : {
3248 : /*
3249 : * We can optimize field selection from a whole-row Var into a
3250 : * simple Var. (This case won't be generated directly by the
3251 : * parser, because ParseComplexProjection short-circuits it.
3252 : * But it can arise while simplifying functions.) Also, we
3253 : * can optimize field selection from a RowExpr construct, or
3254 : * of course from a constant.
3255 : *
3256 : * However, replacing a whole-row Var in this way has a
3257 : * pitfall: if we've already built the rel targetlist for the
3258 : * source relation, then the whole-row Var is scheduled to be
3259 : * produced by the relation scan, but the simple Var probably
3260 : * isn't, which will lead to a failure in setrefs.c. This is
3261 : * not a problem when handling simple single-level queries, in
3262 : * which expression simplification always happens first. It
3263 : * is a risk for lateral references from subqueries, though.
3264 : * To avoid such failures, don't optimize uplevel references.
3265 : *
3093 tgl 3266 ECB : * We must also check that the declared type of the field is
3267 : * still the same as when the FieldSelect was created --- this
3268 : * can change if someone did ALTER COLUMN TYPE on the rowtype.
1922 3269 : * If it isn't, we skip the optimization; the case will
3270 : * probably fail at runtime, but that's not our problem here.
3271 : */
4150 bruce 3272 CBC 2072 : FieldSelect *fselect = (FieldSelect *) node;
3273 : FieldSelect *newfselect;
4150 bruce 3274 ECB : Node *arg;
3275 :
4150 bruce 3276 GIC 2072 : arg = eval_const_expressions_mutator((Node *) fselect->arg,
3277 : context);
3278 2072 : if (arg && IsA(arg, Var) &&
3093 tgl 3279 CBC 246 : ((Var *) arg)->varattno == InvalidAttrNumber &&
3280 42 : ((Var *) arg)->varlevelsup == 0)
3281 : {
4150 bruce 3282 GIC 36 : if (rowtype_field_matches(((Var *) arg)->vartype,
3283 36 : fselect->fieldnum,
4150 bruce 3284 ECB : fselect->resulttype,
3285 : fselect->resulttypmod,
3286 : fselect->resultcollid))
4150 bruce 3287 GIC 36 : return (Node *) makeVar(((Var *) arg)->varno,
3288 36 : fselect->fieldnum,
3289 : fselect->resulttype,
3290 : fselect->resulttypmod,
3291 : fselect->resultcollid,
3292 : ((Var *) arg)->varlevelsup);
3293 : }
3294 2036 : if (arg && IsA(arg, RowExpr))
4150 bruce 3295 ECB : {
4150 bruce 3296 GIC 12 : RowExpr *rowexpr = (RowExpr *) arg;
4150 bruce 3297 ECB :
4150 bruce 3298 CBC 24 : if (fselect->fieldnum > 0 &&
3299 12 : fselect->fieldnum <= list_length(rowexpr->args))
4150 bruce 3300 ECB : {
4150 bruce 3301 CBC 12 : Node *fld = (Node *) list_nth(rowexpr->args,
2118 tgl 3302 12 : fselect->fieldnum - 1);
3303 :
4150 bruce 3304 12 : if (rowtype_field_matches(rowexpr->row_typeid,
4150 bruce 3305 GIC 12 : fselect->fieldnum,
3306 : fselect->resulttype,
3307 : fselect->resulttypmod,
3308 12 : fselect->resultcollid) &&
3309 24 : fselect->resulttype == exprType(fld) &&
3310 24 : fselect->resulttypmod == exprTypmod(fld) &&
4150 bruce 3311 CBC 12 : fselect->resultcollid == exprCollation(fld))
4150 bruce 3312 GBC 12 : return fld;
3313 : }
3314 : }
4150 bruce 3315 GIC 2024 : newfselect = makeNode(FieldSelect);
4150 bruce 3316 CBC 2024 : newfselect->arg = (Expr *) arg;
3317 2024 : newfselect->fieldnum = fselect->fieldnum;
3318 2024 : newfselect->resulttype = fselect->resulttype;
3319 2024 : newfselect->resulttypmod = fselect->resulttypmod;
3320 2024 : newfselect->resultcollid = fselect->resultcollid;
1922 tgl 3321 2024 : if (arg && IsA(arg, Const))
3322 : {
3323 212 : Const *con = (Const *) arg;
3324 :
1922 tgl 3325 GIC 212 : if (rowtype_field_matches(con->consttype,
3326 212 : newfselect->fieldnum,
3327 : newfselect->resulttype,
3328 : newfselect->resulttypmod,
3329 : newfselect->resultcollid))
3330 212 : return ece_evaluate_expr(newfselect);
3331 : }
4150 bruce 3332 1812 : return (Node *) newfselect;
3333 : }
tgl 3334 16797 : case T_NullTest:
3335 : {
bruce 3336 16797 : NullTest *ntest = (NullTest *) node;
3337 : NullTest *newntest;
3338 : Node *arg;
3339 :
3340 16797 : arg = eval_const_expressions_mutator((Node *) ntest->arg,
3341 : context);
2448 tgl 3342 16796 : if (ntest->argisrow && arg && IsA(arg, RowExpr))
3343 : {
3344 : /*
3345 : * We break ROW(...) IS [NOT] NULL into separate tests on
3346 : * its component fields. This form is usually more
3347 : * efficient to evaluate, as well as being more amenable
3348 : * to optimization.
4150 bruce 3349 ECB : */
4150 bruce 3350 GIC 15 : RowExpr *rarg = (RowExpr *) arg;
3351 15 : List *newargs = NIL;
3352 : ListCell *l;
4150 bruce 3353 ECB :
4150 bruce 3354 GIC 60 : foreach(l, rarg->args)
4150 bruce 3355 ECB : {
4150 bruce 3356 CBC 45 : Node *relem = (Node *) lfirst(l);
4150 bruce 3357 ECB :
3358 : /*
3359 : * A constant field refutes the whole NullTest if it's
3360 : * of the wrong nullness; else we can discard it.
3361 : */
4150 bruce 3362 GIC 45 : if (relem && IsA(relem, Const))
4150 bruce 3363 UIC 0 : {
4150 bruce 3364 LBC 0 : Const *carg = (Const *) relem;
4150 bruce 3365 ECB :
4150 bruce 3366 UIC 0 : if (carg->constisnull ?
3367 0 : (ntest->nulltesttype == IS_NOT_NULL) :
3368 0 : (ntest->nulltesttype == IS_NULL))
3369 0 : return makeBoolConst(false, false);
3370 0 : continue;
4150 bruce 3371 ECB : }
3372 :
2448 tgl 3373 : /*
3374 : * Else, make a scalar (argisrow == false) NullTest
3375 : * for this field. Scalar semantics are required
3376 : * because IS [NOT] NULL doesn't recurse; see comments
3377 : * in ExecEvalRowNullInt().
3378 : */
4150 bruce 3379 CBC 45 : newntest = makeNode(NullTest);
4150 bruce 3380 GIC 45 : newntest->arg = (Expr *) relem;
4150 bruce 3381 CBC 45 : newntest->nulltesttype = ntest->nulltesttype;
2448 tgl 3382 45 : newntest->argisrow = false;
2968 tgl 3383 GIC 45 : newntest->location = ntest->location;
4150 bruce 3384 45 : newargs = lappend(newargs, newntest);
4150 bruce 3385 ECB : }
3386 : /* If all the inputs were constants, result is TRUE */
4150 bruce 3387 CBC 15 : if (newargs == NIL)
4150 bruce 3388 LBC 0 : return makeBoolConst(true, false);
4150 bruce 3389 ECB : /* If only one nonconst input, it's the result */
4150 bruce 3390 GIC 15 : if (list_length(newargs) == 1)
4150 bruce 3391 UIC 0 : return (Node *) linitial(newargs);
4150 bruce 3392 ECB : /* Else we need an AND node */
4150 bruce 3393 CBC 15 : return (Node *) make_andclause(newargs);
4150 bruce 3394 ECB : }
4150 bruce 3395 CBC 16781 : if (!ntest->argisrow && arg && IsA(arg, Const))
4150 bruce 3396 ECB : {
4150 bruce 3397 CBC 175 : Const *carg = (Const *) arg;
4150 bruce 3398 ECB : bool result;
3399 :
4150 bruce 3400 CBC 175 : switch (ntest->nulltesttype)
3401 : {
3402 153 : case IS_NULL:
3403 153 : result = carg->constisnull;
4150 bruce 3404 GIC 153 : break;
3405 22 : case IS_NOT_NULL:
3406 22 : result = !carg->constisnull;
4150 bruce 3407 CBC 22 : break;
4150 bruce 3408 UIC 0 : default:
4150 bruce 3409 LBC 0 : elog(ERROR, "unrecognized nulltesttype: %d",
3410 : (int) ntest->nulltesttype);
2118 tgl 3411 ECB : result = false; /* keep compiler quiet */
3412 : break;
4150 bruce 3413 : }
3414 :
4150 bruce 3415 GIC 175 : return makeBoolConst(result, false);
3416 : }
4150 bruce 3417 ECB :
6037 tgl 3418 GIC 16606 : newntest = makeNode(NullTest);
4150 bruce 3419 CBC 16606 : newntest->arg = (Expr *) arg;
6037 tgl 3420 GIC 16606 : newntest->nulltesttype = ntest->nulltesttype;
4150 bruce 3421 16606 : newntest->argisrow = ntest->argisrow;
2968 tgl 3422 16606 : newntest->location = ntest->location;
4150 bruce 3423 16606 : return (Node *) newntest;
3424 : }
tgl 3425 418 : case T_BooleanTest:
3426 : {
1922 tgl 3427 ECB : /*
3428 : * This case could be folded into the generic handling used
3429 : * for ArrayExpr etc. But because the simplification logic is
3430 : * so trivial, applying evaluate_expr() to perform it would be
851 3431 : * a heavy overhead. BooleanTest is probably common enough to
3432 : * justify keeping this bespoke implementation.
1922 3433 : */
4150 bruce 3434 GIC 418 : BooleanTest *btest = (BooleanTest *) node;
3435 : BooleanTest *newbtest;
3436 : Node *arg;
3437 :
3438 418 : arg = eval_const_expressions_mutator((Node *) btest->arg,
4150 bruce 3439 ECB : context);
4150 bruce 3440 GBC 418 : if (arg && IsA(arg, Const))
4150 bruce 3441 EUB : {
4150 bruce 3442 GIC 111 : Const *carg = (Const *) arg;
4150 bruce 3443 EUB : bool result;
3444 :
4150 bruce 3445 GBC 111 : switch (btest->booltesttype)
4150 bruce 3446 EUB : {
4150 bruce 3447 UBC 0 : case IS_TRUE:
4150 bruce 3448 UIC 0 : result = (!carg->constisnull &&
3449 0 : DatumGetBool(carg->constvalue));
3450 0 : break;
4150 bruce 3451 GIC 111 : case IS_NOT_TRUE:
3452 222 : result = (carg->constisnull ||
3453 111 : !DatumGetBool(carg->constvalue));
3454 111 : break;
4150 bruce 3455 UIC 0 : case IS_FALSE:
4150 bruce 3456 LBC 0 : result = (!carg->constisnull &&
3457 0 : !DatumGetBool(carg->constvalue));
3458 0 : break;
3459 0 : case IS_NOT_FALSE:
3460 0 : result = (carg->constisnull ||
3461 0 : DatumGetBool(carg->constvalue));
4150 bruce 3462 UIC 0 : break;
3463 0 : case IS_UNKNOWN:
4150 bruce 3464 LBC 0 : result = carg->constisnull;
4150 bruce 3465 UBC 0 : break;
4150 bruce 3466 UIC 0 : case IS_NOT_UNKNOWN:
4150 bruce 3467 LBC 0 : result = !carg->constisnull;
4150 bruce 3468 UBC 0 : break;
4150 bruce 3469 UIC 0 : default:
4150 bruce 3470 LBC 0 : elog(ERROR, "unrecognized booltesttype: %d",
3471 : (int) btest->booltesttype);
2118 tgl 3472 ECB : result = false; /* keep compiler quiet */
3473 : break;
4150 bruce 3474 : }
3475 :
4150 bruce 3476 GIC 111 : return makeBoolConst(result, false);
4150 bruce 3477 ECB : }
3478 :
4150 bruce 3479 CBC 307 : newbtest = makeNode(BooleanTest);
3480 307 : newbtest->arg = (Expr *) arg;
3481 307 : newbtest->booltesttype = btest->booltesttype;
2968 tgl 3482 307 : newbtest->location = btest->location;
4150 bruce 3483 307 : return (Node *) newbtest;
4150 bruce 3484 ECB : }
1578 tgl 3485 GBC 67281 : case T_CoerceToDomain:
1578 tgl 3486 EUB : {
3487 : /*
3488 : * If the domain currently has no constraints, we replace the
3489 : * CoerceToDomain node with a simple RelabelType, which is
3490 : * both far faster to execute and more amenable to later
3491 : * optimization. We must then mark the plan as needing to be
1578 tgl 3492 ECB : * rebuilt if the domain's constraints change.
3493 : *
3494 : * Also, in estimation mode, always replace CoerceToDomain
3495 : * nodes, effectively assuming that the coercion will succeed.
3496 : */
1578 tgl 3497 CBC 67281 : CoerceToDomain *cdomain = (CoerceToDomain *) node;
1578 tgl 3498 ECB : CoerceToDomain *newcdomain;
3499 : Node *arg;
3500 :
1578 tgl 3501 GIC 67281 : arg = eval_const_expressions_mutator((Node *) cdomain->arg,
1578 tgl 3502 ECB : context);
1578 tgl 3503 GIC 67269 : if (context->estimate ||
3504 67257 : !DomainHasConstraints(cdomain->resulttype))
3505 : {
3506 : /* Record dependency, if this isn't estimation mode */
3507 44862 : if (context->root && !context->estimate)
3508 44826 : record_plan_type_dependency(context->root,
3509 : cdomain->resulttype);
3510 :
1578 tgl 3511 ECB : /* Generate RelabelType to substitute for CoerceToDomain */
963 tgl 3512 GIC 44862 : return applyRelabelType(arg,
3513 : cdomain->resulttype,
3514 : cdomain->resulttypmod,
963 tgl 3515 ECB : cdomain->resultcollid,
3516 : cdomain->coercionformat,
3517 : cdomain->location,
3518 : true);
1578 3519 : }
3520 :
1578 tgl 3521 GIC 22407 : newcdomain = makeNode(CoerceToDomain);
1578 tgl 3522 CBC 22407 : newcdomain->arg = (Expr *) arg;
1578 tgl 3523 GIC 22407 : newcdomain->resulttype = cdomain->resulttype;
1578 tgl 3524 GBC 22407 : newcdomain->resulttypmod = cdomain->resulttypmod;
3525 22407 : newcdomain->resultcollid = cdomain->resultcollid;
3526 22407 : newcdomain->coercionformat = cdomain->coercionformat;
3527 22407 : newcdomain->location = cdomain->location;
1578 tgl 3528 CBC 22407 : return (Node *) newcdomain;
1578 tgl 3529 ECB : }
4150 tgl 3530 CBC 1060 : case T_PlaceHolderVar:
4150 bruce 3531 ECB :
4150 bruce 3532 EUB : /*
3533 : * In estimation mode, just strip the PlaceHolderVar node
3534 : * altogether; this amounts to estimating that the contained value
3535 : * won't be forced to null by an outer join. In regular mode we
3536 : * just use the default behavior (ie, simplify the expression but
3537 : * leave the PlaceHolderVar node intact).
3538 : */
4150 tgl 3539 GBC 1060 : if (context->estimate)
4150 tgl 3540 EUB : {
4150 tgl 3541 GBC 360 : PlaceHolderVar *phv = (PlaceHolderVar *) node;
5283 tgl 3542 EUB :
4150 tgl 3543 GBC 360 : return eval_const_expressions_mutator((Node *) phv->phexpr,
4150 tgl 3544 EUB : context);
3545 : }
4150 tgl 3546 GBC 700 : break;
1615 rhodiumtoad 3547 39 : case T_ConvertRowtypeExpr:
3548 : {
1615 rhodiumtoad 3549 GIC 39 : ConvertRowtypeExpr *cre = castNode(ConvertRowtypeExpr, node);
3550 : Node *arg;
3551 : ConvertRowtypeExpr *newcre;
3552 :
1615 rhodiumtoad 3553 CBC 39 : arg = eval_const_expressions_mutator((Node *) cre->arg,
3554 : context);
3555 :
3556 39 : newcre = makeNode(ConvertRowtypeExpr);
3557 39 : newcre->resulttype = cre->resulttype;
3558 39 : newcre->convertformat = cre->convertformat;
3559 39 : newcre->location = cre->location;
1615 rhodiumtoad 3560 ECB :
3561 : /*
3562 : * In case of a nested ConvertRowtypeExpr, we can convert the
3563 : * leaf row directly to the topmost row format without any
3564 : * intermediate conversions. (This works because
3565 : * ConvertRowtypeExpr is used only for child->parent
3566 : * conversion in inheritance trees, which works by exact match
3567 : * of column name, and a column absent in an intermediate
3568 : * result can't be present in the final result.)
3569 : *
3570 : * No need to check more than one level deep, because the
3571 : * above recursion will have flattened anything else.
3572 : */
1615 rhodiumtoad 3573 GIC 39 : if (arg != NULL && IsA(arg, ConvertRowtypeExpr))
1615 rhodiumtoad 3574 ECB : {
1615 rhodiumtoad 3575 GIC 6 : ConvertRowtypeExpr *argcre = (ConvertRowtypeExpr *) arg;
3576 :
3577 6 : arg = (Node *) argcre->arg;
1615 rhodiumtoad 3578 ECB :
3579 : /*
3580 : * Make sure an outer implicit conversion can't hide an
3581 : * inner explicit one.
3582 : */
1615 rhodiumtoad 3583 GIC 6 : if (newcre->convertformat == COERCE_IMPLICIT_CAST)
1615 rhodiumtoad 3584 LBC 0 : newcre->convertformat = argcre->convertformat;
1615 rhodiumtoad 3585 ECB : }
3586 :
1615 rhodiumtoad 3587 GIC 39 : newcre->arg = (Expr *) arg;
3588 :
1615 rhodiumtoad 3589 CBC 39 : if (arg != NULL && IsA(arg, Const))
1615 rhodiumtoad 3590 GIC 9 : return ece_evaluate_expr((Node *) newcre);
3591 30 : return (Node *) newcre;
3592 : }
4150 tgl 3593 2626131 : default:
3594 2626131 : break;
3595 : }
3596 :
3597 : /*
1922 tgl 3598 ECB : * For any node type not handled above, copy the node unchanged but
3599 : * const-simplify its subexpressions. This is the correct thing for node
3600 : * types whose behavior might change between planning and execution, such
1578 3601 : * as CurrentOfExpr. It's also a safe default for new node types not
1922 3602 : * known to this routine.
3603 : */
1922 tgl 3604 CBC 2626882 : return ece_generic_processing(node);
1922 tgl 3605 ECB : }
3606 :
3607 : /*
3608 : * Subroutine for eval_const_expressions: check for non-Const nodes.
3609 : *
3610 : * We can abort recursion immediately on finding a non-Const node. This is
3611 : * critical for performance, else eval_const_expressions_mutator would take
3612 : * O(N^2) time on non-simplifiable trees. However, we do need to descend
3613 : * into List nodes since expression_tree_walker sometimes invokes the walker
3614 : * function directly on List subtrees.
3615 : */
3616 : static bool
1922 tgl 3617 GIC 81680 : contain_non_const_walker(Node *node, void *context)
1922 tgl 3618 ECB : {
1922 tgl 3619 GIC 81680 : if (node == NULL)
1922 tgl 3620 CBC 299 : return false;
1922 tgl 3621 GIC 81381 : if (IsA(node, Const))
3622 42461 : return false;
1922 tgl 3623 CBC 38920 : if (IsA(node, List))
3624 15165 : return expression_tree_walker(node, contain_non_const_walker, context);
3625 : /* Otherwise, abort the tree traversal and return true */
3626 23755 : return true;
3627 : }
3628 :
3629 : /*
1922 tgl 3630 ECB : * Subroutine for eval_const_expressions: check if a function is OK to evaluate
3631 : */
3632 : static bool
1922 tgl 3633 CBC 115 : ece_function_is_safe(Oid funcid, eval_const_expressions_context *context)
1922 tgl 3634 ECB : {
1922 tgl 3635 CBC 115 : char provolatile = func_volatile(funcid);
1922 tgl 3636 ECB :
3637 : /*
3638 : * Ordinarily we are only allowed to simplify immutable functions. But for
3639 : * purposes of estimation, we consider it okay to simplify functions that
3640 : * are merely stable; the risk that the result might change from planning
3641 : * time to execution time is worth taking in preference to not being able
3642 : * to estimate the value at all.
3643 : */
1922 tgl 3644 GIC 115 : if (provolatile == PROVOLATILE_IMMUTABLE)
3645 115 : return true;
1922 tgl 3646 UIC 0 : if (context->estimate && provolatile == PROVOLATILE_STABLE)
3647 0 : return true;
3648 0 : return false;
3649 : }
8596 tgl 3650 ECB :
3651 : /*
6560 3652 : * Subroutine for eval_const_expressions: process arguments of an OR clause
3653 : *
3654 : * This includes flattening of nested ORs as well as recursion to
3655 : * eval_const_expressions to simplify the OR arguments.
3656 : *
3657 : * After simplification, OR arguments are handled as follows:
3658 : * non constant: keep
3659 : * FALSE: drop (does not affect result)
7027 3660 : * TRUE: force result to TRUE
7027 tgl 3661 EUB : * NULL: keep only one
3662 : * We must keep one NULL input because OR expressions evaluate to NULL when no
3663 : * input is TRUE and at least one is NULL. We don't actually include the NULL
6560 tgl 3664 ECB : * here, that's supposed to be done by the caller.
3665 : *
2062 peter_e 3666 : * The output arguments *haveNull and *forceTrue must be initialized false
3667 : * by the caller. They will be set true if a NULL constant or TRUE constant,
7027 tgl 3668 : * respectively, is detected anywhere in the argument list.
3669 : */
3670 : static List *
6560 tgl 3671 CBC 4837 : simplify_or_arguments(List *args,
3672 : eval_const_expressions_context *context,
3673 : bool *haveNull, bool *forceTrue)
3674 : {
7027 tgl 3675 GIC 4837 : List *newargs = NIL;
3676 : List *unprocessed_args;
3677 :
3678 : /*
3679 : * We want to ensure that any OR immediately beneath another OR gets
3680 : * flattened into a single OR-list, so as to simplify later reasoning.
3219 tgl 3681 ECB : *
3682 : * To avoid stack overflow from recursion of eval_const_expressions, we
3683 : * resort to some tenseness here: we keep a list of not-yet-processed
3684 : * inputs, and handle flattening of nested ORs by prepending to the to-do
3685 : * list instead of recursing. Now that the parser generates N-argument
3686 : * ORs from simple lists, this complexity is probably less necessary than
3687 : * it once was, but we might as well keep the logic.
3688 : */
6560 tgl 3689 GIC 4837 : unprocessed_args = list_copy(args);
3690 16522 : while (unprocessed_args)
3691 : {
3692 11753 : Node *arg = (Node *) linitial(unprocessed_args);
3693 :
6560 tgl 3694 CBC 11753 : unprocessed_args = list_delete_first(unprocessed_args);
3695 :
6560 tgl 3696 ECB : /* flatten nested ORs as per above comment */
1531 tgl 3697 CBC 11753 : if (is_orclause(arg))
6560 3698 3 : {
1336 3699 3 : List *subargs = ((BoolExpr *) arg)->args;
3700 3 : List *oldlist = unprocessed_args;
6560 tgl 3701 ECB :
1336 tgl 3702 GIC 3 : unprocessed_args = list_concat_copy(subargs, unprocessed_args);
1336 tgl 3703 ECB : /* perhaps-overly-tense code to avoid leaking old lists */
1336 tgl 3704 GIC 3 : list_free(oldlist);
6560 3705 3 : continue;
3706 : }
3707 :
3708 : /* If it's not an OR, simplify it */
3709 11750 : arg = eval_const_expressions_mutator(arg, context);
6560 tgl 3710 ECB :
3711 : /*
6385 bruce 3712 : * It is unlikely but not impossible for simplification of a non-OR
3713 : * clause to produce an OR. Recheck, but don't be too tense about it
3714 : * since it's not a mainstream case. In particular we don't worry
3715 : * about const-simplifying the input twice, nor about list leakage.
3716 : */
1531 tgl 3717 GIC 11750 : if (is_orclause(arg))
6560 tgl 3718 UIC 0 : {
1336 3719 0 : List *subargs = ((BoolExpr *) arg)->args;
3720 :
1336 tgl 3721 LBC 0 : unprocessed_args = list_concat_copy(subargs, unprocessed_args);
6560 3722 0 : continue;
6560 tgl 3723 EUB : }
7027 3724 :
6560 3725 : /*
3726 : * OK, we have a const-simplified non-OR argument. Process it per
3727 : * comments above.
3728 : */
7027 tgl 3729 GIC 11750 : if (IsA(arg, Const))
3730 69 : {
6797 bruce 3731 137 : Const *const_input = (Const *) arg;
3732 :
7027 tgl 3733 137 : if (const_input->constisnull)
3734 24 : *haveNull = true;
3735 113 : else if (DatumGetBool(const_input->constvalue))
3736 : {
3737 68 : *forceTrue = true;
3738 :
3739 : /*
3740 : * Once we detect a TRUE result we can just exit the loop
3741 : * immediately. However, if we ever add a notion of
3742 : * non-removable functions, we'd need to keep scanning.
3743 : */
3744 68 : return NIL;
3745 : }
3746 : /* otherwise, we can drop the constant-false input */
6560 3747 69 : continue;
7027 tgl 3748 ECB : }
3749 :
3750 : /* else emit the simplified arg into the result list */
6560 tgl 3751 GIC 11613 : newargs = lappend(newargs, arg);
7027 tgl 3752 ECB : }
3753 :
7027 tgl 3754 GIC 4769 : return newargs;
3755 : }
3756 :
3757 : /*
3758 : * Subroutine for eval_const_expressions: process arguments of an AND clause
3759 : *
3760 : * This includes flattening of nested ANDs as well as recursion to
3761 : * eval_const_expressions to simplify the AND arguments.
3762 : *
3763 : * After simplification, AND arguments are handled as follows:
3764 : * non constant: keep
3765 : * TRUE: drop (does not affect result)
7027 tgl 3766 ECB : * FALSE: force result to FALSE
3767 : * NULL: keep only one
3768 : * We must keep one NULL input because AND expressions evaluate to NULL when
2217 andres 3769 : * no input is FALSE and at least one is NULL. We don't actually include the
3770 : * NULL here, that's supposed to be done by the caller.
7027 tgl 3771 : *
3772 : * The output arguments *haveNull and *forceFalse must be initialized false
3773 : * by the caller. They will be set true if a null constant or false constant,
3774 : * respectively, is detected anywhere in the argument list.
3775 : */
3776 : static List *
6560 tgl 3777 CBC 49463 : simplify_and_arguments(List *args,
3778 : eval_const_expressions_context *context,
6560 tgl 3779 ECB : bool *haveNull, bool *forceFalse)
3780 : {
7027 tgl 3781 CBC 49463 : List *newargs = NIL;
6560 tgl 3782 ECB : List *unprocessed_args;
3783 :
3784 : /* See comments in simplify_or_arguments */
6560 tgl 3785 GIC 49463 : unprocessed_args = list_copy(args);
6560 tgl 3786 CBC 180864 : while (unprocessed_args)
3787 : {
6560 tgl 3788 GIC 132104 : Node *arg = (Node *) linitial(unprocessed_args);
3789 :
3790 132104 : unprocessed_args = list_delete_first(unprocessed_args);
3791 :
3792 : /* flatten nested ANDs as per above comment */
1531 3793 132104 : if (is_andclause(arg))
6560 tgl 3794 CBC 518 : {
1336 tgl 3795 GBC 518 : List *subargs = ((BoolExpr *) arg)->args;
3796 518 : List *oldlist = unprocessed_args;
3797 :
3798 518 : unprocessed_args = list_concat_copy(subargs, unprocessed_args);
1336 tgl 3799 EUB : /* perhaps-overly-tense code to avoid leaking old lists */
1336 tgl 3800 GIC 518 : list_free(oldlist);
6560 3801 518 : continue;
3802 : }
3803 :
3804 : /* If it's not an AND, simplify it */
3805 131586 : arg = eval_const_expressions_mutator(arg, context);
6560 tgl 3806 ECB :
3807 : /*
6385 bruce 3808 : * It is unlikely but not impossible for simplification of a non-AND
3809 : * clause to produce an AND. Recheck, but don't be too tense about it
1336 tgl 3810 : * since it's not a mainstream case. In particular we don't worry
3811 : * about const-simplifying the input twice, nor about list leakage.
6560 3812 : */
1531 tgl 3813 GIC 131586 : if (is_andclause(arg))
6560 tgl 3814 CBC 15 : {
1336 tgl 3815 GIC 15 : List *subargs = ((BoolExpr *) arg)->args;
3816 :
3817 15 : unprocessed_args = list_concat_copy(subargs, unprocessed_args);
6560 3818 15 : continue;
3819 : }
3820 :
6560 tgl 3821 ECB : /*
3822 : * OK, we have a const-simplified non-AND argument. Process it per
3823 : * comments above.
3824 : */
7027 tgl 3825 GIC 131571 : if (IsA(arg, Const))
3826 792 : {
6797 bruce 3827 1495 : Const *const_input = (Const *) arg;
7027 tgl 3828 ECB :
7027 tgl 3829 GIC 1495 : if (const_input->constisnull)
3830 9 : *haveNull = true;
7027 tgl 3831 CBC 1486 : else if (!DatumGetBool(const_input->constvalue))
3832 : {
7027 tgl 3833 GIC 703 : *forceFalse = true;
3834 :
3835 : /*
3836 : * Once we detect a FALSE result we can just exit the loop
3837 : * immediately. However, if we ever add a notion of
3838 : * non-removable functions, we'd need to keep scanning.
3839 : */
3840 703 : return NIL;
3841 : }
3842 : /* otherwise, we can drop the constant-true input */
6560 3843 792 : continue;
3844 : }
3845 :
3846 : /* else emit the simplified arg into the result list */
3847 130076 : newargs = lappend(newargs, arg);
3848 : }
3849 :
7027 3850 48760 : return newargs;
3851 : }
3852 :
3853 : /*
6587 tgl 3854 ECB : * Subroutine for eval_const_expressions: try to simplify boolean equality
3855 : * or inequality condition
3856 : *
3857 : * Inputs are the operator OID and the simplified arguments to the operator.
3858 : * Returns a simplified expression if successful, or NULL if cannot
3859 : * simplify the expression.
3860 : *
3861 : * The idea here is to reduce "x = true" to "x" and "x = false" to "NOT x",
5011 3862 : * or similarly "x <> true" to "NOT x" and "x <> false" to "x".
6587 3863 : * This is only marginally useful in itself, but doing it in constant folding
3864 : * ensures that we will recognize these forms as being equivalent in, for
3865 : * example, partial index matching.
3866 : *
3867 : * We come here only if simplify_function has failed; therefore we cannot
3868 : * see two constant inputs, nor a constant-NULL input.
3869 : */
4564 3870 : static Node *
5011 tgl 3871 CBC 225 : simplify_boolean_equality(Oid opno, List *args)
6587 tgl 3872 ECB : {
4564 3873 : Node *leftop;
3874 : Node *rightop;
6587 3875 :
6587 tgl 3876 GIC 225 : Assert(list_length(args) == 2);
6587 tgl 3877 CBC 225 : leftop = linitial(args);
3878 225 : rightop = lsecond(args);
6587 tgl 3879 GIC 225 : if (leftop && IsA(leftop, Const))
3880 : {
6587 tgl 3881 UIC 0 : Assert(!((Const *) leftop)->constisnull);
5011 tgl 3882 LBC 0 : if (opno == BooleanEqualOperator)
3883 : {
5011 tgl 3884 UIC 0 : if (DatumGetBool(((Const *) leftop)->constvalue))
4790 bruce 3885 0 : return rightop; /* true = foo */
3886 : else
4382 3887 0 : return negate_clause(rightop); /* false = foo */
3888 : }
3889 : else
5011 tgl 3890 ECB : {
5011 tgl 3891 LBC 0 : if (DatumGetBool(((Const *) leftop)->constvalue))
4382 bruce 3892 0 : return negate_clause(rightop); /* true <> foo */
3893 : else
4790 3894 0 : return rightop; /* false <> foo */
5011 tgl 3895 ECB : }
3896 : }
6587 tgl 3897 GIC 225 : if (rightop && IsA(rightop, Const))
3898 : {
3899 150 : Assert(!((Const *) rightop)->constisnull);
5011 3900 150 : if (opno == BooleanEqualOperator)
3901 : {
5011 tgl 3902 CBC 117 : if (DatumGetBool(((Const *) rightop)->constvalue))
4790 bruce 3903 60 : return leftop; /* foo = true */
5011 tgl 3904 ECB : else
4564 tgl 3905 GIC 57 : return negate_clause(leftop); /* foo = false */
5011 tgl 3906 ECB : }
6587 3907 : else
5011 3908 : {
5011 tgl 3909 GIC 33 : if (DatumGetBool(((Const *) rightop)->constvalue))
4564 tgl 3910 CBC 30 : return negate_clause(leftop); /* foo <> true */
3911 : else
4790 bruce 3912 GIC 3 : return leftop; /* foo <> false */
3913 : }
3914 : }
6587 tgl 3915 75 : return NULL;
3916 : }
6587 tgl 3917 ECB :
3918 : /*
3919 : * Subroutine for eval_const_expressions: try to simplify a function call
7423 3920 : * (which might originally have been an operator; we don't care)
3921 : *
3922 : * Inputs are the function OID, actual result type OID (which is needed for
3923 : * polymorphic functions), result typmod, result collation, the input
4034 3924 : * collation to use for the function, the original argument list (not
3925 : * const-simplified yet, unless process_args is false), and some flags;
3926 : * also the context data for eval_const_expressions.
8421 3927 : *
3928 : * Returns a simplified expression if successful, or NULL if cannot
3929 : * simplify the function call.
3930 : *
3931 : * This function is also responsible for converting named-notation argument
3932 : * lists into positional notation and/or adding any needed default argument
3933 : * expressions; which is a bit grotty, but it avoids extra fetches of the
3934 : * function's pg_proc tuple. For this reason, the args list is
3935 : * pass-by-reference. Conversion and const-simplification of the args list
3936 : * will be done even if simplification of the function call itself is not
3937 : * possible.
3938 : */
3939 : static Expr *
4034 tgl 3940 GIC 504856 : simplify_function(Oid funcid, Oid result_type, int32 result_typmod,
3941 : Oid result_collid, Oid input_collid, List **args_p,
3942 : bool funcvariadic, bool process_args, bool allow_non_const,
3943 : eval_const_expressions_context *context)
3944 : {
3945 504856 : List *args = *args_p;
3946 : HeapTuple func_tuple;
3947 : Form_pg_proc func_form;
8397 bruce 3948 ECB : Expr *newexpr;
3949 :
3950 : /*
3951 : * We have three strategies for simplification: execute the function to
3952 : * deliver a constant result, use a transform function to generate a
4310 rhaas 3953 : * substitute node tree, or expand in-line the body of the function
3954 : * definition (which only works for simple SQL-language functions, but
3260 bruce 3955 : * that is a common case). Each case needs access to the function's
4034 tgl 3956 : * pg_proc tuple, so fetch it just once.
3957 : *
4034 tgl 3958 EUB : * Note: the allow_non_const flag suppresses both the second and third
3955 bruce 3959 : * strategies; so if !allow_non_const, simplify_function can only return a
3960 : * Const or NULL. Argument-list rewriting happens anyway, though.
8421 tgl 3961 : */
4802 rhaas 3962 GBC 504856 : func_tuple = SearchSysCache1(PROCOID, ObjectIdGetDatum(funcid));
8421 tgl 3963 GIC 504856 : if (!HeapTupleIsValid(func_tuple))
7198 tgl 3964 UBC 0 : elog(ERROR, "cache lookup failed for function %u", funcid);
4034 tgl 3965 GIC 504856 : func_form = (Form_pg_proc) GETSTRUCT(func_tuple);
3966 :
3967 : /*
4034 tgl 3968 EUB : * Process the function arguments, unless the caller did it already.
3969 : *
3970 : * Here we must deal with named or defaulted arguments, and then
3971 : * recursively apply eval_const_expressions to the whole argument list.
3972 : */
4034 tgl 3973 GIC 504856 : if (process_args)
4034 tgl 3974 ECB : {
668 tgl 3975 GIC 504191 : args = expand_function_arguments(args, false, result_type, func_tuple);
4034 tgl 3976 CBC 504191 : args = (List *) expression_tree_mutator((Node *) args,
2118 tgl 3977 ECB : eval_const_expressions_mutator,
3978 : (void *) context);
4034 3979 : /* Argument processing done, give it back to the caller */
4034 tgl 3980 CBC 504136 : *args_p = args;
3981 : }
4034 tgl 3982 ECB :
3983 : /* Now attempt simplification of the function call proper. */
3984 :
4404 tgl 3985 GIC 504801 : newexpr = evaluate_function(funcid, result_type, result_typmod,
3730 tgl 3986 ECB : result_collid, input_collid,
3987 : args, funcvariadic,
3988 : func_tuple, context);
7434 3989 :
1520 tgl 3990 GIC 503287 : if (!newexpr && allow_non_const && OidIsValid(func_form->prosupport))
3991 : {
4034 tgl 3992 ECB : /*
3993 : * Build a SupportRequestSimplify node to pass to the support
3994 : * function, pointing to a dummy FuncExpr node containing the
3995 : * simplified arg list. We use this approach to present a uniform
3996 : * interface to the support function regardless of how the target
3997 : * function is actually being invoked.
3998 : */
3999 : SupportRequestSimplify req;
4000 : FuncExpr fexpr;
4001 :
4034 tgl 4002 GIC 13485 : fexpr.xpr.type = T_FuncExpr;
4003 13485 : fexpr.funcid = funcid;
4004 13485 : fexpr.funcresulttype = result_type;
4005 13485 : fexpr.funcretset = func_form->proretset;
3730 4006 13485 : fexpr.funcvariadic = funcvariadic;
3831 4007 13485 : fexpr.funcformat = COERCE_EXPLICIT_CALL;
4034 4008 13485 : fexpr.funccollid = result_collid;
4009 13485 : fexpr.inputcollid = input_collid;
4010 13485 : fexpr.args = args;
4011 13485 : fexpr.location = -1;
4012 :
1520 4013 13485 : req.type = T_SupportRequestSimplify;
4014 13485 : req.root = context->root;
4015 13485 : req.fcall = &fexpr;
4016 :
4034 tgl 4017 ECB : newexpr = (Expr *)
1520 tgl 4018 GIC 13485 : DatumGetPointer(OidFunctionCall1(func_form->prosupport,
4019 : PointerGetDatum(&req)));
4020 :
4021 : /* catch a possible API misunderstanding */
1520 tgl 4022 CBC 13485 : Assert(newexpr != (Expr *) &fexpr);
4023 : }
4024 :
4034 tgl 4025 GIC 503287 : if (!newexpr && allow_non_const)
4398 4026 418901 : newexpr = inline_function(funcid, result_type, result_collid,
4027 : input_collid, args, funcvariadic,
4028 : func_tuple, context);
4029 :
8179 4030 503276 : ReleaseSysCache(func_tuple);
4031 :
7434 4032 503276 : return newexpr;
4033 : }
4034 :
4035 : /*
4036 : * expand_function_arguments: convert named-notation args to positional args
4037 : * and/or insert default args, as needed
4038 : *
668 tgl 4039 ECB : * Returns a possibly-transformed version of the args list.
4040 : *
668 tgl 4041 EUB : * If include_out_arguments is true, then the args list and the result
668 tgl 4042 ECB : * include OUT arguments.
4043 : *
4044 : * The expected result type of the call must be given, for sanity-checking
4045 : * purposes. Also, we ask the caller to provide the function's actual
4046 : * pg_proc tuple, not just its OID.
4047 : *
4048 : * If we need to change anything, the input argument list is copied, not
4049 : * modified.
4034 4050 : *
4051 : * Note: this gets applied to operator argument lists too, even though the
4052 : * cases it handles should never occur there. This should be OK since it
4053 : * will fall through very quickly if there's nothing to do.
4054 : */
4055 : List *
668 tgl 4056 GIC 505812 : expand_function_arguments(List *args, bool include_out_arguments,
668 tgl 4057 ECB : Oid result_type, HeapTuple func_tuple)
4058 : {
4034 tgl 4059 GIC 505812 : Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
668 4060 505812 : Oid *proargtypes = funcform->proargtypes.values;
4061 505812 : int pronargs = funcform->pronargs;
4034 tgl 4062 CBC 505812 : bool has_named_args = false;
4063 : ListCell *lc;
4064 :
4065 : /*
4066 : * If we are asked to match to OUT arguments, then use the proallargtypes
668 tgl 4067 ECB : * array (which includes those); otherwise use proargtypes (which
4068 : * doesn't). Of course, if proallargtypes is null, we always use
4069 : * proargtypes. (Fetching proallargtypes is annoyingly expensive
4070 : * considering that we may have nothing to do here, but fortunately the
4071 : * common case is include_out_arguments == false.)
4072 : */
668 tgl 4073 GIC 505812 : if (include_out_arguments)
4074 : {
4075 : Datum proallargtypes;
4076 : bool isNull;
4077 :
4078 193 : proallargtypes = SysCacheGetAttr(PROCOID, func_tuple,
668 tgl 4079 ECB : Anum_pg_proc_proallargtypes,
4080 : &isNull);
668 tgl 4081 CBC 193 : if (!isNull)
668 tgl 4082 ECB : {
668 tgl 4083 CBC 76 : ArrayType *arr = DatumGetArrayTypeP(proallargtypes);
668 tgl 4084 ECB :
668 tgl 4085 CBC 76 : pronargs = ARR_DIMS(arr)[0];
4086 76 : if (ARR_NDIM(arr) != 1 ||
4087 76 : pronargs < 0 ||
4088 76 : ARR_HASNULL(arr) ||
668 tgl 4089 GIC 76 : ARR_ELEMTYPE(arr) != OIDOID)
668 tgl 4090 LBC 0 : elog(ERROR, "proallargtypes is not a 1-D Oid array or it contains nulls");
668 tgl 4091 CBC 76 : Assert(pronargs >= funcform->pronargs);
4092 76 : proargtypes = (Oid *) ARR_DATA_PTR(arr);
4093 : }
4094 : }
668 tgl 4095 ECB :
4096 : /* Do we have any named arguments? */
4034 tgl 4097 GIC 1305211 : foreach(lc, args)
4098 : {
4034 tgl 4099 CBC 805330 : Node *arg = (Node *) lfirst(lc);
4100 :
4034 tgl 4101 GIC 805330 : if (IsA(arg, NamedArgExpr))
4034 tgl 4102 ECB : {
4034 tgl 4103 CBC 5931 : has_named_args = true;
4034 tgl 4104 GIC 5931 : break;
4105 : }
4106 : }
4034 tgl 4107 ECB :
4108 : /* If so, we must apply reorder_function_arguments */
4034 tgl 4109 CBC 505812 : if (has_named_args)
4110 : {
668 tgl 4111 GIC 5931 : args = reorder_function_arguments(args, pronargs, func_tuple);
4112 : /* Recheck argument types and add casts if needed */
4113 5931 : recheck_cast_function_args(args, result_type,
4114 : proargtypes, pronargs,
4115 : func_tuple);
4116 : }
4117 499881 : else if (list_length(args) < pronargs)
4118 : {
4119 : /* No named args, but we seem to be short some defaults */
4120 1921 : args = add_function_defaults(args, pronargs, func_tuple);
4121 : /* Recheck argument types and add casts if needed */
4122 1921 : recheck_cast_function_args(args, result_type,
4123 : proargtypes, pronargs,
4124 : func_tuple);
4125 : }
4126 :
4034 4127 505812 : return args;
4128 : }
4129 :
4130 : /*
4131 : * reorder_function_arguments: convert named-notation args to positional args
4132 : *
4931 tgl 4133 ECB : * This function also inserts default argument values as needed, since it's
4134 : * impossible to form a truly valid positional call without that.
4135 : */
4136 : static List *
668 tgl 4137 CBC 5931 : reorder_function_arguments(List *args, int pronargs, HeapTuple func_tuple)
4931 tgl 4138 ECB : {
4931 tgl 4139 CBC 5931 : Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4931 tgl 4140 GIC 5931 : int nargsprovided = list_length(args);
4141 : Node *argarray[FUNC_MAX_ARGS];
4142 : ListCell *lc;
4143 : int i;
4144 :
4145 5931 : Assert(nargsprovided <= pronargs);
801 4146 5931 : if (pronargs < 0 || pronargs > FUNC_MAX_ARGS)
4931 tgl 4147 UIC 0 : elog(ERROR, "too many function arguments");
801 tgl 4148 GIC 5931 : memset(argarray, 0, pronargs * sizeof(Node *));
4149 :
4931 tgl 4150 ECB : /* Deconstruct the argument list into an array indexed by argnumber */
4931 tgl 4151 GIC 5931 : i = 0;
4152 23957 : foreach(lc, args)
4153 : {
4790 bruce 4154 18026 : Node *arg = (Node *) lfirst(lc);
4931 tgl 4155 ECB :
4931 tgl 4156 GIC 18026 : if (!IsA(arg, NamedArgExpr))
4157 : {
4931 tgl 4158 ECB : /* positional argument, assumed to precede all named args */
4931 tgl 4159 GIC 727 : Assert(argarray[i] == NULL);
4931 tgl 4160 CBC 727 : argarray[i++] = arg;
4161 : }
4931 tgl 4162 ECB : else
4163 : {
4931 tgl 4164 CBC 17299 : NamedArgExpr *na = (NamedArgExpr *) arg;
4931 tgl 4165 ECB :
668 tgl 4166 CBC 17299 : Assert(na->argnumber >= 0 && na->argnumber < pronargs);
4931 tgl 4167 GBC 17299 : Assert(argarray[na->argnumber] == NULL);
4931 tgl 4168 CBC 17299 : argarray[na->argnumber] = (Node *) na->arg;
4931 tgl 4169 ECB : }
4170 : }
4171 :
4172 : /*
4173 : * Fetch default expressions, if needed, and insert into array at proper
4790 bruce 4174 : * locations (they aren't necessarily consecutive or all used)
4175 : */
4931 tgl 4176 CBC 5931 : if (nargsprovided < pronargs)
4177 : {
4790 bruce 4178 2786 : List *defaults = fetch_function_defaults(func_tuple);
4179 :
4931 tgl 4180 2786 : i = pronargs - funcform->pronargdefaults;
4181 15950 : foreach(lc, defaults)
4182 : {
4931 tgl 4183 GIC 13164 : if (argarray[i] == NULL)
4184 5467 : argarray[i] = (Node *) lfirst(lc);
4185 13164 : i++;
4931 tgl 4186 ECB : }
4187 : }
4188 :
4189 : /* Now reconstruct the args list in proper order */
4931 tgl 4190 CBC 5931 : args = NIL;
4931 tgl 4191 GIC 29424 : for (i = 0; i < pronargs; i++)
4192 : {
4193 23493 : Assert(argarray[i] != NULL);
4931 tgl 4194 CBC 23493 : args = lappend(args, argarray[i]);
4195 : }
4196 :
4197 5931 : return args;
4198 : }
4931 tgl 4199 ECB :
4200 : /*
4201 : * add_function_defaults: add missing function arguments from its defaults
4202 : *
4203 : * This is used only when the argument list was positional to begin with,
4204 : * and so we know we just need to add defaults at the end.
4205 : */
4206 : static List *
668 tgl 4207 GIC 1921 : add_function_defaults(List *args, int pronargs, HeapTuple func_tuple)
4208 : {
5206 4209 1921 : int nargsprovided = list_length(args);
4210 : List *defaults;
4211 : int ndelete;
4212 :
4213 : /* Get all the default expressions from the pg_proc tuple */
4931 tgl 4214 CBC 1921 : defaults = fetch_function_defaults(func_tuple);
4215 :
5225 tgl 4216 ECB : /* Delete any unused defaults from the list */
668 tgl 4217 CBC 1921 : ndelete = nargsprovided + list_length(defaults) - pronargs;
5225 tgl 4218 GIC 1921 : if (ndelete < 0)
5225 tgl 4219 UIC 0 : elog(ERROR, "not enough default arguments");
1362 tgl 4220 GIC 1921 : if (ndelete > 0)
523 4221 104 : defaults = list_delete_first_n(defaults, ndelete);
4931 tgl 4222 ECB :
4034 4223 : /* And form the combined argument list, not modifying the input list */
1336 tgl 4224 GBC 1921 : return list_concat_copy(args, defaults);
4931 tgl 4225 ECB : }
4226 :
4227 : /*
4228 : * fetch_function_defaults: get function's default arguments as expression list
4229 : */
4230 : static List *
4931 tgl 4231 CBC 4707 : fetch_function_defaults(HeapTuple func_tuple)
4232 : {
4931 tgl 4233 ECB : List *defaults;
4234 : Datum proargdefaults;
4235 : char *str;
4236 :
15 dgustafsson 4237 GNC 4707 : proargdefaults = SysCacheGetAttrNotNull(PROCOID, func_tuple,
4238 : Anum_pg_proc_proargdefaults);
4931 tgl 4239 CBC 4707 : str = TextDatumGetCString(proargdefaults);
2238 peter_e 4240 4707 : defaults = castNode(List, stringToNode(str));
4931 tgl 4241 GIC 4707 : pfree(str);
4242 4707 : return defaults;
4243 : }
4244 :
4245 : /*
4246 : * recheck_cast_function_args: recheck function args and typecast as needed
4247 : * after adding defaults.
4931 tgl 4248 ECB : *
4249 : * It is possible for some of the defaulted arguments to be polymorphic;
4250 : * therefore we can't assume that the default expressions have the correct
4251 : * data types already. We have to re-resolve polymorphics and do coercion
4252 : * just like the parser did.
4253 : *
4254 : * This should be a no-op if there are no polymorphic arguments,
4255 : * but we do it anyway to be sure.
4256 : *
4034 4257 : * Note: if any casts are needed, the args list is modified in-place;
4258 : * caller should have already copied the list structure.
4259 : */
4260 : static void
668 tgl 4261 GIC 7852 : recheck_cast_function_args(List *args, Oid result_type,
668 tgl 4262 ECB : Oid *proargtypes, int pronargs,
4263 : HeapTuple func_tuple)
4264 : {
4931 tgl 4265 CBC 7852 : Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4931 tgl 4266 ECB : int nargs;
4267 : Oid actual_arg_types[FUNC_MAX_ARGS];
4268 : Oid declared_arg_types[FUNC_MAX_ARGS];
4269 : Oid rettype;
4270 : ListCell *lc;
4271 :
5225 tgl 4272 GIC 7852 : if (list_length(args) > FUNC_MAX_ARGS)
5225 tgl 4273 UIC 0 : elog(ERROR, "too many function arguments");
5225 tgl 4274 GIC 7852 : nargs = 0;
4275 37816 : foreach(lc, args)
4276 : {
4277 29964 : actual_arg_types[nargs++] = exprType((Node *) lfirst(lc));
4278 : }
668 tgl 4279 CBC 7852 : Assert(nargs == pronargs);
668 tgl 4280 GIC 7852 : memcpy(declared_arg_types, proargtypes, pronargs * sizeof(Oid));
5225 tgl 4281 CBC 7852 : rettype = enforce_generic_type_consistency(actual_arg_types,
4282 : declared_arg_types,
4283 : nargs,
4284 : funcform->prorettype,
4285 : false);
5225 tgl 4286 ECB : /* let's just check we got the same answer as the parser did ... */
5225 tgl 4287 GIC 7852 : if (rettype != result_type)
5225 tgl 4288 UIC 0 : elog(ERROR, "function's resolved result type changed during planning");
5225 tgl 4289 ECB :
4290 : /* perform any necessary typecasting of arguments */
5225 tgl 4291 GBC 7852 : make_fn_arguments(NULL, args, actual_arg_types, declared_arg_types);
5225 tgl 4292 CBC 7852 : }
5225 tgl 4293 ECB :
4294 : /*
4295 : * evaluate_function: try to pre-evaluate a function call
7434 4296 : *
4297 : * We can do this if the function is strict and has any constant-null inputs
4298 : * (just return a null constant), or if the function is immutable and has all
4299 : * constant inputs (call it and return the result as a Const node). In
4300 : * estimation mode we are willing to pre-evaluate stable functions too.
4301 : *
4302 : * Returns a simplified expression if successful, or NULL if cannot
7423 4303 : * simplify the function.
4304 : */
4305 : static Expr *
4404 tgl 4306 GIC 504801 : evaluate_function(Oid funcid, Oid result_type, int32 result_typmod,
4307 : Oid result_collid, Oid input_collid, List *args,
4308 : bool funcvariadic,
4398 tgl 4309 ECB : HeapTuple func_tuple,
4310 : eval_const_expressions_context *context)
7434 4311 : {
7434 tgl 4312 CBC 504801 : Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4313 504801 : bool has_nonconst_input = false;
4314 504801 : bool has_null_input = false;
4315 : ListCell *arg;
4316 : FuncExpr *newexpr;
4317 :
4318 : /*
4319 : * Can't simplify if it returns a set.
4320 : */
7434 tgl 4321 GIC 504801 : if (funcform->proretset)
8421 4322 22904 : return NULL;
4323 :
4324 : /*
4325 : * Can't simplify if it returns RECORD. The immediate problem is that it
4326 : * will be needing an expected tupdesc which we can't supply here.
4327 : *
4328 : * In the case where it has OUT parameters, it could get by without an
4329 : * expected tupdesc, but we still have issues: get_expr_result_type()
4330 : * doesn't know how to extract type info from a RECORD constant, and in
4331 : * the case of a NULL function result there doesn't seem to be any clean
4332 : * way to fix that. In view of the likelihood of there being still other
6385 bruce 4333 ECB : * gotchas, seems best to leave the function call unreduced.
4334 : */
6489 tgl 4335 GIC 481897 : if (funcform->prorettype == RECORDOID)
6569 4336 1846 : return NULL;
6569 tgl 4337 ECB :
4338 : /*
4339 : * Check for constant inputs and especially constant-NULL inputs.
4340 : */
7434 tgl 4341 GIC 1243546 : foreach(arg, args)
4342 : {
4343 763495 : if (IsA(lfirst(arg), Const))
7434 tgl 4344 CBC 348413 : has_null_input |= ((Const *) lfirst(arg))->constisnull;
7434 tgl 4345 EUB : else
7434 tgl 4346 CBC 415082 : has_nonconst_input = true;
8351 tgl 4347 ECB : }
4348 :
4349 : /*
4350 : * If the function is strict and has a constant-NULL input, it will never
6385 bruce 4351 : * be called at all, so we can replace the call by a NULL constant, even
4352 : * if there are other inputs that aren't constant, and even if the
4353 : * function is not otherwise immutable.
4354 : */
7434 tgl 4355 GIC 480051 : if (funcform->proisstrict && has_null_input)
4398 4356 272 : return (Expr *) makeNullConst(result_type, result_typmod,
4357 : result_collid);
4358 :
7434 tgl 4359 ECB : /*
6725 tgl 4360 EUB : * Otherwise, can simplify only if all inputs are constants. (For a
4361 : * non-strict function, constant NULL inputs are treated the same as
4362 : * constant non-NULL inputs.)
7434 tgl 4363 ECB : */
6725 tgl 4364 CBC 479779 : if (has_nonconst_input)
6725 tgl 4365 GIC 320442 : return NULL;
4366 :
4367 : /*
4368 : * Ordinarily we are only allowed to simplify immutable functions. But for
4369 : * purposes of estimation, we consider it okay to simplify functions that
4370 : * are merely stable; the risk that the result might change from planning
4371 : * time to execution time is worth taking in preference to not being able
4372 : * to estimate the value at all.
4373 : */
4374 159337 : if (funcform->provolatile == PROVOLATILE_IMMUTABLE)
4375 : /* okay */ ;
4376 75030 : else if (context->estimate && funcform->provolatile == PROVOLATILE_STABLE)
4377 : /* okay */ ;
6725 tgl 4378 ECB : else
8351 tgl 4379 GIC 73727 : return NULL;
4380 :
4381 : /*
4382 : * OK, looks like we can simplify this operator/function.
4383 : *
7423 tgl 4384 ECB : * Build a new FuncExpr node containing the already-simplified arguments.
8421 4385 : */
7423 tgl 4386 CBC 85610 : newexpr = makeNode(FuncExpr);
7423 tgl 4387 GIC 85610 : newexpr->funcid = funcid;
7306 4388 85610 : newexpr->funcresulttype = result_type;
7423 4389 85610 : newexpr->funcretset = false;
3730 4390 85610 : newexpr->funcvariadic = funcvariadic;
3602 bruce 4391 85610 : newexpr->funcformat = COERCE_EXPLICIT_CALL; /* doesn't matter */
2118 tgl 4392 85610 : newexpr->funccollid = result_collid; /* doesn't matter */
4404 tgl 4393 CBC 85610 : newexpr->inputcollid = input_collid;
8421 4394 85610 : newexpr->args = args;
5337 tgl 4395 GIC 85610 : newexpr->location = -1;
4396 :
4398 4397 85610 : return evaluate_expr((Expr *) newexpr, result_type, result_typmod,
4398 : result_collid);
4399 : }
4400 :
4401 : /*
4402 : * inline_function: try to expand a function call inline
4403 : *
4404 : * If the function is a sufficiently simple SQL-language function
4405 : * (just "SELECT expression"), then we can inline it and avoid the rather
4406 : * high per-call overhead of SQL functions. Furthermore, this can expose
7434 tgl 4407 ECB : * opportunities for constant-folding within the function expression.
4408 : *
4409 : * We have to beware of some special cases however. A directly or
4410 : * indirectly recursive function would cause us to recurse forever,
4411 : * so we keep track of which functions we are already expanding and
4412 : * do not re-expand them. Also, if a parameter is used more than once
4413 : * in the SQL-function body, we require it not to contain any volatile
4414 : * functions (volatiles might deliver inconsistent answers) nor to be
3260 bruce 4415 : * unreasonably expensive to evaluate. The expensiveness check not only
7189 tgl 4416 : * prevents us from doing multiple evaluations of an expensive parameter
4417 : * at runtime, but is a safety value to limit growth of an expression due
4418 : * to repeated inlining.
4419 : *
4420 : * We must also beware of changing the volatility or strictness status of
4421 : * functions by inlining them.
4422 : *
4423 : * Also, at the moment we can't inline functions returning RECORD. This
4424 : * doesn't work in the general case because it discards information such
4425 : * as OUT-parameter declarations.
4426 : *
2435 4427 : * Also, context-dependent expression nodes in the argument list are trouble.
4428 : *
4429 : * Returns a simplified expression if successful, or NULL if cannot
4430 : * simplify the function.
4431 : */
4432 : static Expr *
4398 tgl 4433 GIC 418901 : inline_function(Oid funcid, Oid result_type, Oid result_collid,
4434 : Oid input_collid, List *args,
4435 : bool funcvariadic,
6876 tgl 4436 ECB : HeapTuple func_tuple,
4437 : eval_const_expressions_context *context)
4438 : {
7434 tgl 4439 GIC 418901 : Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
4440 : char *src;
4441 : Datum tmp;
4442 : bool isNull;
4443 : MemoryContext oldcxt;
4444 : MemoryContext mycxt;
4445 : inline_error_callback_arg callback_arg;
7195 tgl 4446 ECB : ErrorContextCallback sqlerrcontext;
4447 : FuncExpr *fexpr;
4399 4448 : SQLFunctionParseInfoPtr pinfo;
4449 : TupleDesc rettupdesc;
4450 : ParseState *pstate;
7434 4451 : List *raw_parsetree_list;
4452 : List *querytree_list;
4453 : Query *querytree;
4454 : Node *newexpr;
4455 : int *usecounts;
4456 : ListCell *arg;
4457 : int i;
4458 :
4459 : /*
6385 bruce 4460 : * Forget it if the function is not SQL-language or has other showstopper
1864 peter_e 4461 : * properties. (The prokind and nargs checks are just paranoia.)
7434 tgl 4462 : */
7434 tgl 4463 CBC 418901 : if (funcform->prolang != SQLlanguageId ||
1864 peter_e 4464 11886 : funcform->prokind != PROKIND_FUNCTION ||
1850 tgl 4465 11886 : funcform->prosecdef ||
7434 4466 11880 : funcform->proretset ||
4512 4467 11230 : funcform->prorettype == RECORDOID ||
1838 andrew 4468 GIC 22157 : !heap_attisnull(func_tuple, Anum_pg_proc_proconfig, NULL) ||
6888 neilc 4469 CBC 11074 : funcform->pronargs != list_length(args))
7434 tgl 4470 GIC 407827 : return NULL;
4471 :
4472 : /* Check for recursive function, and give up trying to expand if so */
6876 4473 11074 : if (list_member_oid(context->active_fns, funcid))
7434 4474 4254 : return NULL;
4475 :
4476 : /* Check permission to call function (fail later, if not) */
147 peter 4477 GNC 6820 : if (object_aclcheck(ProcedureRelationId, funcid, GetUserId(), ACL_EXECUTE) != ACLCHECK_OK)
7434 tgl 4478 GIC 7 : return NULL;
4479 :
4480 : /* Check whether a plugin wants to hook function entry/exit */
4500 rhaas 4481 6813 : if (FmgrHookIsNeeded(funcid))
4500 rhaas 4482 UIC 0 : return NULL;
4483 :
4484 : /*
4485 : * Make a temporary memory context, so that we don't leak all the stuff
4486 : * that parsing might create.
4487 : */
7434 tgl 4488 GIC 6813 : mycxt = AllocSetContextCreate(CurrentMemoryContext,
4489 : "inline_function",
4490 : ALLOCSET_DEFAULT_SIZES);
4491 6813 : oldcxt = MemoryContextSwitchTo(mycxt);
4492 :
4493 : /*
4494 : * We need a dummy FuncExpr node containing the already-simplified
4495 : * arguments. (In some cases we don't really need it, but building it is
4496 : * cheap enough that it's not worth contortions to avoid.)
4497 : */
684 4498 6813 : fexpr = makeNode(FuncExpr);
4499 6813 : fexpr->funcid = funcid;
4500 6813 : fexpr->funcresulttype = result_type;
4501 6813 : fexpr->funcretset = false;
4502 6813 : fexpr->funcvariadic = funcvariadic;
4503 6813 : fexpr->funcformat = COERCE_EXPLICIT_CALL; /* doesn't matter */
4504 6813 : fexpr->funccollid = result_collid; /* doesn't matter */
684 tgl 4505 CBC 6813 : fexpr->inputcollid = input_collid;
684 tgl 4506 GIC 6813 : fexpr->args = args;
4507 6813 : fexpr->location = -1;
4508 :
4509 : /* Fetch the function body */
15 dgustafsson 4510 GNC 6813 : tmp = SysCacheGetAttrNotNull(PROCOID, func_tuple, Anum_pg_proc_prosrc);
724 tgl 4511 GIC 6813 : src = TextDatumGetCString(tmp);
4512 :
4513 : /*
4514 : * Setup error traceback support for ereport(). This is so that we can
4515 : * finger the function that bad information came from.
4516 : */
4769 4517 6813 : callback_arg.proname = NameStr(funcform->proname);
724 4518 6813 : callback_arg.prosrc = src;
4519 :
4769 4520 6813 : sqlerrcontext.callback = sql_inline_error_callback;
4521 6813 : sqlerrcontext.arg = (void *) &callback_arg;
4522 6813 : sqlerrcontext.previous = error_context_stack;
4523 6813 : error_context_stack = &sqlerrcontext;
4524 :
4525 : /* If we have prosqlbody, pay attention to that not prosrc */
732 peter 4526 6813 : tmp = SysCacheGetAttr(PROCOID,
4527 : func_tuple,
4528 : Anum_pg_proc_prosqlbody,
4529 : &isNull);
724 tgl 4530 CBC 6813 : if (!isNull)
732 peter 4531 ECB : {
4532 : Node *n;
4533 : List *query_list;
4534 :
732 peter 4535 CBC 1241 : n = stringToNode(TextDatumGetCString(tmp));
4536 1241 : if (IsA(n, List))
186 drowley 4537 GNC 895 : query_list = linitial_node(List, castNode(List, n));
4538 : else
4539 346 : query_list = list_make1(n);
4540 1241 : if (list_length(query_list) != 1)
732 peter 4541 CBC 3 : goto fail;
186 drowley 4542 GNC 1238 : querytree = linitial(query_list);
4543 :
586 tgl 4544 ECB : /*
4545 : * Because we'll insist below that the querytree have an empty rtable
4546 : * and no sublinks, it cannot have any relation references that need
4547 : * to be locked or rewritten. So we can omit those steps.
4548 : */
732 peter 4549 EUB : }
4550 : else
4551 : {
4552 : /* Set up to handle parameters while parsing the function body. */
697 tgl 4553 GIC 5572 : pinfo = prepare_sql_fn_parse_info(func_tuple,
4554 : (Node *) fexpr,
697 tgl 4555 ECB : input_collid);
4556 :
4557 : /*
4558 : * We just do parsing and parse analysis, not rewriting, because
4559 : * rewriting will not affect table-free-SELECT-only queries, which is
4560 : * all that we care about. Also, we can punt as soon as we detect
4561 : * more than one command in the function body.
4562 : */
697 tgl 4563 GIC 5572 : raw_parsetree_list = pg_parse_query(src);
4564 5572 : if (list_length(raw_parsetree_list) != 1)
697 tgl 4565 CBC 37 : goto fail;
7434 tgl 4566 ECB :
697 tgl 4567 CBC 5535 : pstate = make_parsestate(NULL);
4568 5535 : pstate->p_sourcetext = src;
4569 5535 : sql_fn_parser_setup(pstate, pinfo);
4399 tgl 4570 ECB :
697 tgl 4571 CBC 5535 : querytree = transformTopLevelStmt(pstate, linitial(raw_parsetree_list));
4399 tgl 4572 ECB :
697 tgl 4573 CBC 5528 : free_parsestate(pstate);
732 peter 4574 ECB : }
4575 :
4576 : /*
7434 tgl 4577 : * The single command must be a simple "SELECT expression".
2063 4578 : *
4579 : * Note: if you change the tests involved in this, see also plpgsql's
4580 : * exec_simple_check_plan(). That generally needs to have the same idea
4581 : * of what's a "simple expression", so that inlining a function that
4582 : * previously wasn't inlined won't change plpgsql's conclusion.
4583 : */
7434 tgl 4584 CBC 6766 : if (!IsA(querytree, Query) ||
4585 6766 : querytree->commandType != CMD_SELECT ||
7434 tgl 4586 GIC 6709 : querytree->hasAggs ||
5215 tgl 4587 CBC 6671 : querytree->hasWindowFuncs ||
2399 4588 6671 : querytree->hasTargetSRFs ||
7434 4589 6671 : querytree->hasSubLinks ||
5300 4590 6389 : querytree->cteList ||
7434 tgl 4591 GIC 6389 : querytree->rtable ||
4592 5731 : querytree->jointree->fromlist ||
7434 tgl 4593 CBC 5731 : querytree->jointree->quals ||
7434 tgl 4594 GIC 5731 : querytree->groupClause ||
2885 andres 4595 5731 : querytree->groupingSets ||
7434 tgl 4596 5731 : querytree->havingQual ||
5215 tgl 4597 CBC 5731 : querytree->windowClause ||
7434 tgl 4598 GIC 5731 : querytree->distinctClause ||
4599 5731 : querytree->sortClause ||
4600 5731 : querytree->limitOffset ||
4601 5731 : querytree->limitCount ||
7434 tgl 4602 CBC 11372 : querytree->setOperations ||
6888 neilc 4603 5686 : list_length(querytree->targetList) != 1)
7434 tgl 4604 1110 : goto fail;
4605 :
684 tgl 4606 ECB : /* If the function result is composite, resolve it */
684 tgl 4607 CBC 5656 : (void) get_expr_result_type((Node *) fexpr,
684 tgl 4608 ECB : NULL,
4609 : &rettupdesc);
4610 :
4611 : /*
4612 : * Make sure the function (still) returns what it's declared to. This
4613 : * will raise an error if wrong, but that's okay since the function would
4614 : * fail at runtime anyway. Note that check_sql_fn_retval will also insert
4615 : * a coercion if needed to make the tlist expression match the declared
4616 : * type of the function.
4617 : *
4618 : * Note: we do not try this until we have verified that no rewriting was
4619 : * needed; that's probably not important, but let's be careful.
7222 4620 : */
1187 tgl 4621 GIC 5656 : querytree_list = list_make1(querytree);
902 4622 5656 : if (check_sql_fn_retval(list_make1(querytree_list),
4623 : result_type, rettupdesc,
4624 : false, NULL))
5878 4625 6 : goto fail; /* reject whole-tuple-result cases */
4626 :
4627 : /*
4628 : * Given the tests above, check_sql_fn_retval shouldn't have decided to
4629 : * inject a projection step, but let's just make sure.
1187 tgl 4630 ECB : */
1187 tgl 4631 CBC 5647 : if (querytree != linitial(querytree_list))
1187 tgl 4632 LBC 0 : goto fail;
4633 :
5500 tgl 4634 ECB : /* Now we can grab the tlist expression */
5500 tgl 4635 CBC 5647 : newexpr = (Node *) ((TargetEntry *) linitial(querytree->targetList))->expr;
5500 tgl 4636 ECB :
4637 : /*
1850 4638 : * If the SQL function returns VOID, we can only inline it if it is a
4639 : * SELECT of an expression returning VOID (ie, it's just a redirection to
4640 : * another VOID-returning function). In all non-VOID-returning cases,
4641 : * check_sql_fn_retval should ensure that newexpr returns the function's
4642 : * declared result type, so this test shouldn't fail otherwise; but we may
4643 : * as well cope gracefully if it does.
4644 : */
1850 tgl 4645 GIC 5647 : if (exprType(newexpr) != result_type)
4646 9 : goto fail;
4647 :
4648 : /*
4649 : * Additional validity checks on the expression. It mustn't be more
4650 : * volatile than the surrounding function (this is to avoid breaking hacks
2399 tgl 4651 ECB : * that involve pretending a function is immutable when it really ain't).
4652 : * If the surrounding function is declared strict, then the expression
4653 : * must contain only strict constructs and must use all of the function
4654 : * parameters (this is overkill, but an exact analysis is hard).
7434 4655 : */
7434 tgl 4656 CBC 5997 : if (funcform->provolatile == PROVOLATILE_IMMUTABLE &&
4657 359 : contain_mutable_functions(newexpr))
4658 3 : goto fail;
4659 5934 : else if (funcform->provolatile == PROVOLATILE_STABLE &&
7188 bruce 4660 299 : contain_volatile_functions(newexpr))
7434 tgl 4661 LBC 0 : goto fail;
7434 tgl 4662 ECB :
7434 tgl 4663 CBC 6323 : if (funcform->proisstrict &&
4664 688 : contain_nonstrict_functions(newexpr))
4665 71 : goto fail;
7434 tgl 4666 ECB :
2435 4667 : /*
4668 : * If any parameter expression contains a context-dependent node, we can't
4669 : * inline, for fear of putting such a node into the wrong context.
4670 : */
2435 tgl 4671 CBC 5564 : if (contain_context_dependent_node((Node *) args))
2435 tgl 4672 GIC 3 : goto fail;
4673 :
7434 tgl 4674 ECB : /*
4675 : * We may be able to do it; there are still checks on parameter usage to
4676 : * make, but those are most easily done in combination with the actual
4677 : * substitution of the inputs. So start building expression with inputs
4678 : * substituted.
4679 : */
6882 tgl 4680 GIC 5561 : usecounts = (int *) palloc0(funcform->pronargs * sizeof(int));
7434 4681 5561 : newexpr = substitute_actual_parameters(newexpr, funcform->pronargs,
4682 : args, usecounts);
4683 :
4684 : /* Now check for parameter usage */
4685 5561 : i = 0;
4686 7652 : foreach(arg, args)
4687 : {
7188 bruce 4688 CBC 2095 : Node *param = lfirst(arg);
7434 tgl 4689 ECB :
7434 tgl 4690 GIC 2095 : if (usecounts[i] == 0)
4691 : {
7434 tgl 4692 ECB : /* Param not used at all: uncool if func is strict */
7434 tgl 4693 GIC 82 : if (funcform->proisstrict)
4694 4 : goto fail;
4695 : }
4696 2013 : else if (usecounts[i] != 1)
4697 : {
7189 tgl 4698 ECB : /* Param used multiple times: uncool if expensive or volatile */
7189 tgl 4699 EUB : QualCost eval_cost;
4700 :
4701 : /*
6385 bruce 4702 ECB : * We define "expensive" as "contains any subplan or more than 10
4703 : * operators". Note that the subplan search has to be done
4704 : * explicitly, since cost_qual_eval() will barf on unplanned
4705 : * subselects.
4706 : */
7189 tgl 4707 GIC 82 : if (contain_subplans(param))
4708 4 : goto fail;
5890 4709 82 : cost_qual_eval(&eval_cost, list_make1(param), NULL);
7189 4710 82 : if (eval_cost.startup + eval_cost.per_tuple >
4711 82 : 10 * cpu_operator_cost)
7189 tgl 4712 LBC 0 : goto fail;
7188 bruce 4713 ECB :
4714 : /*
4715 : * Check volatility last since this is more expensive than the
4716 : * above tests
4717 : */
7189 tgl 4718 GIC 82 : if (contain_volatile_functions(param))
7434 4719 4 : goto fail;
4720 : }
4721 2091 : i++;
4722 : }
7434 tgl 4723 ECB :
4724 : /*
6385 bruce 4725 : * Whew --- we can make the substitution. Copy the modified expression
4726 : * out of the temporary memory context, and clean up.
7434 tgl 4727 : */
7434 tgl 4728 GBC 5557 : MemoryContextSwitchTo(oldcxt);
4729 :
7434 tgl 4730 CBC 5557 : newexpr = copyObject(newexpr);
7434 tgl 4731 ECB :
7434 tgl 4732 CBC 5557 : MemoryContextDelete(mycxt);
4733 :
4734 : /*
4735 : * If the result is of a collatable type, force the result to expose the
4736 : * correct collation. In most cases this does not matter, but it's
4737 : * possible that the function result is used directly as a sort key or in
4382 bruce 4738 ECB : * other places where we expect exprCollation() to tell the truth.
4399 tgl 4739 : */
4398 tgl 4740 GIC 5557 : if (OidIsValid(result_collid))
4741 : {
4382 bruce 4742 576 : Oid exprcoll = exprCollation(newexpr);
4743 :
4398 tgl 4744 576 : if (OidIsValid(exprcoll) && exprcoll != result_collid)
4745 : {
4382 bruce 4746 18 : CollateExpr *newnode = makeNode(CollateExpr);
4399 tgl 4747 ECB :
4399 tgl 4748 CBC 18 : newnode->arg = (Expr *) newexpr;
4398 tgl 4749 GIC 18 : newnode->collOid = result_collid;
4399 4750 18 : newnode->location = -1;
4751 :
4399 tgl 4752 CBC 18 : newexpr = (Node *) newnode;
4399 tgl 4753 ECB : }
4754 : }
4755 :
4756 : /*
5050 bruce 4757 : * Since there is now no trace of the function in the plan tree, we must
4758 : * explicitly record the plan's dependency on the function.
4759 : */
4236 tgl 4760 CBC 5557 : if (context->root)
4761 5496 : record_plan_function_dependency(context->root, funcid);
4762 :
7434 tgl 4763 ECB : /*
4764 : * Recursively try to simplify the modified expression. Here we must add
4765 : * the current function to the context list of active functions.
4766 : */
1362 tgl 4767 GIC 5557 : context->active_fns = lappend_oid(context->active_fns, funcid);
6876 4768 5557 : newexpr = eval_const_expressions_mutator(newexpr, context);
1362 4769 5556 : context->active_fns = list_delete_last(context->active_fns);
4770 :
7195 4771 5556 : error_context_stack = sqlerrcontext.previous;
4772 :
7434 4773 5556 : return (Expr *) newexpr;
7434 tgl 4774 ECB :
4775 : /* Here if func is not inlinable: release temp memory and return NULL */
7434 tgl 4776 CBC 1246 : fail:
4777 1246 : MemoryContextSwitchTo(oldcxt);
4778 1246 : MemoryContextDelete(mycxt);
7195 tgl 4779 GBC 1246 : error_context_stack = sqlerrcontext.previous;
4780 :
7434 tgl 4781 GIC 1246 : return NULL;
4782 : }
4783 :
4784 : /*
7434 tgl 4785 ECB : * Replace Param nodes by appropriate actual parameters
4786 : */
4787 : static Node *
7434 tgl 4788 CBC 5561 : substitute_actual_parameters(Node *expr, int nargs, List *args,
4789 : int *usecounts)
4790 : {
4791 : substitute_actual_parameters_context context;
4792 :
7188 bruce 4793 GIC 5561 : context.nargs = nargs;
7434 tgl 4794 5561 : context.args = args;
7434 tgl 4795 CBC 5561 : context.usecounts = usecounts;
4796 :
4797 5561 : return substitute_actual_parameters_mutator(expr, &context);
4798 : }
7434 tgl 4799 ECB :
4800 : static Node *
7434 tgl 4801 GIC 15820 : substitute_actual_parameters_mutator(Node *node,
4802 : substitute_actual_parameters_context *context)
4803 : {
4804 15820 : if (node == NULL)
4805 4450 : return NULL;
4806 11370 : if (IsA(node, Param))
7434 tgl 4807 ECB : {
7434 tgl 4808 GIC 2103 : Param *param = (Param *) node;
7434 tgl 4809 ECB :
6196 tgl 4810 GIC 2103 : if (param->paramkind != PARAM_EXTERN)
6196 tgl 4811 LBC 0 : elog(ERROR, "unexpected paramkind: %d", (int) param->paramkind);
7434 tgl 4812 GIC 2103 : if (param->paramid <= 0 || param->paramid > context->nargs)
7198 tgl 4813 LBC 0 : elog(ERROR, "invalid paramid: %d", param->paramid);
4814 :
7434 tgl 4815 ECB : /* Count usage of parameter */
7434 tgl 4816 CBC 2103 : context->usecounts[param->paramid - 1]++;
7434 tgl 4817 ECB :
4818 : /* Select the appropriate actual arg and replace the Param with it */
4819 : /* We don't need to copy at this time (it'll get done later) */
6888 neilc 4820 GIC 2103 : return list_nth(context->args, param->paramid - 1);
4821 : }
7434 tgl 4822 9267 : return expression_tree_mutator(node, substitute_actual_parameters_mutator,
4823 : (void *) context);
4824 : }
4825 :
4826 : /*
7195 tgl 4827 ECB : * error context callback to let us supply a call-stack traceback
4828 : */
4829 : static void
7195 tgl 4830 GIC 14 : sql_inline_error_callback(void *arg)
4831 : {
4769 4832 14 : inline_error_callback_arg *callback_arg = (inline_error_callback_arg *) arg;
4833 : int syntaxerrposition;
6958 tgl 4834 ECB :
4835 : /* If it's a syntax error, convert to internal syntax error report */
724 tgl 4836 CBC 14 : syntaxerrposition = geterrposition();
724 tgl 4837 GIC 14 : if (syntaxerrposition > 0)
6958 tgl 4838 ECB : {
724 tgl 4839 GIC 4 : errposition(0);
724 tgl 4840 CBC 4 : internalerrposition(syntaxerrposition);
724 tgl 4841 GIC 4 : internalerrquery(callback_arg->prosrc);
4842 : }
7195 tgl 4843 ECB :
4769 tgl 4844 CBC 14 : errcontext("SQL function \"%s\" during inlining", callback_arg->proname);
7195 4845 14 : }
7195 tgl 4846 ECB :
4847 : /*
7306 4848 : * evaluate_expr: pre-evaluate a constant expression
4849 : *
4850 : * We use the executor's routine ExecEvalExpr() to avoid duplication of
4851 : * code and ensure we get the same result as the executor would get.
4852 : */
4853 : Expr *
4398 tgl 4854 GIC 99722 : evaluate_expr(Expr *expr, Oid result_type, int32 result_typmod,
4398 tgl 4855 ECB : Oid result_collation)
4856 : {
4857 : EState *estate;
4858 : ExprState *exprstate;
4859 : MemoryContext oldcontext;
7306 4860 : Datum const_val;
4861 : bool const_is_null;
4862 : int16 resultTypLen;
4863 : bool resultTypByVal;
4864 :
4865 : /*
4866 : * To use the executor, we need an EState.
4867 : */
7306 tgl 4868 CBC 99722 : estate = CreateExecutorState();
4869 :
4870 : /* We can use the estate's working context to avoid memory leaks. */
4871 99722 : oldcontext = MemoryContextSwitchTo(estate->es_query_cxt);
7306 tgl 4872 ECB :
5203 4873 : /* Make sure any opfuncids are filled in. */
5203 tgl 4874 GIC 99722 : fix_opfuncids((Node *) expr);
5203 tgl 4875 ECB :
4876 : /*
3260 bruce 4877 : * Prepare expr for execution. (Note: we can't use ExecPrepareExpr
5203 tgl 4878 EUB : * because it'd result in recursively invoking eval_const_expressions.)
7306 tgl 4879 ECB : */
5203 tgl 4880 GBC 99722 : exprstate = ExecInitExpr(expr, NULL);
4881 :
4882 : /*
7306 tgl 4883 ECB : * And evaluate it.
4884 : *
4885 : * It is OK to use a default econtext because none of the ExecEvalExpr()
4886 : * code used in this situation will use econtext. That might seem
6347 bruce 4887 : * fortuitous, but it's not so unreasonable --- a constant expression does
4888 : * not depend on context, by definition, n'est ce pas?
7306 tgl 4889 : */
7306 tgl 4890 GIC 99713 : const_val = ExecEvalExprSwitchContext(exprstate,
4891 99713 : GetPerTupleExprContext(estate),
4892 : &const_is_null);
4893 :
4894 : /* Get info needed about result datatype */
4895 98193 : get_typlenbyval(result_type, &resultTypLen, &resultTypByVal);
4896 :
7306 tgl 4897 ECB : /* Get back to outer memory context */
7306 tgl 4898 GIC 98193 : MemoryContextSwitchTo(oldcontext);
7306 tgl 4899 ECB :
4900 : /*
4901 : * Must copy result out of sub-context used by expression eval.
4902 : *
5659 4903 : * Also, if it's varlena, forcibly detoast it. This protects us against
4904 : * storing TOAST pointers into plans that might outlive the referenced
4905 : * data. (makeConst would handle detoasting anyway, but it's worth a few
2635 4906 : * extra lines here so that we can do the copy and detoast in one step.)
5659 4907 : */
7306 tgl 4908 CBC 98193 : if (!const_is_null)
4909 : {
5659 tgl 4910 GIC 97521 : if (resultTypLen == -1)
5659 tgl 4911 CBC 37811 : const_val = PointerGetDatum(PG_DETOAST_DATUM_COPY(const_val));
5659 tgl 4912 ECB : else
5659 tgl 4913 GIC 59710 : const_val = datumCopy(const_val, resultTypByVal, resultTypLen);
4914 : }
4915 :
4916 : /* Release all the junk we just created */
7306 4917 98193 : FreeExecutorState(estate);
4918 :
4919 : /*
4920 : * Make the constant result node.
7306 tgl 4921 ECB : */
4398 tgl 4922 GIC 98193 : return (Expr *) makeConst(result_type, result_typmod, result_collation,
4923 : resultTypLen,
4924 : const_val, const_is_null,
4925 : resultTypByVal);
4926 : }
4927 :
4928 :
4929 : /*
4930 : * inline_set_returning_function
4931 : * Attempt to "inline" a set-returning function in the FROM clause.
4932 : *
4933 : * "rte" is an RTE_FUNCTION rangetable entry. If it represents a call of a
4934 : * set-returning SQL function that can safely be inlined, expand the function
5295 tgl 4935 ECB : * and return the substitute Query structure. Otherwise, return NULL.
4936 : *
4937 : * We assume that the RTE's expression has already been put through
1347 4938 : * eval_const_expressions(), which among other things will take care of
4939 : * default arguments and named-argument notation.
4940 : *
5500 4941 : * This has a good deal of similarity to inline_function(), but that's
4942 : * for the non-set-returning case, and there are enough differences to
4943 : * justify separate functions.
4944 : */
4945 : Query *
5295 tgl 4946 GIC 17937 : inline_set_returning_function(PlannerInfo *root, RangeTblEntry *rte)
5500 tgl 4947 ECB : {
4948 : RangeTblFunction *rtfunc;
4949 : FuncExpr *fexpr;
4950 : Oid func_oid;
4951 : HeapTuple func_tuple;
4952 : Form_pg_proc funcform;
4953 : char *src;
4954 : Datum tmp;
4955 : bool isNull;
4956 : MemoryContext oldcxt;
4957 : MemoryContext mycxt;
4769 4958 : inline_error_callback_arg callback_arg;
4959 : ErrorContextCallback sqlerrcontext;
4960 : SQLFunctionParseInfoPtr pinfo;
4961 : TypeFuncClass functypclass;
1187 4962 : TupleDesc rettupdesc;
4963 : List *raw_parsetree_list;
4964 : List *querytree_list;
5500 4965 : Query *querytree;
4966 :
5295 tgl 4967 GIC 17937 : Assert(rte->rtekind == RTE_FUNCTION);
4968 :
4969 : /*
4970 : * It doesn't make a lot of sense for a SQL SRF to refer to itself in its
4971 : * own FROM clause, since that must cause infinite recursion at runtime.
4972 : * It will cause this code to recurse too, so check for stack overflow.
4973 : * (There's no need to do more.)
4974 : */
5500 tgl 4975 CBC 17937 : check_stack_depth();
4976 :
3426 tgl 4977 ECB : /* Fail if the RTE has ORDINALITY - we don't implement that here. */
3541 stark 4978 CBC 17937 : if (rte->funcordinality)
3541 stark 4979 GIC 308 : return NULL;
3541 stark 4980 ECB :
4981 : /* Fail if RTE isn't a single, simple FuncExpr */
3426 tgl 4982 GIC 17629 : if (list_length(rte->functions) != 1)
5500 4983 36 : return NULL;
3426 tgl 4984 CBC 17593 : rtfunc = (RangeTblFunction *) linitial(rte->functions);
4985 :
3426 tgl 4986 GIC 17593 : if (!IsA(rtfunc->funcexpr, FuncExpr))
4987 195 : return NULL;
4988 17398 : fexpr = (FuncExpr *) rtfunc->funcexpr;
3541 stark 4989 ECB :
4931 tgl 4990 GIC 17398 : func_oid = fexpr->funcid;
4991 :
4992 : /*
4993 : * The function must be declared to return a set, else inlining would
4994 : * change the results if the contained SELECT didn't return exactly one
4995 : * row.
4996 : */
5500 4997 17398 : if (!fexpr->funcretset)
4998 1990 : return NULL;
4999 :
5000 : /*
5001 : * Refuse to inline if the arguments contain any volatile functions or
5002 : * sub-selects. Volatile functions are rejected because inlining may
5003 : * result in the arguments being evaluated multiple times, risking a
5004 : * change in behavior. Sub-selects are rejected partly for implementation
5005 : * reasons (pushing them down another level might change their behavior)
5006 : * and partly because they're likely to be expensive and so multiple
5007 : * evaluation would be bad.
5008 : */
5009 30764 : if (contain_volatile_functions((Node *) fexpr->args) ||
5010 15356 : contain_subplans((Node *) fexpr->args))
5011 175 : return NULL;
5012 :
5500 tgl 5013 ECB : /* Check permission to call function (fail later, if not) */
147 peter 5014 GNC 15233 : if (object_aclcheck(ProcedureRelationId, func_oid, GetUserId(), ACL_EXECUTE) != ACLCHECK_OK)
5500 tgl 5015 GIC 4 : return NULL;
5016 :
5017 : /* Check whether a plugin wants to hook function entry/exit */
4500 rhaas 5018 15229 : if (FmgrHookIsNeeded(func_oid))
4500 rhaas 5019 UIC 0 : return NULL;
5020 :
5021 : /*
5022 : * OK, let's take a look at the function's pg_proc entry.
5023 : */
4802 rhaas 5024 GIC 15229 : func_tuple = SearchSysCache1(PROCOID, ObjectIdGetDatum(func_oid));
5500 tgl 5025 15229 : if (!HeapTupleIsValid(func_tuple))
4931 tgl 5026 UIC 0 : elog(ERROR, "cache lookup failed for function %u", func_oid);
5500 tgl 5027 GIC 15229 : funcform = (Form_pg_proc) GETSTRUCT(func_tuple);
5028 :
5029 : /*
5030 : * Forget it if the function is not SQL-language or has other showstopper
5031 : * properties. In particular it mustn't be declared STRICT, since we
5032 : * couldn't enforce that. It also mustn't be VOLATILE, because that is
5033 : * supposed to cause it to be executed with its own snapshot, rather than
1850 tgl 5034 ECB : * sharing the snapshot of the calling query. We also disallow returning
5035 : * SETOF VOID, because inlining would result in exposing the actual result
5036 : * of the function's last SELECT, which should not happen in that case.
5037 : * (Rechecking prokind, proretset, and pronargs is just paranoia.)
5038 : */
5500 tgl 5039 GIC 15229 : if (funcform->prolang != SQLlanguageId ||
1850 5040 279 : funcform->prokind != PROKIND_FUNCTION ||
5500 5041 279 : funcform->proisstrict ||
5500 tgl 5042 CBC 249 : funcform->provolatile == PROVOLATILE_VOLATILE ||
1850 tgl 5043 GIC 75 : funcform->prorettype == VOIDOID ||
5500 5044 72 : funcform->prosecdef ||
5500 tgl 5045 CBC 72 : !funcform->proretset ||
1347 5046 72 : list_length(fexpr->args) != funcform->pronargs ||
1838 andrew 5047 GIC 72 : !heap_attisnull(func_tuple, Anum_pg_proc_proconfig, NULL))
5048 : {
5500 tgl 5049 CBC 15157 : ReleaseSysCache(func_tuple);
5050 15157 : return NULL;
5500 tgl 5051 ECB : }
5052 :
5053 : /*
5054 : * Make a temporary memory context, so that we don't leak all the stuff
5055 : * that parsing might create.
5056 : */
5500 tgl 5057 CBC 72 : mycxt = AllocSetContextCreate(CurrentMemoryContext,
5058 : "inline_set_returning_function",
5059 : ALLOCSET_DEFAULT_SIZES);
5500 tgl 5060 GIC 72 : oldcxt = MemoryContextSwitchTo(mycxt);
5061 :
5062 : /* Fetch the function body */
15 dgustafsson 5063 GNC 72 : tmp = SysCacheGetAttrNotNull(PROCOID, func_tuple, Anum_pg_proc_prosrc);
724 tgl 5064 GIC 72 : src = TextDatumGetCString(tmp);
5065 :
5066 : /*
5067 : * Setup error traceback support for ereport(). This is so that we can
5068 : * finger the function that bad information came from.
5069 : */
4769 5070 72 : callback_arg.proname = NameStr(funcform->proname);
724 tgl 5071 CBC 72 : callback_arg.prosrc = src;
4769 tgl 5072 ECB :
4769 tgl 5073 CBC 72 : sqlerrcontext.callback = sql_inline_error_callback;
4769 tgl 5074 GIC 72 : sqlerrcontext.arg = (void *) &callback_arg;
5075 72 : sqlerrcontext.previous = error_context_stack;
4769 tgl 5076 CBC 72 : error_context_stack = &sqlerrcontext;
4769 tgl 5077 ECB :
5078 : /* If we have prosqlbody, pay attention to that not prosrc */
732 peter 5079 GIC 72 : tmp = SysCacheGetAttr(PROCOID,
732 peter 5080 ECB : func_tuple,
724 tgl 5081 EUB : Anum_pg_proc_prosqlbody,
5082 : &isNull);
724 tgl 5083 GIC 72 : if (!isNull)
5084 : {
5085 : Node *n;
732 peter 5086 ECB :
732 peter 5087 CBC 6 : n = stringToNode(TextDatumGetCString(tmp));
732 peter 5088 GBC 6 : if (IsA(n, List))
732 peter 5089 CBC 6 : querytree_list = linitial_node(List, castNode(List, n));
5090 : else
732 peter 5091 UIC 0 : querytree_list = list_make1(n);
732 peter 5092 GIC 6 : if (list_length(querytree_list) != 1)
732 peter 5093 UIC 0 : goto fail;
732 peter 5094 GIC 6 : querytree = linitial(querytree_list);
5095 :
5096 : /* Acquire necessary locks, then apply rewriter. */
586 tgl 5097 6 : AcquireRewriteLocks(querytree, true, false);
732 peter 5098 6 : querytree_list = pg_rewrite_query(querytree);
5099 6 : if (list_length(querytree_list) != 1)
732 peter 5100 UIC 0 : goto fail;
732 peter 5101 CBC 6 : querytree = linitial(querytree_list);
732 peter 5102 ECB : }
5103 : else
5104 : {
697 tgl 5105 : /*
5106 : * Set up to handle parameters while parsing the function body. We
5107 : * can use the FuncExpr just created as the input for
5108 : * prepare_sql_fn_parse_info.
5109 : */
697 tgl 5110 GIC 66 : pinfo = prepare_sql_fn_parse_info(func_tuple,
697 tgl 5111 ECB : (Node *) fexpr,
5112 : fexpr->inputcollid);
5113 :
5114 : /*
5115 : * Parse, analyze, and rewrite (unlike inline_function(), we can't
5116 : * skip rewriting here). We can fail as soon as we find more than one
5117 : * query, though.
5118 : */
697 tgl 5119 CBC 66 : raw_parsetree_list = pg_parse_query(src);
697 tgl 5120 GIC 66 : if (list_length(raw_parsetree_list) != 1)
697 tgl 5121 UIC 0 : goto fail;
5500 tgl 5122 ECB :
401 peter 5123 GIC 66 : querytree_list = pg_analyze_and_rewrite_withcb(linitial(raw_parsetree_list),
5124 : src,
697 tgl 5125 ECB : (ParserSetupHook) sql_fn_parser_setup,
5126 : pinfo, NULL);
697 tgl 5127 GIC 66 : if (list_length(querytree_list) != 1)
697 tgl 5128 UIC 0 : goto fail;
697 tgl 5129 GIC 66 : querytree = linitial(querytree_list);
5130 : }
5131 :
732 peter 5132 ECB : /*
5133 : * Also resolve the actual function result tupdesc, if composite. If the
5134 : * function is just declared to return RECORD, dig the info out of the AS
5135 : * clause.
5136 : */
732 peter 5137 CBC 72 : functypclass = get_expr_result_type((Node *) fexpr, NULL, &rettupdesc);
5138 72 : if (functypclass == TYPEFUNC_RECORD)
732 peter 5139 GIC 12 : rettupdesc = BuildDescFromLists(rtfunc->funccolnames,
5140 : rtfunc->funccoltypes,
732 peter 5141 ECB : rtfunc->funccoltypmods,
5142 : rtfunc->funccolcollations);
5143 :
5144 : /*
4038 tgl 5145 : * The single command must be a plain SELECT.
5146 : */
5500 tgl 5147 GIC 72 : if (!IsA(querytree, Query) ||
2276 5148 72 : querytree->commandType != CMD_SELECT)
5500 tgl 5149 LBC 0 : goto fail;
5500 tgl 5150 ECB :
5151 : /*
5152 : * Make sure the function (still) returns what it's declared to. This
5500 tgl 5153 EUB : * will raise an error if wrong, but that's okay since the function would
3260 bruce 5154 ECB : * fail at runtime anyway. Note that check_sql_fn_retval will also insert
1187 tgl 5155 EUB : * coercions if needed to make the tlist expression(s) match the declared
1187 tgl 5156 ECB : * type of the function. We also ask it to insert dummy NULL columns for
5157 : * any dropped columns in rettupdesc, so that the elements of the modified
5158 : * tlist match up to the attribute numbers.
5500 5159 : *
5050 bruce 5160 : * If the function returns a composite type, don't inline unless the check
5161 : * shows it's returning a whole tuple result; otherwise what it's
1187 tgl 5162 EUB : * returning is a single composite column which is not what we need.
5500 tgl 5163 ECB : */
902 tgl 5164 GIC 72 : if (!check_sql_fn_retval(list_make1(querytree_list),
5165 : fexpr->funcresulttype, rettupdesc,
1187 5166 45 : true, NULL) &&
5167 45 : (functypclass == TYPEFUNC_COMPOSITE ||
5168 45 : functypclass == TYPEFUNC_COMPOSITE_DOMAIN ||
5169 : functypclass == TYPEFUNC_RECORD))
5500 tgl 5170 UIC 0 : goto fail; /* reject not-whole-tuple-result cases */
5171 :
4864 tgl 5172 ECB : /*
5173 : * check_sql_fn_retval might've inserted a projection step, but that's
5174 : * fine; just make sure we use the upper Query.
5175 : */
902 tgl 5176 GIC 69 : querytree = linitial_node(Query, querytree_list);
5177 :
5178 : /*
5179 : * Looks good --- substitute parameters into the query.
5180 : */
5500 tgl 5181 CBC 69 : querytree = substitute_actual_srf_parameters(querytree,
5182 69 : funcform->pronargs,
5500 tgl 5183 EUB : fexpr->args);
5184 :
5500 tgl 5185 ECB : /*
5186 : * Copy the modified query out of the temporary memory context, and clean
5187 : * up.
5188 : */
5500 tgl 5189 CBC 69 : MemoryContextSwitchTo(oldcxt);
5500 tgl 5190 EUB :
5500 tgl 5191 CBC 69 : querytree = copyObject(querytree);
5192 :
5500 tgl 5193 GIC 69 : MemoryContextDelete(mycxt);
5194 69 : error_context_stack = sqlerrcontext.previous;
5195 69 : ReleaseSysCache(func_tuple);
5196 :
5197 : /*
5198 : * We don't have to fix collations here because the upper query is already
4382 bruce 5199 ECB : * parsed, ie, the collations in the RTE are what count.
4399 tgl 5200 : */
5201 :
5202 : /*
5203 : * Since there is now no trace of the function in the plan tree, we must
5204 : * explicitly record the plan's dependency on the function.
5205 : */
4236 tgl 5206 GIC 69 : record_plan_function_dependency(root, func_oid);
5207 :
5500 5208 69 : return querytree;
5500 tgl 5209 ECB :
5210 : /* Here if func is not inlinable: release temp memory and return NULL */
5500 tgl 5211 UBC 0 : fail:
5500 tgl 5212 UIC 0 : MemoryContextSwitchTo(oldcxt);
5213 0 : MemoryContextDelete(mycxt);
5214 0 : error_context_stack = sqlerrcontext.previous;
5215 0 : ReleaseSysCache(func_tuple);
5216 :
5217 0 : return NULL;
5218 : }
5219 :
5220 : /*
5221 : * Replace Param nodes by appropriate actual parameters
5222 : *
5223 : * This is just enough different from substitute_actual_parameters()
5224 : * that it needs its own code.
5225 : */
5500 tgl 5226 ECB : static Query *
5500 tgl 5227 GIC 69 : substitute_actual_srf_parameters(Query *expr, int nargs, List *args)
5500 tgl 5228 ECB : {
5229 : substitute_actual_srf_parameters_context context;
5230 :
5500 tgl 5231 GIC 69 : context.nargs = nargs;
5500 tgl 5232 GBC 69 : context.args = args;
5500 tgl 5233 GIC 69 : context.sublevels_up = 1;
5234 :
5235 69 : return query_tree_mutator(expr,
5236 : substitute_actual_srf_parameters_mutator,
5237 : &context,
5500 tgl 5238 ECB : 0);
5239 : }
5240 :
5241 : static Node *
5500 tgl 5242 GIC 2409 : substitute_actual_srf_parameters_mutator(Node *node,
2118 tgl 5243 ECB : substitute_actual_srf_parameters_context *context)
5500 5244 : {
5245 : Node *result;
5246 :
5500 tgl 5247 GIC 2409 : if (node == NULL)
5248 1254 : return NULL;
5249 1155 : if (IsA(node, Query))
5250 : {
5500 tgl 5251 CBC 39 : context->sublevels_up++;
5500 tgl 5252 GIC 39 : result = (Node *) query_tree_mutator((Query *) node,
2118 tgl 5253 ECB : substitute_actual_srf_parameters_mutator,
5254 : (void *) context,
5500 5255 : 0);
5500 tgl 5256 CBC 39 : context->sublevels_up--;
5257 39 : return result;
5258 : }
5500 tgl 5259 GIC 1116 : if (IsA(node, Param))
5260 : {
5261 51 : Param *param = (Param *) node;
5262 :
5263 51 : if (param->paramkind == PARAM_EXTERN)
5264 : {
5265 51 : if (param->paramid <= 0 || param->paramid > context->nargs)
5500 tgl 5266 UIC 0 : elog(ERROR, "invalid paramid: %d", param->paramid);
5267 :
5500 tgl 5268 ECB : /*
5269 : * Since the parameter is being inserted into a subquery, we must
5050 bruce 5270 : * adjust levels.
5271 : */
5500 tgl 5272 GIC 51 : result = copyObject(list_nth(context->args, param->paramid - 1));
5500 tgl 5273 GBC 51 : IncrementVarSublevelsUp(result, context->sublevels_up, 0);
5274 51 : return result;
5500 tgl 5275 EUB : }
5276 : }
5500 tgl 5277 GBC 1065 : return expression_tree_mutator(node,
5278 : substitute_actual_srf_parameters_mutator,
5500 tgl 5279 EUB : (void *) context);
5280 : }
5281 :
5282 : /*
5283 : * pull_paramids
5284 : * Returns a Bitmapset containing the paramids of all Params in 'expr'.
5285 : */
5286 : Bitmapset *
501 drowley 5287 GIC 502 : pull_paramids(Expr *expr)
5288 : {
501 drowley 5289 CBC 502 : Bitmapset *result = NULL;
5290 :
501 drowley 5291 GIC 502 : (void) pull_paramids_walker((Node *) expr, &result);
5292 :
501 drowley 5293 CBC 502 : return result;
501 drowley 5294 ECB : }
5295 :
5296 : static bool
501 drowley 5297 CBC 1109 : pull_paramids_walker(Node *node, Bitmapset **context)
5298 : {
501 drowley 5299 GIC 1109 : if (node == NULL)
5300 9 : return false;
5301 1100 : if (IsA(node, Param))
5302 : {
332 tgl 5303 511 : Param *param = (Param *) node;
501 drowley 5304 ECB :
501 drowley 5305 GIC 511 : *context = bms_add_member(*context, param->paramid);
5306 511 : return false;
5307 : }
5308 589 : return expression_tree_walker(node, pull_paramids_walker,
501 drowley 5309 ECB : (void *) context);
5310 : }
|
