| 1 | /*------------------------------------------------------------------------- |
| 2 | * |
| 3 | * clauses.c |
| 4 | * routines to manipulate qualification clauses |
| 5 | * |
| 6 | * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group |
| 7 | * Portions Copyright (c) 1994, Regents of the University of California |
| 8 | * |
| 9 | * |
| 10 | * IDENTIFICATION |
| 11 | * src/backend/optimizer/util/clauses.c |
| 12 | * |
| 13 | * HISTORY |
| 14 | * AUTHOR DATE MAJOR EVENT |
| 15 | * Andrew Yu Nov 3, 1994 clause.c and clauses.c combined |
| 16 | * |
| 17 | *------------------------------------------------------------------------- |
| 18 | */ |
| 19 | |
| 20 | #include "postgres.h" |
| 21 | |
| 22 | #include "access/htup_details.h" |
| 23 | #include "catalog/pg_aggregate.h" |
| 24 | #include "catalog/pg_class.h" |
| 25 | #include "catalog/pg_language.h" |
| 26 | #include "catalog/pg_operator.h" |
| 27 | #include "catalog/pg_proc.h" |
| 28 | #include "catalog/pg_type.h" |
| 29 | #include "executor/executor.h" |
| 30 | #include "executor/functions.h" |
| 31 | #include "funcapi.h" |
| 32 | #include "miscadmin.h" |
| 33 | #include "nodes/makefuncs.h" |
| 34 | #include "nodes/nodeFuncs.h" |
| 35 | #include "nodes/supportnodes.h" |
| 36 | #include "optimizer/clauses.h" |
| 37 | #include "optimizer/cost.h" |
| 38 | #include "optimizer/optimizer.h" |
| 39 | #include "optimizer/plancat.h" |
| 40 | #include "optimizer/planmain.h" |
| 41 | #include "parser/analyze.h" |
| 42 | #include "parser/parse_agg.h" |
| 43 | #include "parser/parse_coerce.h" |
| 44 | #include "parser/parse_func.h" |
| 45 | #include "rewrite/rewriteManip.h" |
| 46 | #include "tcop/tcopprot.h" |
| 47 | #include "utils/acl.h" |
| 48 | #include "utils/builtins.h" |
| 49 | #include "utils/datum.h" |
| 50 | #include "utils/fmgroids.h" |
| 51 | #include "utils/lsyscache.h" |
| 52 | #include "utils/memutils.h" |
| 53 | #include "utils/syscache.h" |
| 54 | #include "utils/typcache.h" |
| 55 | |
| 56 | |
| 57 | typedef struct |
| 58 | { |
| 59 | PlannerInfo *root; |
| 60 | AggSplit aggsplit; |
| 61 | AggClauseCosts *costs; |
| 62 | } get_agg_clause_costs_context; |
| 63 | |
| 64 | typedef struct |
| 65 | { |
| 66 | ParamListInfo boundParams; |
| 67 | PlannerInfo *root; |
| 68 | List *active_fns; |
| 69 | Node *case_val; |
| 70 | bool estimate; |
| 71 | } eval_const_expressions_context; |
| 72 | |
| 73 | typedef struct |
| 74 | { |
| 75 | int nargs; |
| 76 | List *args; |
| 77 | int *usecounts; |
| 78 | } substitute_actual_parameters_context; |
| 79 | |
| 80 | typedef struct |
| 81 | { |
| 82 | int nargs; |
| 83 | List *args; |
| 84 | int sublevels_up; |
| 85 | } substitute_actual_srf_parameters_context; |
| 86 | |
| 87 | typedef struct |
| 88 | { |
| 89 | char *proname; |
| 90 | char *prosrc; |
| 91 | } inline_error_callback_arg; |
| 92 | |
| 93 | typedef struct |
| 94 | { |
| 95 | char max_hazard; /* worst proparallel hazard found so far */ |
| 96 | char max_interesting; /* worst proparallel hazard of interest */ |
| 97 | List *safe_param_ids; /* PARAM_EXEC Param IDs to treat as safe */ |
| 98 | } max_parallel_hazard_context; |
| 99 | |
| 100 | static bool contain_agg_clause_walker(Node *node, void *context); |
| 101 | static bool get_agg_clause_costs_walker(Node *node, |
| 102 | get_agg_clause_costs_context *context); |
| 103 | static bool find_window_functions_walker(Node *node, WindowFuncLists *lists); |
| 104 | static bool contain_subplans_walker(Node *node, void *context); |
| 105 | static bool contain_mutable_functions_walker(Node *node, void *context); |
| 106 | static bool contain_volatile_functions_walker(Node *node, void *context); |
| 107 | static bool contain_volatile_functions_not_nextval_walker(Node *node, void *context); |
| 108 | static bool max_parallel_hazard_walker(Node *node, |
| 109 | max_parallel_hazard_context *context); |
| 110 | static bool contain_nonstrict_functions_walker(Node *node, void *context); |
| 111 | static bool contain_context_dependent_node(Node *clause); |
| 112 | static bool contain_context_dependent_node_walker(Node *node, int *flags); |
| 113 | static bool contain_leaked_vars_walker(Node *node, void *context); |
| 114 | static Relids find_nonnullable_rels_walker(Node *node, bool top_level); |
| 115 | static List *find_nonnullable_vars_walker(Node *node, bool top_level); |
| 116 | static bool is_strict_saop(ScalarArrayOpExpr *expr, bool falseOK); |
| 117 | static Node *eval_const_expressions_mutator(Node *node, |
| 118 | eval_const_expressions_context *context); |
| 119 | static bool contain_non_const_walker(Node *node, void *context); |
| 120 | static bool ece_function_is_safe(Oid funcid, |
| 121 | eval_const_expressions_context *context); |
| 122 | static List *simplify_or_arguments(List *args, |
| 123 | eval_const_expressions_context *context, |
| 124 | bool *haveNull, bool *forceTrue); |
| 125 | static List *simplify_and_arguments(List *args, |
| 126 | eval_const_expressions_context *context, |
| 127 | bool *haveNull, bool *forceFalse); |
| 128 | static Node *simplify_boolean_equality(Oid opno, List *args); |
| 129 | static Expr *simplify_function(Oid funcid, |
| 130 | Oid result_type, int32 result_typmod, |
| 131 | Oid result_collid, Oid input_collid, List **args_p, |
| 132 | bool funcvariadic, bool process_args, bool allow_non_const, |
| 133 | eval_const_expressions_context *context); |
| 134 | static List *reorder_function_arguments(List *args, HeapTuple func_tuple); |
| 135 | static List *add_function_defaults(List *args, HeapTuple func_tuple); |
| 136 | static List *fetch_function_defaults(HeapTuple func_tuple); |
| 137 | static void recheck_cast_function_args(List *args, Oid result_type, |
| 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 tlist_matches_coltypelist(List *tlist, List *coltypelist); |
| 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 |
| 179 | contain_agg_clause(Node *clause) |
| 180 | { |
| 181 | return contain_agg_clause_walker(clause, NULL); |
| 182 | } |
| 183 | |
| 184 | static bool |
| 185 | contain_agg_clause_walker(Node *node, void *context) |
| 186 | { |
| 187 | if (node == NULL) |
| 188 | return false; |
| 189 | if (IsA(node, Aggref)) |
| 190 | { |
| 191 | Assert(((Aggref *) node)->agglevelsup == 0); |
| 192 | return true; /* abort the tree traversal and return true */ |
| 193 | } |
| 194 | if (IsA(node, GroupingFunc)) |
| 195 | { |
| 196 | Assert(((GroupingFunc *) node)->agglevelsup == 0); |
| 197 | return true; /* abort the tree traversal and return true */ |
| 198 | } |
| 199 | Assert(!IsA(node, SubLink)); |
| 200 | return expression_tree_walker(node, contain_agg_clause_walker, context); |
| 201 | } |
| 202 | |
| 203 | /* |
| 204 | * get_agg_clause_costs |
| 205 | * Recursively find the Aggref nodes in an expression tree, and |
| 206 | * accumulate cost information about them. |
| 207 | * |
| 208 | * 'aggsplit' tells us the expected partial-aggregation mode, which affects |
| 209 | * the cost estimates. |
| 210 | * |
| 211 | * NOTE that the counts/costs are ADDED to those already in *costs ... so |
| 212 | * the caller is responsible for zeroing the struct initially. |
| 213 | * |
| 214 | * We count the nodes, estimate their execution costs, and estimate the total |
| 215 | * space needed for their transition state values if all are evaluated in |
| 216 | * parallel (as would be done in a HashAgg plan). Also, we check whether |
| 217 | * partial aggregation is feasible. See AggClauseCosts for the exact set |
| 218 | * of statistics collected. |
| 219 | * |
| 220 | * In addition, we mark Aggref nodes with the correct aggtranstype, so |
| 221 | * that that doesn't need to be done repeatedly. (That makes this function's |
| 222 | * name a bit of a misnomer.) |
| 223 | * |
| 224 | * This does not descend into subqueries, and so should be used only after |
| 225 | * reduction of sublinks to subplans, or in contexts where it's known there |
| 226 | * are no subqueries. There mustn't be outer-aggregate references either. |
| 227 | */ |
| 228 | void |
| 229 | get_agg_clause_costs(PlannerInfo *root, Node *clause, AggSplit aggsplit, |
| 230 | AggClauseCosts *costs) |
| 231 | { |
| 232 | get_agg_clause_costs_context context; |
| 233 | |
| 234 | context.root = root; |
| 235 | context.aggsplit = aggsplit; |
| 236 | context.costs = costs; |
| 237 | (void) get_agg_clause_costs_walker(clause, &context); |
| 238 | } |
| 239 | |
| 240 | static bool |
| 241 | get_agg_clause_costs_walker(Node *node, get_agg_clause_costs_context *context) |
| 242 | { |
| 243 | if (node == NULL) |
| 244 | return false; |
| 245 | if (IsA(node, Aggref)) |
| 246 | { |
| 247 | Aggref *aggref = (Aggref *) node; |
| 248 | AggClauseCosts *costs = context->costs; |
| 249 | HeapTuple aggTuple; |
| 250 | Form_pg_aggregate aggform; |
| 251 | Oid aggtransfn; |
| 252 | Oid aggfinalfn; |
| 253 | Oid aggcombinefn; |
| 254 | Oid aggserialfn; |
| 255 | Oid aggdeserialfn; |
| 256 | Oid aggtranstype; |
| 257 | int32 aggtransspace; |
| 258 | QualCost argcosts; |
| 259 | |
| 260 | Assert(aggref->agglevelsup == 0); |
| 261 | |
| 262 | /* |
| 263 | * Fetch info about aggregate from pg_aggregate. Note it's correct to |
| 264 | * ignore the moving-aggregate variant, since what we're concerned |
| 265 | * with here is aggregates not window functions. |
| 266 | */ |
| 267 | aggTuple = SearchSysCache1(AGGFNOID, |
| 268 | ObjectIdGetDatum(aggref->aggfnoid)); |
| 269 | if (!HeapTupleIsValid(aggTuple)) |
| 270 | elog(ERROR, "cache lookup failed for aggregate %u" , |
| 271 | aggref->aggfnoid); |
| 272 | aggform = (Form_pg_aggregate) GETSTRUCT(aggTuple); |
| 273 | aggtransfn = aggform->aggtransfn; |
| 274 | aggfinalfn = aggform->aggfinalfn; |
| 275 | aggcombinefn = aggform->aggcombinefn; |
| 276 | aggserialfn = aggform->aggserialfn; |
| 277 | aggdeserialfn = aggform->aggdeserialfn; |
| 278 | aggtranstype = aggform->aggtranstype; |
| 279 | aggtransspace = aggform->aggtransspace; |
| 280 | ReleaseSysCache(aggTuple); |
| 281 | |
| 282 | /* |
| 283 | * Resolve the possibly-polymorphic aggregate transition type, unless |
| 284 | * already done in a previous pass over the expression. |
| 285 | */ |
| 286 | if (OidIsValid(aggref->aggtranstype)) |
| 287 | aggtranstype = aggref->aggtranstype; |
| 288 | else |
| 289 | { |
| 290 | Oid inputTypes[FUNC_MAX_ARGS]; |
| 291 | int numArguments; |
| 292 | |
| 293 | /* extract argument types (ignoring any ORDER BY expressions) */ |
| 294 | numArguments = get_aggregate_argtypes(aggref, inputTypes); |
| 295 | |
| 296 | /* resolve actual type of transition state, if polymorphic */ |
| 297 | aggtranstype = resolve_aggregate_transtype(aggref->aggfnoid, |
| 298 | aggtranstype, |
| 299 | inputTypes, |
| 300 | numArguments); |
| 301 | aggref->aggtranstype = aggtranstype; |
| 302 | } |
| 303 | |
| 304 | /* |
| 305 | * Count it, and check for cases requiring ordered input. Note that |
| 306 | * ordered-set aggs always have nonempty aggorder. Any ordered-input |
| 307 | * case also defeats partial aggregation. |
| 308 | */ |
| 309 | costs->numAggs++; |
| 310 | if (aggref->aggorder != NIL || aggref->aggdistinct != NIL) |
| 311 | { |
| 312 | costs->numOrderedAggs++; |
| 313 | costs->hasNonPartial = true; |
| 314 | } |
| 315 | |
| 316 | /* |
| 317 | * Check whether partial aggregation is feasible, unless we already |
| 318 | * found out that we can't do it. |
| 319 | */ |
| 320 | if (!costs->hasNonPartial) |
| 321 | { |
| 322 | /* |
| 323 | * If there is no combine function, then partial aggregation is |
| 324 | * not possible. |
| 325 | */ |
| 326 | if (!OidIsValid(aggcombinefn)) |
| 327 | costs->hasNonPartial = true; |
| 328 | |
| 329 | /* |
| 330 | * If we have any aggs with transtype INTERNAL then we must check |
| 331 | * whether they have serialization/deserialization functions; if |
| 332 | * not, we can't serialize partial-aggregation results. |
| 333 | */ |
| 334 | else if (aggtranstype == INTERNALOID && |
| 335 | (!OidIsValid(aggserialfn) || !OidIsValid(aggdeserialfn))) |
| 336 | costs->hasNonSerial = true; |
| 337 | } |
| 338 | |
| 339 | /* |
| 340 | * Add the appropriate component function execution costs to |
| 341 | * appropriate totals. |
| 342 | */ |
| 343 | if (DO_AGGSPLIT_COMBINE(context->aggsplit)) |
| 344 | { |
| 345 | /* charge for combining previously aggregated states */ |
| 346 | add_function_cost(context->root, aggcombinefn, NULL, |
| 347 | &costs->transCost); |
| 348 | } |
| 349 | else |
| 350 | add_function_cost(context->root, aggtransfn, NULL, |
| 351 | &costs->transCost); |
| 352 | if (DO_AGGSPLIT_DESERIALIZE(context->aggsplit) && |
| 353 | OidIsValid(aggdeserialfn)) |
| 354 | add_function_cost(context->root, aggdeserialfn, NULL, |
| 355 | &costs->transCost); |
| 356 | if (DO_AGGSPLIT_SERIALIZE(context->aggsplit) && |
| 357 | OidIsValid(aggserialfn)) |
| 358 | add_function_cost(context->root, aggserialfn, NULL, |
| 359 | &costs->finalCost); |
| 360 | if (!DO_AGGSPLIT_SKIPFINAL(context->aggsplit) && |
| 361 | OidIsValid(aggfinalfn)) |
| 362 | add_function_cost(context->root, aggfinalfn, NULL, |
| 363 | &costs->finalCost); |
| 364 | |
| 365 | /* |
| 366 | * These costs are incurred only by the initial aggregate node, so we |
| 367 | * mustn't include them again at upper levels. |
| 368 | */ |
| 369 | if (!DO_AGGSPLIT_COMBINE(context->aggsplit)) |
| 370 | { |
| 371 | /* add the input expressions' cost to per-input-row costs */ |
| 372 | cost_qual_eval_node(&argcosts, (Node *) aggref->args, context->root); |
| 373 | costs->transCost.startup += argcosts.startup; |
| 374 | costs->transCost.per_tuple += argcosts.per_tuple; |
| 375 | |
| 376 | /* |
| 377 | * Add any filter's cost to per-input-row costs. |
| 378 | * |
| 379 | * XXX Ideally we should reduce input expression costs according |
| 380 | * to filter selectivity, but it's not clear it's worth the |
| 381 | * trouble. |
| 382 | */ |
| 383 | if (aggref->aggfilter) |
| 384 | { |
| 385 | cost_qual_eval_node(&argcosts, (Node *) aggref->aggfilter, |
| 386 | context->root); |
| 387 | costs->transCost.startup += argcosts.startup; |
| 388 | costs->transCost.per_tuple += argcosts.per_tuple; |
| 389 | } |
| 390 | } |
| 391 | |
| 392 | /* |
| 393 | * If there are direct arguments, treat their evaluation cost like the |
| 394 | * cost of the finalfn. |
| 395 | */ |
| 396 | if (aggref->aggdirectargs) |
| 397 | { |
| 398 | cost_qual_eval_node(&argcosts, (Node *) aggref->aggdirectargs, |
| 399 | context->root); |
| 400 | costs->finalCost.startup += argcosts.startup; |
| 401 | costs->finalCost.per_tuple += argcosts.per_tuple; |
| 402 | } |
| 403 | |
| 404 | /* |
| 405 | * If the transition type is pass-by-value then it doesn't add |
| 406 | * anything to the required size of the hashtable. If it is |
| 407 | * pass-by-reference then we have to add the estimated size of the |
| 408 | * value itself, plus palloc overhead. |
| 409 | */ |
| 410 | if (!get_typbyval(aggtranstype)) |
| 411 | { |
| 412 | int32 avgwidth; |
| 413 | |
| 414 | /* Use average width if aggregate definition gave one */ |
| 415 | if (aggtransspace > 0) |
| 416 | avgwidth = aggtransspace; |
| 417 | else if (aggtransfn == F_ARRAY_APPEND) |
| 418 | { |
| 419 | /* |
| 420 | * If the transition function is array_append(), it'll use an |
| 421 | * expanded array as transvalue, which will occupy at least |
| 422 | * ALLOCSET_SMALL_INITSIZE and possibly more. Use that as the |
| 423 | * estimate for lack of a better idea. |
| 424 | */ |
| 425 | avgwidth = ALLOCSET_SMALL_INITSIZE; |
| 426 | } |
| 427 | else |
| 428 | { |
| 429 | /* |
| 430 | * If transition state is of same type as first aggregated |
| 431 | * input, assume it's the same typmod (same width) as well. |
| 432 | * This works for cases like MAX/MIN and is probably somewhat |
| 433 | * reasonable otherwise. |
| 434 | */ |
| 435 | int32 aggtranstypmod = -1; |
| 436 | |
| 437 | if (aggref->args) |
| 438 | { |
| 439 | TargetEntry *tle = (TargetEntry *) linitial(aggref->args); |
| 440 | |
| 441 | if (aggtranstype == exprType((Node *) tle->expr)) |
| 442 | aggtranstypmod = exprTypmod((Node *) tle->expr); |
| 443 | } |
| 444 | |
| 445 | avgwidth = get_typavgwidth(aggtranstype, aggtranstypmod); |
| 446 | } |
| 447 | |
| 448 | avgwidth = MAXALIGN(avgwidth); |
| 449 | costs->transitionSpace += avgwidth + 2 * sizeof(void *); |
| 450 | } |
| 451 | else if (aggtranstype == INTERNALOID) |
| 452 | { |
| 453 | /* |
| 454 | * INTERNAL transition type is a special case: although INTERNAL |
| 455 | * is pass-by-value, it's almost certainly being used as a pointer |
| 456 | * to some large data structure. The aggregate definition can |
| 457 | * provide an estimate of the size. If it doesn't, then we assume |
| 458 | * ALLOCSET_DEFAULT_INITSIZE, which is a good guess if the data is |
| 459 | * being kept in a private memory context, as is done by |
| 460 | * array_agg() for instance. |
| 461 | */ |
| 462 | if (aggtransspace > 0) |
| 463 | costs->transitionSpace += aggtransspace; |
| 464 | else |
| 465 | costs->transitionSpace += ALLOCSET_DEFAULT_INITSIZE; |
| 466 | } |
| 467 | |
| 468 | /* |
| 469 | * We assume that the parser checked that there are no aggregates (of |
| 470 | * this level anyway) in the aggregated arguments, direct arguments, |
| 471 | * or filter clause. Hence, we need not recurse into any of them. |
| 472 | */ |
| 473 | return false; |
| 474 | } |
| 475 | Assert(!IsA(node, SubLink)); |
| 476 | return expression_tree_walker(node, get_agg_clause_costs_walker, |
| 477 | (void *) context); |
| 478 | } |
| 479 | |
| 480 | |
| 481 | /***************************************************************************** |
| 482 | * Window-function clause manipulation |
| 483 | *****************************************************************************/ |
| 484 | |
| 485 | /* |
| 486 | * contain_window_function |
| 487 | * Recursively search for WindowFunc nodes within a clause. |
| 488 | * |
| 489 | * Since window functions don't have level fields, but are hard-wired to |
| 490 | * be associated with the current query level, this is just the same as |
| 491 | * rewriteManip.c's function. |
| 492 | */ |
| 493 | bool |
| 494 | contain_window_function(Node *clause) |
| 495 | { |
| 496 | return contain_windowfuncs(clause); |
| 497 | } |
| 498 | |
| 499 | /* |
| 500 | * find_window_functions |
| 501 | * Locate all the WindowFunc nodes in an expression tree, and organize |
| 502 | * them by winref ID number. |
| 503 | * |
| 504 | * Caller must provide an upper bound on the winref IDs expected in the tree. |
| 505 | */ |
| 506 | WindowFuncLists * |
| 507 | find_window_functions(Node *clause, Index maxWinRef) |
| 508 | { |
| 509 | WindowFuncLists *lists = palloc(sizeof(WindowFuncLists)); |
| 510 | |
| 511 | lists->numWindowFuncs = 0; |
| 512 | lists->maxWinRef = maxWinRef; |
| 513 | lists->windowFuncs = (List **) palloc0((maxWinRef + 1) * sizeof(List *)); |
| 514 | (void) find_window_functions_walker(clause, lists); |
| 515 | return lists; |
| 516 | } |
| 517 | |
| 518 | static bool |
| 519 | find_window_functions_walker(Node *node, WindowFuncLists *lists) |
| 520 | { |
| 521 | if (node == NULL) |
| 522 | return false; |
| 523 | if (IsA(node, WindowFunc)) |
| 524 | { |
| 525 | WindowFunc *wfunc = (WindowFunc *) node; |
| 526 | |
| 527 | /* winref is unsigned, so one-sided test is OK */ |
| 528 | if (wfunc->winref > lists->maxWinRef) |
| 529 | elog(ERROR, "WindowFunc contains out-of-range winref %u" , |
| 530 | wfunc->winref); |
| 531 | /* eliminate duplicates, so that we avoid repeated computation */ |
| 532 | if (!list_member(lists->windowFuncs[wfunc->winref], wfunc)) |
| 533 | { |
| 534 | lists->windowFuncs[wfunc->winref] = |
| 535 | lappend(lists->windowFuncs[wfunc->winref], wfunc); |
| 536 | lists->numWindowFuncs++; |
| 537 | } |
| 538 | |
| 539 | /* |
| 540 | * We assume that the parser checked that there are no window |
| 541 | * functions in the arguments or filter clause. Hence, we need not |
| 542 | * recurse into them. (If either the parser or the planner screws up |
| 543 | * on this point, the executor will still catch it; see ExecInitExpr.) |
| 544 | */ |
| 545 | return false; |
| 546 | } |
| 547 | Assert(!IsA(node, SubLink)); |
| 548 | return expression_tree_walker(node, find_window_functions_walker, |
| 549 | (void *) lists); |
| 550 | } |
| 551 | |
| 552 | |
| 553 | /***************************************************************************** |
| 554 | * Support for expressions returning sets |
| 555 | *****************************************************************************/ |
| 556 | |
| 557 | /* |
| 558 | * expression_returns_set_rows |
| 559 | * Estimate the number of rows returned by a set-returning expression. |
| 560 | * The result is 1 if it's not a set-returning expression. |
| 561 | * |
| 562 | * We should only examine the top-level function or operator; it used to be |
| 563 | * appropriate to recurse, but not anymore. (Even if there are more SRFs in |
| 564 | * the function's inputs, their multipliers are accounted for separately.) |
| 565 | * |
| 566 | * Note: keep this in sync with expression_returns_set() in nodes/nodeFuncs.c. |
| 567 | */ |
| 568 | double |
| 569 | expression_returns_set_rows(PlannerInfo *root, Node *clause) |
| 570 | { |
| 571 | if (clause == NULL) |
| 572 | return 1.0; |
| 573 | if (IsA(clause, FuncExpr)) |
| 574 | { |
| 575 | FuncExpr *expr = (FuncExpr *) clause; |
| 576 | |
| 577 | if (expr->funcretset) |
| 578 | return clamp_row_est(get_function_rows(root, expr->funcid, clause)); |
| 579 | } |
| 580 | if (IsA(clause, OpExpr)) |
| 581 | { |
| 582 | OpExpr *expr = (OpExpr *) clause; |
| 583 | |
| 584 | if (expr->opretset) |
| 585 | { |
| 586 | set_opfuncid(expr); |
| 587 | return clamp_row_est(get_function_rows(root, expr->opfuncid, clause)); |
| 588 | } |
| 589 | } |
| 590 | return 1.0; |
| 591 | } |
| 592 | |
| 593 | |
| 594 | /***************************************************************************** |
| 595 | * Subplan clause manipulation |
| 596 | *****************************************************************************/ |
| 597 | |
| 598 | /* |
| 599 | * contain_subplans |
| 600 | * Recursively search for subplan nodes within a clause. |
| 601 | * |
| 602 | * If we see a SubLink node, we will return true. This is only possible if |
| 603 | * the expression tree hasn't yet been transformed by subselect.c. We do not |
| 604 | * know whether the node will produce a true subplan or just an initplan, |
| 605 | * but we make the conservative assumption that it will be a subplan. |
| 606 | * |
| 607 | * Returns true if any subplan found. |
| 608 | */ |
| 609 | bool |
| 610 | contain_subplans(Node *clause) |
| 611 | { |
| 612 | return contain_subplans_walker(clause, NULL); |
| 613 | } |
| 614 | |
| 615 | static bool |
| 616 | contain_subplans_walker(Node *node, void *context) |
| 617 | { |
| 618 | if (node == NULL) |
| 619 | return false; |
| 620 | if (IsA(node, SubPlan) || |
| 621 | IsA(node, AlternativeSubPlan) || |
| 622 | IsA(node, SubLink)) |
| 623 | return true; /* abort the tree traversal and return true */ |
| 624 | return expression_tree_walker(node, contain_subplans_walker, context); |
| 625 | } |
| 626 | |
| 627 | |
| 628 | /***************************************************************************** |
| 629 | * Check clauses for mutable functions |
| 630 | *****************************************************************************/ |
| 631 | |
| 632 | /* |
| 633 | * contain_mutable_functions |
| 634 | * Recursively search for mutable functions within a clause. |
| 635 | * |
| 636 | * Returns true if any mutable function (or operator implemented by a |
| 637 | * mutable function) is found. This test is needed so that we don't |
| 638 | * mistakenly think that something like "WHERE random() < 0.5" can be treated |
| 639 | * as a constant qualification. |
| 640 | * |
| 641 | * We will recursively look into Query nodes (i.e., SubLink sub-selects) |
| 642 | * but not into SubPlans. See comments for contain_volatile_functions(). |
| 643 | */ |
| 644 | bool |
| 645 | contain_mutable_functions(Node *clause) |
| 646 | { |
| 647 | return contain_mutable_functions_walker(clause, NULL); |
| 648 | } |
| 649 | |
| 650 | static bool |
| 651 | contain_mutable_functions_checker(Oid func_id, void *context) |
| 652 | { |
| 653 | return (func_volatile(func_id) != PROVOLATILE_IMMUTABLE); |
| 654 | } |
| 655 | |
| 656 | static bool |
| 657 | contain_mutable_functions_walker(Node *node, void *context) |
| 658 | { |
| 659 | if (node == NULL) |
| 660 | return false; |
| 661 | /* Check for mutable functions in node itself */ |
| 662 | if (check_functions_in_node(node, contain_mutable_functions_checker, |
| 663 | context)) |
| 664 | return true; |
| 665 | |
| 666 | if (IsA(node, SQLValueFunction)) |
| 667 | { |
| 668 | /* all variants of SQLValueFunction are stable */ |
| 669 | return true; |
| 670 | } |
| 671 | |
| 672 | if (IsA(node, NextValueExpr)) |
| 673 | { |
| 674 | /* NextValueExpr is volatile */ |
| 675 | return true; |
| 676 | } |
| 677 | |
| 678 | /* |
| 679 | * It should be safe to treat MinMaxExpr as immutable, because it will |
| 680 | * depend on a non-cross-type btree comparison function, and those should |
| 681 | * always be immutable. Treating XmlExpr as immutable is more dubious, |
| 682 | * and treating CoerceToDomain as immutable is outright dangerous. But we |
| 683 | * have done so historically, and changing this would probably cause more |
| 684 | * problems than it would fix. In practice, if you have a non-immutable |
| 685 | * domain constraint you are in for pain anyhow. |
| 686 | */ |
| 687 | |
| 688 | /* Recurse to check arguments */ |
| 689 | if (IsA(node, Query)) |
| 690 | { |
| 691 | /* Recurse into subselects */ |
| 692 | return query_tree_walker((Query *) node, |
| 693 | contain_mutable_functions_walker, |
| 694 | context, 0); |
| 695 | } |
| 696 | return expression_tree_walker(node, contain_mutable_functions_walker, |
| 697 | context); |
| 698 | } |
| 699 | |
| 700 | |
| 701 | /***************************************************************************** |
| 702 | * Check clauses for volatile functions |
| 703 | *****************************************************************************/ |
| 704 | |
| 705 | /* |
| 706 | * contain_volatile_functions |
| 707 | * Recursively search for volatile functions within a clause. |
| 708 | * |
| 709 | * Returns true if any volatile function (or operator implemented by a |
| 710 | * volatile function) is found. This test prevents, for example, |
| 711 | * invalid conversions of volatile expressions into indexscan quals. |
| 712 | * |
| 713 | * We will recursively look into Query nodes (i.e., SubLink sub-selects) |
| 714 | * but not into SubPlans. This is a bit odd, but intentional. If we are |
| 715 | * looking at a SubLink, we are probably deciding whether a query tree |
| 716 | * transformation is safe, and a contained sub-select should affect that; |
| 717 | * for example, duplicating a sub-select containing a volatile function |
| 718 | * would be bad. However, once we've got to the stage of having SubPlans, |
| 719 | * subsequent planning need not consider volatility within those, since |
| 720 | * the executor won't change its evaluation rules for a SubPlan based on |
| 721 | * volatility. |
| 722 | */ |
| 723 | bool |
| 724 | contain_volatile_functions(Node *clause) |
| 725 | { |
| 726 | return contain_volatile_functions_walker(clause, NULL); |
| 727 | } |
| 728 | |
| 729 | static bool |
| 730 | contain_volatile_functions_checker(Oid func_id, void *context) |
| 731 | { |
| 732 | return (func_volatile(func_id) == PROVOLATILE_VOLATILE); |
| 733 | } |
| 734 | |
| 735 | static bool |
| 736 | contain_volatile_functions_walker(Node *node, void *context) |
| 737 | { |
| 738 | if (node == NULL) |
| 739 | return false; |
| 740 | /* Check for volatile functions in node itself */ |
| 741 | if (check_functions_in_node(node, contain_volatile_functions_checker, |
| 742 | context)) |
| 743 | return true; |
| 744 | |
| 745 | if (IsA(node, NextValueExpr)) |
| 746 | { |
| 747 | /* NextValueExpr is volatile */ |
| 748 | return true; |
| 749 | } |
| 750 | |
| 751 | /* |
| 752 | * See notes in contain_mutable_functions_walker about why we treat |
| 753 | * MinMaxExpr, XmlExpr, and CoerceToDomain as immutable, while |
| 754 | * SQLValueFunction is stable. Hence, none of them are of interest here. |
| 755 | */ |
| 756 | |
| 757 | /* Recurse to check arguments */ |
| 758 | if (IsA(node, Query)) |
| 759 | { |
| 760 | /* Recurse into subselects */ |
| 761 | return query_tree_walker((Query *) node, |
| 762 | contain_volatile_functions_walker, |
| 763 | context, 0); |
| 764 | } |
| 765 | return expression_tree_walker(node, contain_volatile_functions_walker, |
| 766 | context); |
| 767 | } |
| 768 | |
| 769 | /* |
| 770 | * Special purpose version of contain_volatile_functions() for use in COPY: |
| 771 | * ignore nextval(), but treat all other functions normally. |
| 772 | */ |
| 773 | bool |
| 774 | contain_volatile_functions_not_nextval(Node *clause) |
| 775 | { |
| 776 | return contain_volatile_functions_not_nextval_walker(clause, NULL); |
| 777 | } |
| 778 | |
| 779 | static bool |
| 780 | contain_volatile_functions_not_nextval_checker(Oid func_id, void *context) |
| 781 | { |
| 782 | return (func_id != F_NEXTVAL_OID && |
| 783 | func_volatile(func_id) == PROVOLATILE_VOLATILE); |
| 784 | } |
| 785 | |
| 786 | static bool |
| 787 | contain_volatile_functions_not_nextval_walker(Node *node, void *context) |
| 788 | { |
| 789 | if (node == NULL) |
| 790 | return false; |
| 791 | /* Check for volatile functions in node itself */ |
| 792 | if (check_functions_in_node(node, |
| 793 | contain_volatile_functions_not_nextval_checker, |
| 794 | context)) |
| 795 | return true; |
| 796 | |
| 797 | /* |
| 798 | * See notes in contain_mutable_functions_walker about why we treat |
| 799 | * MinMaxExpr, XmlExpr, and CoerceToDomain as immutable, while |
| 800 | * SQLValueFunction is stable. Hence, none of them are of interest here. |
| 801 | * Also, since we're intentionally ignoring nextval(), presumably we |
| 802 | * should ignore NextValueExpr. |
| 803 | */ |
| 804 | |
| 805 | /* Recurse to check arguments */ |
| 806 | if (IsA(node, Query)) |
| 807 | { |
| 808 | /* Recurse into subselects */ |
| 809 | return query_tree_walker((Query *) node, |
| 810 | contain_volatile_functions_not_nextval_walker, |
| 811 | context, 0); |
| 812 | } |
| 813 | return expression_tree_walker(node, |
| 814 | contain_volatile_functions_not_nextval_walker, |
| 815 | context); |
| 816 | } |
| 817 | |
| 818 | |
| 819 | /***************************************************************************** |
| 820 | * Check queries for parallel unsafe and/or restricted constructs |
| 821 | *****************************************************************************/ |
| 822 | |
| 823 | /* |
| 824 | * max_parallel_hazard |
| 825 | * Find the worst parallel-hazard level in the given query |
| 826 | * |
| 827 | * Returns the worst function hazard property (the earliest in this list: |
| 828 | * PROPARALLEL_UNSAFE, PROPARALLEL_RESTRICTED, PROPARALLEL_SAFE) that can |
| 829 | * be found in the given parsetree. We use this to find out whether the query |
| 830 | * can be parallelized at all. The caller will also save the result in |
| 831 | * PlannerGlobal so as to short-circuit checks of portions of the querytree |
| 832 | * later, in the common case where everything is SAFE. |
| 833 | */ |
| 834 | char |
| 835 | max_parallel_hazard(Query *parse) |
| 836 | { |
| 837 | max_parallel_hazard_context context; |
| 838 | |
| 839 | context.max_hazard = PROPARALLEL_SAFE; |
| 840 | context.max_interesting = PROPARALLEL_UNSAFE; |
| 841 | context.safe_param_ids = NIL; |
| 842 | (void) max_parallel_hazard_walker((Node *) parse, &context); |
| 843 | return context.max_hazard; |
| 844 | } |
| 845 | |
| 846 | /* |
| 847 | * is_parallel_safe |
| 848 | * Detect whether the given expr contains only parallel-safe functions |
| 849 | * |
| 850 | * root->glob->maxParallelHazard must previously have been set to the |
| 851 | * result of max_parallel_hazard() on the whole query. |
| 852 | */ |
| 853 | bool |
| 854 | is_parallel_safe(PlannerInfo *root, Node *node) |
| 855 | { |
| 856 | max_parallel_hazard_context context; |
| 857 | PlannerInfo *proot; |
| 858 | ListCell *l; |
| 859 | |
| 860 | /* |
| 861 | * Even if the original querytree contained nothing unsafe, we need to |
| 862 | * search the expression if we have generated any PARAM_EXEC Params while |
| 863 | * planning, because those are parallel-restricted and there might be one |
| 864 | * in this expression. But otherwise we don't need to look. |
| 865 | */ |
| 866 | if (root->glob->maxParallelHazard == PROPARALLEL_SAFE && |
| 867 | root->glob->paramExecTypes == NIL) |
| 868 | return true; |
| 869 | /* Else use max_parallel_hazard's search logic, but stop on RESTRICTED */ |
| 870 | context.max_hazard = PROPARALLEL_SAFE; |
| 871 | context.max_interesting = PROPARALLEL_RESTRICTED; |
| 872 | context.safe_param_ids = NIL; |
| 873 | |
| 874 | /* |
| 875 | * The params that refer to the same or parent query level are considered |
| 876 | * parallel-safe. The idea is that we compute such params at Gather or |
| 877 | * Gather Merge node and pass their value to workers. |
| 878 | */ |
| 879 | for (proot = root; proot != NULL; proot = proot->parent_root) |
| 880 | { |
| 881 | foreach(l, proot->init_plans) |
| 882 | { |
| 883 | SubPlan *initsubplan = (SubPlan *) lfirst(l); |
| 884 | ListCell *l2; |
| 885 | |
| 886 | foreach(l2, initsubplan->setParam) |
| 887 | context.safe_param_ids = lcons_int(lfirst_int(l2), |
| 888 | context.safe_param_ids); |
| 889 | } |
| 890 | } |
| 891 | |
| 892 | return !max_parallel_hazard_walker(node, &context); |
| 893 | } |
| 894 | |
| 895 | /* core logic for all parallel-hazard checks */ |
| 896 | static bool |
| 897 | max_parallel_hazard_test(char proparallel, max_parallel_hazard_context *context) |
| 898 | { |
| 899 | switch (proparallel) |
| 900 | { |
| 901 | case PROPARALLEL_SAFE: |
| 902 | /* nothing to see here, move along */ |
| 903 | break; |
| 904 | case PROPARALLEL_RESTRICTED: |
| 905 | /* increase max_hazard to RESTRICTED */ |
| 906 | Assert(context->max_hazard != PROPARALLEL_UNSAFE); |
| 907 | context->max_hazard = proparallel; |
| 908 | /* done if we are not expecting any unsafe functions */ |
| 909 | if (context->max_interesting == proparallel) |
| 910 | return true; |
| 911 | break; |
| 912 | case PROPARALLEL_UNSAFE: |
| 913 | context->max_hazard = proparallel; |
| 914 | /* we're always done at the first unsafe construct */ |
| 915 | return true; |
| 916 | default: |
| 917 | elog(ERROR, "unrecognized proparallel value \"%c\"" , proparallel); |
| 918 | break; |
| 919 | } |
| 920 | return false; |
| 921 | } |
| 922 | |
| 923 | /* check_functions_in_node callback */ |
| 924 | static bool |
| 925 | max_parallel_hazard_checker(Oid func_id, void *context) |
| 926 | { |
| 927 | return max_parallel_hazard_test(func_parallel(func_id), |
| 928 | (max_parallel_hazard_context *) context); |
| 929 | } |
| 930 | |
| 931 | static bool |
| 932 | max_parallel_hazard_walker(Node *node, max_parallel_hazard_context *context) |
| 933 | { |
| 934 | if (node == NULL) |
| 935 | return false; |
| 936 | |
| 937 | /* Check for hazardous functions in node itself */ |
| 938 | if (check_functions_in_node(node, max_parallel_hazard_checker, |
| 939 | context)) |
| 940 | return true; |
| 941 | |
| 942 | /* |
| 943 | * It should be OK to treat MinMaxExpr as parallel-safe, since btree |
| 944 | * opclass support functions are generally parallel-safe. XmlExpr is a |
| 945 | * bit more dubious but we can probably get away with it. We err on the |
| 946 | * side of caution by treating CoerceToDomain as parallel-restricted. |
| 947 | * (Note: in principle that's wrong because a domain constraint could |
| 948 | * contain a parallel-unsafe function; but useful constraints probably |
| 949 | * never would have such, and assuming they do would cripple use of |
| 950 | * parallel query in the presence of domain types.) SQLValueFunction |
| 951 | * should be safe in all cases. NextValueExpr is parallel-unsafe. |
| 952 | */ |
| 953 | if (IsA(node, CoerceToDomain)) |
| 954 | { |
| 955 | if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context)) |
| 956 | return true; |
| 957 | } |
| 958 | |
| 959 | else if (IsA(node, NextValueExpr)) |
| 960 | { |
| 961 | if (max_parallel_hazard_test(PROPARALLEL_UNSAFE, context)) |
| 962 | return true; |
| 963 | } |
| 964 | |
| 965 | /* |
| 966 | * Treat window functions as parallel-restricted because we aren't sure |
| 967 | * whether the input row ordering is fully deterministic, and the output |
| 968 | * of window functions might vary across workers if not. (In some cases, |
| 969 | * like where the window frame orders by a primary key, we could relax |
| 970 | * this restriction. But it doesn't currently seem worth expending extra |
| 971 | * effort to do so.) |
| 972 | */ |
| 973 | else if (IsA(node, WindowFunc)) |
| 974 | { |
| 975 | if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context)) |
| 976 | return true; |
| 977 | } |
| 978 | |
| 979 | /* |
| 980 | * As a notational convenience for callers, look through RestrictInfo. |
| 981 | */ |
| 982 | else if (IsA(node, RestrictInfo)) |
| 983 | { |
| 984 | RestrictInfo *rinfo = (RestrictInfo *) node; |
| 985 | |
| 986 | return max_parallel_hazard_walker((Node *) rinfo->clause, context); |
| 987 | } |
| 988 | |
| 989 | /* |
| 990 | * Really we should not see SubLink during a max_interesting == restricted |
| 991 | * scan, but if we do, return true. |
| 992 | */ |
| 993 | else if (IsA(node, SubLink)) |
| 994 | { |
| 995 | if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context)) |
| 996 | return true; |
| 997 | } |
| 998 | |
| 999 | /* |
| 1000 | * Only parallel-safe SubPlans can be sent to workers. Within the |
| 1001 | * testexpr of the SubPlan, Params representing the output columns of the |
| 1002 | * subplan can be treated as parallel-safe, so temporarily add their IDs |
| 1003 | * to the safe_param_ids list while examining the testexpr. |
| 1004 | */ |
| 1005 | else if (IsA(node, SubPlan)) |
| 1006 | { |
| 1007 | SubPlan *subplan = (SubPlan *) node; |
| 1008 | List *save_safe_param_ids; |
| 1009 | |
| 1010 | if (!subplan->parallel_safe && |
| 1011 | max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context)) |
| 1012 | return true; |
| 1013 | save_safe_param_ids = context->safe_param_ids; |
| 1014 | context->safe_param_ids = list_concat(list_copy(subplan->paramIds), |
| 1015 | context->safe_param_ids); |
| 1016 | if (max_parallel_hazard_walker(subplan->testexpr, context)) |
| 1017 | return true; /* no need to restore safe_param_ids */ |
| 1018 | context->safe_param_ids = save_safe_param_ids; |
| 1019 | /* we must also check args, but no special Param treatment there */ |
| 1020 | if (max_parallel_hazard_walker((Node *) subplan->args, context)) |
| 1021 | return true; |
| 1022 | /* don't want to recurse normally, so we're done */ |
| 1023 | return false; |
| 1024 | } |
| 1025 | |
| 1026 | /* |
| 1027 | * We can't pass Params to workers at the moment either, so they are also |
| 1028 | * parallel-restricted, unless they are PARAM_EXTERN Params or are |
| 1029 | * PARAM_EXEC Params listed in safe_param_ids, meaning they could be |
| 1030 | * either generated within the worker or can be computed in master and |
| 1031 | * then their value can be passed to the worker. |
| 1032 | */ |
| 1033 | else if (IsA(node, Param)) |
| 1034 | { |
| 1035 | Param *param = (Param *) node; |
| 1036 | |
| 1037 | if (param->paramkind == PARAM_EXTERN) |
| 1038 | return false; |
| 1039 | |
| 1040 | if (param->paramkind != PARAM_EXEC || |
| 1041 | !list_member_int(context->safe_param_ids, param->paramid)) |
| 1042 | { |
| 1043 | if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context)) |
| 1044 | return true; |
| 1045 | } |
| 1046 | return false; /* nothing to recurse to */ |
| 1047 | } |
| 1048 | |
| 1049 | /* |
| 1050 | * When we're first invoked on a completely unplanned tree, we must |
| 1051 | * recurse into subqueries so to as to locate parallel-unsafe constructs |
| 1052 | * anywhere in the tree. |
| 1053 | */ |
| 1054 | else if (IsA(node, Query)) |
| 1055 | { |
| 1056 | Query *query = (Query *) node; |
| 1057 | |
| 1058 | /* SELECT FOR UPDATE/SHARE must be treated as unsafe */ |
| 1059 | if (query->rowMarks != NULL) |
| 1060 | { |
| 1061 | context->max_hazard = PROPARALLEL_UNSAFE; |
| 1062 | return true; |
| 1063 | } |
| 1064 | |
| 1065 | /* Recurse into subselects */ |
| 1066 | return query_tree_walker(query, |
| 1067 | max_parallel_hazard_walker, |
| 1068 | context, 0); |
| 1069 | } |
| 1070 | |
| 1071 | /* Recurse to check arguments */ |
| 1072 | return expression_tree_walker(node, |
| 1073 | max_parallel_hazard_walker, |
| 1074 | context); |
| 1075 | } |
| 1076 | |
| 1077 | |
| 1078 | /***************************************************************************** |
| 1079 | * Check clauses for nonstrict functions |
| 1080 | *****************************************************************************/ |
| 1081 | |
| 1082 | /* |
| 1083 | * contain_nonstrict_functions |
| 1084 | * Recursively search for nonstrict functions within a clause. |
| 1085 | * |
| 1086 | * Returns true if any nonstrict construct is found --- ie, anything that |
| 1087 | * could produce non-NULL output with a NULL input. |
| 1088 | * |
| 1089 | * The idea here is that the caller has verified that the expression contains |
| 1090 | * one or more Var or Param nodes (as appropriate for the caller's need), and |
| 1091 | * now wishes to prove that the expression result will be NULL if any of these |
| 1092 | * inputs is NULL. If we return false, then the proof succeeded. |
| 1093 | */ |
| 1094 | bool |
| 1095 | contain_nonstrict_functions(Node *clause) |
| 1096 | { |
| 1097 | return contain_nonstrict_functions_walker(clause, NULL); |
| 1098 | } |
| 1099 | |
| 1100 | static bool |
| 1101 | contain_nonstrict_functions_checker(Oid func_id, void *context) |
| 1102 | { |
| 1103 | return !func_strict(func_id); |
| 1104 | } |
| 1105 | |
| 1106 | static bool |
| 1107 | contain_nonstrict_functions_walker(Node *node, void *context) |
| 1108 | { |
| 1109 | if (node == NULL) |
| 1110 | return false; |
| 1111 | if (IsA(node, Aggref)) |
| 1112 | { |
| 1113 | /* an aggregate could return non-null with null input */ |
| 1114 | return true; |
| 1115 | } |
| 1116 | if (IsA(node, GroupingFunc)) |
| 1117 | { |
| 1118 | /* |
| 1119 | * A GroupingFunc doesn't evaluate its arguments, and therefore must |
| 1120 | * be treated as nonstrict. |
| 1121 | */ |
| 1122 | return true; |
| 1123 | } |
| 1124 | if (IsA(node, WindowFunc)) |
| 1125 | { |
| 1126 | /* a window function could return non-null with null input */ |
| 1127 | return true; |
| 1128 | } |
| 1129 | if (IsA(node, SubscriptingRef)) |
| 1130 | { |
| 1131 | /* |
| 1132 | * subscripting assignment is nonstrict, but subscripting itself is |
| 1133 | * strict |
| 1134 | */ |
| 1135 | if (((SubscriptingRef *) node)->refassgnexpr != NULL) |
| 1136 | return true; |
| 1137 | |
| 1138 | /* else fall through to check args */ |
| 1139 | } |
| 1140 | if (IsA(node, DistinctExpr)) |
| 1141 | { |
| 1142 | /* IS DISTINCT FROM is inherently non-strict */ |
| 1143 | return true; |
| 1144 | } |
| 1145 | if (IsA(node, NullIfExpr)) |
| 1146 | { |
| 1147 | /* NULLIF is inherently non-strict */ |
| 1148 | return true; |
| 1149 | } |
| 1150 | if (IsA(node, BoolExpr)) |
| 1151 | { |
| 1152 | BoolExpr *expr = (BoolExpr *) node; |
| 1153 | |
| 1154 | switch (expr->boolop) |
| 1155 | { |
| 1156 | case AND_EXPR: |
| 1157 | case OR_EXPR: |
| 1158 | /* AND, OR are inherently non-strict */ |
| 1159 | return true; |
| 1160 | default: |
| 1161 | break; |
| 1162 | } |
| 1163 | } |
| 1164 | if (IsA(node, SubLink)) |
| 1165 | { |
| 1166 | /* In some cases a sublink might be strict, but in general not */ |
| 1167 | return true; |
| 1168 | } |
| 1169 | if (IsA(node, SubPlan)) |
| 1170 | return true; |
| 1171 | if (IsA(node, AlternativeSubPlan)) |
| 1172 | return true; |
| 1173 | if (IsA(node, FieldStore)) |
| 1174 | return true; |
| 1175 | if (IsA(node, CoerceViaIO)) |
| 1176 | { |
| 1177 | /* |
| 1178 | * CoerceViaIO is strict regardless of whether the I/O functions are, |
| 1179 | * so just go look at its argument; asking check_functions_in_node is |
| 1180 | * useless expense and could deliver the wrong answer. |
| 1181 | */ |
| 1182 | return contain_nonstrict_functions_walker((Node *) ((CoerceViaIO *) node)->arg, |
| 1183 | context); |
| 1184 | } |
| 1185 | if (IsA(node, ArrayCoerceExpr)) |
| 1186 | { |
| 1187 | /* |
| 1188 | * ArrayCoerceExpr is strict at the array level, regardless of what |
| 1189 | * the per-element expression is; so we should ignore elemexpr and |
| 1190 | * recurse only into the arg. |
| 1191 | */ |
| 1192 | return contain_nonstrict_functions_walker((Node *) ((ArrayCoerceExpr *) node)->arg, |
| 1193 | context); |
| 1194 | } |
| 1195 | if (IsA(node, CaseExpr)) |
| 1196 | return true; |
| 1197 | if (IsA(node, ArrayExpr)) |
| 1198 | return true; |
| 1199 | if (IsA(node, RowExpr)) |
| 1200 | return true; |
| 1201 | if (IsA(node, RowCompareExpr)) |
| 1202 | return true; |
| 1203 | if (IsA(node, CoalesceExpr)) |
| 1204 | return true; |
| 1205 | if (IsA(node, MinMaxExpr)) |
| 1206 | return true; |
| 1207 | if (IsA(node, XmlExpr)) |
| 1208 | return true; |
| 1209 | if (IsA(node, NullTest)) |
| 1210 | return true; |
| 1211 | if (IsA(node, BooleanTest)) |
| 1212 | return true; |
| 1213 | |
| 1214 | /* Check other function-containing nodes */ |
| 1215 | if (check_functions_in_node(node, contain_nonstrict_functions_checker, |
| 1216 | context)) |
| 1217 | return true; |
| 1218 | |
| 1219 | return expression_tree_walker(node, contain_nonstrict_functions_walker, |
| 1220 | context); |
| 1221 | } |
| 1222 | |
| 1223 | /***************************************************************************** |
| 1224 | * Check clauses for context-dependent nodes |
| 1225 | *****************************************************************************/ |
| 1226 | |
| 1227 | /* |
| 1228 | * contain_context_dependent_node |
| 1229 | * Recursively search for context-dependent nodes within a clause. |
| 1230 | * |
| 1231 | * CaseTestExpr nodes must appear directly within the corresponding CaseExpr, |
| 1232 | * not nested within another one, or they'll see the wrong test value. If one |
| 1233 | * appears "bare" in the arguments of a SQL function, then we can't inline the |
| 1234 | * SQL function for fear of creating such a situation. The same applies for |
| 1235 | * CaseTestExpr used within the elemexpr of an ArrayCoerceExpr. |
| 1236 | * |
| 1237 | * CoerceToDomainValue would have the same issue if domain CHECK expressions |
| 1238 | * could get inlined into larger expressions, but presently that's impossible. |
| 1239 | * Still, it might be allowed in future, or other node types with similar |
| 1240 | * issues might get invented. So give this function a generic name, and set |
| 1241 | * up the recursion state to allow multiple flag bits. |
| 1242 | */ |
| 1243 | static bool |
| 1244 | contain_context_dependent_node(Node *clause) |
| 1245 | { |
| 1246 | int flags = 0; |
| 1247 | |
| 1248 | return contain_context_dependent_node_walker(clause, &flags); |
| 1249 | } |
| 1250 | |
| 1251 | #define CCDN_CASETESTEXPR_OK 0x0001 /* CaseTestExpr okay here? */ |
| 1252 | |
| 1253 | static bool |
| 1254 | contain_context_dependent_node_walker(Node *node, int *flags) |
| 1255 | { |
| 1256 | if (node == NULL) |
| 1257 | return false; |
| 1258 | if (IsA(node, CaseTestExpr)) |
| 1259 | return !(*flags & CCDN_CASETESTEXPR_OK); |
| 1260 | else if (IsA(node, CaseExpr)) |
| 1261 | { |
| 1262 | CaseExpr *caseexpr = (CaseExpr *) node; |
| 1263 | |
| 1264 | /* |
| 1265 | * If this CASE doesn't have a test expression, then it doesn't create |
| 1266 | * a context in which CaseTestExprs should appear, so just fall |
| 1267 | * through and treat it as a generic expression node. |
| 1268 | */ |
| 1269 | if (caseexpr->arg) |
| 1270 | { |
| 1271 | int save_flags = *flags; |
| 1272 | bool res; |
| 1273 | |
| 1274 | /* |
| 1275 | * Note: in principle, we could distinguish the various sub-parts |
| 1276 | * of a CASE construct and set the flag bit only for some of them, |
| 1277 | * since we are only expecting CaseTestExprs to appear in the |
| 1278 | * "expr" subtree of the CaseWhen nodes. But it doesn't really |
| 1279 | * seem worth any extra code. If there are any bare CaseTestExprs |
| 1280 | * elsewhere in the CASE, something's wrong already. |
| 1281 | */ |
| 1282 | *flags |= CCDN_CASETESTEXPR_OK; |
| 1283 | res = expression_tree_walker(node, |
| 1284 | contain_context_dependent_node_walker, |
| 1285 | (void *) flags); |
| 1286 | *flags = save_flags; |
| 1287 | return res; |
| 1288 | } |
| 1289 | } |
| 1290 | else if (IsA(node, ArrayCoerceExpr)) |
| 1291 | { |
| 1292 | ArrayCoerceExpr *ac = (ArrayCoerceExpr *) node; |
| 1293 | int save_flags; |
| 1294 | bool res; |
| 1295 | |
| 1296 | /* Check the array expression */ |
| 1297 | if (contain_context_dependent_node_walker((Node *) ac->arg, flags)) |
| 1298 | return true; |
| 1299 | |
| 1300 | /* Check the elemexpr, which is allowed to contain CaseTestExpr */ |
| 1301 | save_flags = *flags; |
| 1302 | *flags |= CCDN_CASETESTEXPR_OK; |
| 1303 | res = contain_context_dependent_node_walker((Node *) ac->elemexpr, |
| 1304 | flags); |
| 1305 | *flags = save_flags; |
| 1306 | return res; |
| 1307 | } |
| 1308 | return expression_tree_walker(node, contain_context_dependent_node_walker, |
| 1309 | (void *) flags); |
| 1310 | } |
| 1311 | |
| 1312 | /***************************************************************************** |
| 1313 | * Check clauses for Vars passed to non-leakproof functions |
| 1314 | *****************************************************************************/ |
| 1315 | |
| 1316 | /* |
| 1317 | * contain_leaked_vars |
| 1318 | * Recursively scan a clause to discover whether it contains any Var |
| 1319 | * nodes (of the current query level) that are passed as arguments to |
| 1320 | * leaky functions. |
| 1321 | * |
| 1322 | * Returns true if the clause contains any non-leakproof functions that are |
| 1323 | * passed Var nodes of the current query level, and which might therefore leak |
| 1324 | * data. Such clauses must be applied after any lower-level security barrier |
| 1325 | * clauses. |
| 1326 | */ |
| 1327 | bool |
| 1328 | contain_leaked_vars(Node *clause) |
| 1329 | { |
| 1330 | return contain_leaked_vars_walker(clause, NULL); |
| 1331 | } |
| 1332 | |
| 1333 | static bool |
| 1334 | contain_leaked_vars_checker(Oid func_id, void *context) |
| 1335 | { |
| 1336 | return !get_func_leakproof(func_id); |
| 1337 | } |
| 1338 | |
| 1339 | static bool |
| 1340 | contain_leaked_vars_walker(Node *node, void *context) |
| 1341 | { |
| 1342 | if (node == NULL) |
| 1343 | return false; |
| 1344 | |
| 1345 | switch (nodeTag(node)) |
| 1346 | { |
| 1347 | case T_Var: |
| 1348 | case T_Const: |
| 1349 | case T_Param: |
| 1350 | case T_ArrayExpr: |
| 1351 | case T_FieldSelect: |
| 1352 | case T_FieldStore: |
| 1353 | case T_NamedArgExpr: |
| 1354 | case T_BoolExpr: |
| 1355 | case T_RelabelType: |
| 1356 | case T_CollateExpr: |
| 1357 | case T_CaseExpr: |
| 1358 | case T_CaseTestExpr: |
| 1359 | case T_RowExpr: |
| 1360 | case T_SQLValueFunction: |
| 1361 | case T_NullTest: |
| 1362 | case T_BooleanTest: |
| 1363 | case T_NextValueExpr: |
| 1364 | case T_List: |
| 1365 | |
| 1366 | /* |
| 1367 | * We know these node types don't contain function calls; but |
| 1368 | * something further down in the node tree might. |
| 1369 | */ |
| 1370 | break; |
| 1371 | |
| 1372 | case T_FuncExpr: |
| 1373 | case T_OpExpr: |
| 1374 | case T_DistinctExpr: |
| 1375 | case T_NullIfExpr: |
| 1376 | case T_ScalarArrayOpExpr: |
| 1377 | case T_CoerceViaIO: |
| 1378 | case T_ArrayCoerceExpr: |
| 1379 | case T_SubscriptingRef: |
| 1380 | |
| 1381 | /* |
| 1382 | * If node contains a leaky function call, and there's any Var |
| 1383 | * underneath it, reject. |
| 1384 | */ |
| 1385 | if (check_functions_in_node(node, contain_leaked_vars_checker, |
| 1386 | context) && |
| 1387 | contain_var_clause(node)) |
| 1388 | return true; |
| 1389 | break; |
| 1390 | |
| 1391 | case T_RowCompareExpr: |
| 1392 | { |
| 1393 | /* |
| 1394 | * It's worth special-casing this because a leaky comparison |
| 1395 | * function only compromises one pair of row elements, which |
| 1396 | * might not contain Vars while others do. |
| 1397 | */ |
| 1398 | RowCompareExpr *rcexpr = (RowCompareExpr *) node; |
| 1399 | ListCell *opid; |
| 1400 | ListCell *larg; |
| 1401 | ListCell *rarg; |
| 1402 | |
| 1403 | forthree(opid, rcexpr->opnos, |
| 1404 | larg, rcexpr->largs, |
| 1405 | rarg, rcexpr->rargs) |
| 1406 | { |
| 1407 | Oid funcid = get_opcode(lfirst_oid(opid)); |
| 1408 | |
| 1409 | if (!get_func_leakproof(funcid) && |
| 1410 | (contain_var_clause((Node *) lfirst(larg)) || |
| 1411 | contain_var_clause((Node *) lfirst(rarg)))) |
| 1412 | return true; |
| 1413 | } |
| 1414 | } |
| 1415 | break; |
| 1416 | |
| 1417 | case T_MinMaxExpr: |
| 1418 | { |
| 1419 | /* |
| 1420 | * MinMaxExpr is leakproof if the comparison function it calls |
| 1421 | * is leakproof. |
| 1422 | */ |
| 1423 | MinMaxExpr *minmaxexpr = (MinMaxExpr *) node; |
| 1424 | TypeCacheEntry *typentry; |
| 1425 | bool leakproof; |
| 1426 | |
| 1427 | /* Look up the btree comparison function for the datatype */ |
| 1428 | typentry = lookup_type_cache(minmaxexpr->minmaxtype, |
| 1429 | TYPECACHE_CMP_PROC); |
| 1430 | if (OidIsValid(typentry->cmp_proc)) |
| 1431 | leakproof = get_func_leakproof(typentry->cmp_proc); |
| 1432 | else |
| 1433 | { |
| 1434 | /* |
| 1435 | * The executor will throw an error, but here we just |
| 1436 | * treat the missing function as leaky. |
| 1437 | */ |
| 1438 | leakproof = false; |
| 1439 | } |
| 1440 | |
| 1441 | if (!leakproof && |
| 1442 | contain_var_clause((Node *) minmaxexpr->args)) |
| 1443 | return true; |
| 1444 | } |
| 1445 | break; |
| 1446 | |
| 1447 | case T_CurrentOfExpr: |
| 1448 | |
| 1449 | /* |
| 1450 | * WHERE CURRENT OF doesn't contain leaky function calls. |
| 1451 | * Moreover, it is essential that this is considered non-leaky, |
| 1452 | * since the planner must always generate a TID scan when CURRENT |
| 1453 | * OF is present -- cf. cost_tidscan. |
| 1454 | */ |
| 1455 | return false; |
| 1456 | |
| 1457 | default: |
| 1458 | |
| 1459 | /* |
| 1460 | * If we don't recognize the node tag, assume it might be leaky. |
| 1461 | * This prevents an unexpected security hole if someone adds a new |
| 1462 | * node type that can call a function. |
| 1463 | */ |
| 1464 | return true; |
| 1465 | } |
| 1466 | return expression_tree_walker(node, contain_leaked_vars_walker, |
| 1467 | context); |
| 1468 | } |
| 1469 | |
| 1470 | /* |
| 1471 | * find_nonnullable_rels |
| 1472 | * Determine which base rels are forced nonnullable by given clause. |
| 1473 | * |
| 1474 | * Returns the set of all Relids that are referenced in the clause in such |
| 1475 | * a way that the clause cannot possibly return TRUE if any of these Relids |
| 1476 | * is an all-NULL row. (It is OK to err on the side of conservatism; hence |
| 1477 | * the analysis here is simplistic.) |
| 1478 | * |
| 1479 | * The semantics here are subtly different from contain_nonstrict_functions: |
| 1480 | * that function is concerned with NULL results from arbitrary expressions, |
| 1481 | * but here we assume that the input is a Boolean expression, and wish to |
| 1482 | * see if NULL inputs will provably cause a FALSE-or-NULL result. We expect |
| 1483 | * the expression to have been AND/OR flattened and converted to implicit-AND |
| 1484 | * format. |
| 1485 | * |
| 1486 | * Note: this function is largely duplicative of find_nonnullable_vars(). |
| 1487 | * The reason not to simplify this function into a thin wrapper around |
| 1488 | * find_nonnullable_vars() is that the tested conditions really are different: |
| 1489 | * a clause like "t1.v1 IS NOT NULL OR t1.v2 IS NOT NULL" does not prove |
| 1490 | * that either v1 or v2 can't be NULL, but it does prove that the t1 row |
| 1491 | * as a whole can't be all-NULL. Also, the behavior for PHVs is different. |
| 1492 | * |
| 1493 | * top_level is true while scanning top-level AND/OR structure; here, showing |
| 1494 | * the result is either FALSE or NULL is good enough. top_level is false when |
| 1495 | * we have descended below a NOT or a strict function: now we must be able to |
| 1496 | * prove that the subexpression goes to NULL. |
| 1497 | * |
| 1498 | * We don't use expression_tree_walker here because we don't want to descend |
| 1499 | * through very many kinds of nodes; only the ones we can be sure are strict. |
| 1500 | */ |
| 1501 | Relids |
| 1502 | find_nonnullable_rels(Node *clause) |
| 1503 | { |
| 1504 | return find_nonnullable_rels_walker(clause, true); |
| 1505 | } |
| 1506 | |
| 1507 | static Relids |
| 1508 | find_nonnullable_rels_walker(Node *node, bool top_level) |
| 1509 | { |
| 1510 | Relids result = NULL; |
| 1511 | ListCell *l; |
| 1512 | |
| 1513 | if (node == NULL) |
| 1514 | return NULL; |
| 1515 | if (IsA(node, Var)) |
| 1516 | { |
| 1517 | Var *var = (Var *) node; |
| 1518 | |
| 1519 | if (var->varlevelsup == 0) |
| 1520 | result = bms_make_singleton(var->varno); |
| 1521 | } |
| 1522 | else if (IsA(node, List)) |
| 1523 | { |
| 1524 | /* |
| 1525 | * At top level, we are examining an implicit-AND list: if any of the |
| 1526 | * arms produces FALSE-or-NULL then the result is FALSE-or-NULL. If |
| 1527 | * not at top level, we are examining the arguments of a strict |
| 1528 | * function: if any of them produce NULL then the result of the |
| 1529 | * function must be NULL. So in both cases, the set of nonnullable |
| 1530 | * rels is the union of those found in the arms, and we pass down the |
| 1531 | * top_level flag unmodified. |
| 1532 | */ |
| 1533 | foreach(l, (List *) node) |
| 1534 | { |
| 1535 | result = bms_join(result, |
| 1536 | find_nonnullable_rels_walker(lfirst(l), |
| 1537 | top_level)); |
| 1538 | } |
| 1539 | } |
| 1540 | else if (IsA(node, FuncExpr)) |
| 1541 | { |
| 1542 | FuncExpr *expr = (FuncExpr *) node; |
| 1543 | |
| 1544 | if (func_strict(expr->funcid)) |
| 1545 | result = find_nonnullable_rels_walker((Node *) expr->args, false); |
| 1546 | } |
| 1547 | else if (IsA(node, OpExpr)) |
| 1548 | { |
| 1549 | OpExpr *expr = (OpExpr *) node; |
| 1550 | |
| 1551 | set_opfuncid(expr); |
| 1552 | if (func_strict(expr->opfuncid)) |
| 1553 | result = find_nonnullable_rels_walker((Node *) expr->args, false); |
| 1554 | } |
| 1555 | else if (IsA(node, ScalarArrayOpExpr)) |
| 1556 | { |
| 1557 | ScalarArrayOpExpr *expr = (ScalarArrayOpExpr *) node; |
| 1558 | |
| 1559 | if (is_strict_saop(expr, true)) |
| 1560 | result = find_nonnullable_rels_walker((Node *) expr->args, false); |
| 1561 | } |
| 1562 | else if (IsA(node, BoolExpr)) |
| 1563 | { |
| 1564 | BoolExpr *expr = (BoolExpr *) node; |
| 1565 | |
| 1566 | switch (expr->boolop) |
| 1567 | { |
| 1568 | case AND_EXPR: |
| 1569 | /* At top level we can just recurse (to the List case) */ |
| 1570 | if (top_level) |
| 1571 | { |
| 1572 | result = find_nonnullable_rels_walker((Node *) expr->args, |
| 1573 | top_level); |
| 1574 | break; |
| 1575 | } |
| 1576 | |
| 1577 | /* |
| 1578 | * Below top level, even if one arm produces NULL, the result |
| 1579 | * could be FALSE (hence not NULL). However, if *all* the |
| 1580 | * arms produce NULL then the result is NULL, so we can take |
| 1581 | * the intersection of the sets of nonnullable rels, just as |
| 1582 | * for OR. Fall through to share code. |
| 1583 | */ |
| 1584 | /* FALL THRU */ |
| 1585 | case OR_EXPR: |
| 1586 | |
| 1587 | /* |
| 1588 | * OR is strict if all of its arms are, so we can take the |
| 1589 | * intersection of the sets of nonnullable rels for each arm. |
| 1590 | * This works for both values of top_level. |
| 1591 | */ |
| 1592 | foreach(l, expr->args) |
| 1593 | { |
| 1594 | Relids subresult; |
| 1595 | |
| 1596 | subresult = find_nonnullable_rels_walker(lfirst(l), |
| 1597 | top_level); |
| 1598 | if (result == NULL) /* first subresult? */ |
| 1599 | result = subresult; |
| 1600 | else |
| 1601 | result = bms_int_members(result, subresult); |
| 1602 | |
| 1603 | /* |
| 1604 | * If the intersection is empty, we can stop looking. This |
| 1605 | * also justifies the test for first-subresult above. |
| 1606 | */ |
| 1607 | if (bms_is_empty(result)) |
| 1608 | break; |
| 1609 | } |
| 1610 | break; |
| 1611 | case NOT_EXPR: |
| 1612 | /* NOT will return null if its arg is null */ |
| 1613 | result = find_nonnullable_rels_walker((Node *) expr->args, |
| 1614 | false); |
| 1615 | break; |
| 1616 | default: |
| 1617 | elog(ERROR, "unrecognized boolop: %d" , (int) expr->boolop); |
| 1618 | break; |
| 1619 | } |
| 1620 | } |
| 1621 | else if (IsA(node, RelabelType)) |
| 1622 | { |
| 1623 | RelabelType *expr = (RelabelType *) node; |
| 1624 | |
| 1625 | result = find_nonnullable_rels_walker((Node *) expr->arg, top_level); |
| 1626 | } |
| 1627 | else if (IsA(node, CoerceViaIO)) |
| 1628 | { |
| 1629 | /* not clear this is useful, but it can't hurt */ |
| 1630 | CoerceViaIO *expr = (CoerceViaIO *) node; |
| 1631 | |
| 1632 | result = find_nonnullable_rels_walker((Node *) expr->arg, top_level); |
| 1633 | } |
| 1634 | else if (IsA(node, ArrayCoerceExpr)) |
| 1635 | { |
| 1636 | /* ArrayCoerceExpr is strict at the array level; ignore elemexpr */ |
| 1637 | ArrayCoerceExpr *expr = (ArrayCoerceExpr *) node; |
| 1638 | |
| 1639 | result = find_nonnullable_rels_walker((Node *) expr->arg, top_level); |
| 1640 | } |
| 1641 | else if (IsA(node, ConvertRowtypeExpr)) |
| 1642 | { |
| 1643 | /* not clear this is useful, but it can't hurt */ |
| 1644 | ConvertRowtypeExpr *expr = (ConvertRowtypeExpr *) node; |
| 1645 | |
| 1646 | result = find_nonnullable_rels_walker((Node *) expr->arg, top_level); |
| 1647 | } |
| 1648 | else if (IsA(node, CollateExpr)) |
| 1649 | { |
| 1650 | CollateExpr *expr = (CollateExpr *) node; |
| 1651 | |
| 1652 | result = find_nonnullable_rels_walker((Node *) expr->arg, top_level); |
| 1653 | } |
| 1654 | else if (IsA(node, NullTest)) |
| 1655 | { |
| 1656 | /* IS NOT NULL can be considered strict, but only at top level */ |
| 1657 | NullTest *expr = (NullTest *) node; |
| 1658 | |
| 1659 | if (top_level && expr->nulltesttype == IS_NOT_NULL && !expr->argisrow) |
| 1660 | result = find_nonnullable_rels_walker((Node *) expr->arg, false); |
| 1661 | } |
| 1662 | else if (IsA(node, BooleanTest)) |
| 1663 | { |
| 1664 | /* Boolean tests that reject NULL are strict at top level */ |
| 1665 | BooleanTest *expr = (BooleanTest *) node; |
| 1666 | |
| 1667 | if (top_level && |
| 1668 | (expr->booltesttype == IS_TRUE || |
| 1669 | expr->booltesttype == IS_FALSE || |
| 1670 | expr->booltesttype == IS_NOT_UNKNOWN)) |
| 1671 | result = find_nonnullable_rels_walker((Node *) expr->arg, false); |
| 1672 | } |
| 1673 | else if (IsA(node, PlaceHolderVar)) |
| 1674 | { |
| 1675 | PlaceHolderVar *phv = (PlaceHolderVar *) node; |
| 1676 | |
| 1677 | /* |
| 1678 | * If the contained expression forces any rels non-nullable, so does |
| 1679 | * the PHV. |
| 1680 | */ |
| 1681 | result = find_nonnullable_rels_walker((Node *) phv->phexpr, top_level); |
| 1682 | |
| 1683 | /* |
| 1684 | * If the PHV's syntactic scope is exactly one rel, it will be forced |
| 1685 | * to be evaluated at that rel, and so it will behave like a Var of |
| 1686 | * that rel: if the rel's entire output goes to null, so will the PHV. |
| 1687 | * (If the syntactic scope is a join, we know that the PHV will go to |
| 1688 | * null if the whole join does; but that is AND semantics while we |
| 1689 | * need OR semantics for find_nonnullable_rels' result, so we can't do |
| 1690 | * anything with the knowledge.) |
| 1691 | */ |
| 1692 | if (phv->phlevelsup == 0 && |
| 1693 | bms_membership(phv->phrels) == BMS_SINGLETON) |
| 1694 | result = bms_add_members(result, phv->phrels); |
| 1695 | } |
| 1696 | return result; |
| 1697 | } |
| 1698 | |
| 1699 | /* |
| 1700 | * find_nonnullable_vars |
| 1701 | * Determine which Vars are forced nonnullable by given clause. |
| 1702 | * |
| 1703 | * Returns a list of all level-zero Vars that are referenced in the clause in |
| 1704 | * such a way that the clause cannot possibly return TRUE if any of these Vars |
| 1705 | * is NULL. (It is OK to err on the side of conservatism; hence the analysis |
| 1706 | * here is simplistic.) |
| 1707 | * |
| 1708 | * The semantics here are subtly different from contain_nonstrict_functions: |
| 1709 | * that function is concerned with NULL results from arbitrary expressions, |
| 1710 | * but here we assume that the input is a Boolean expression, and wish to |
| 1711 | * see if NULL inputs will provably cause a FALSE-or-NULL result. We expect |
| 1712 | * the expression to have been AND/OR flattened and converted to implicit-AND |
| 1713 | * format. |
| 1714 | * |
| 1715 | * The result is a palloc'd List, but we have not copied the member Var nodes. |
| 1716 | * Also, we don't bother trying to eliminate duplicate entries. |
| 1717 | * |
| 1718 | * top_level is true while scanning top-level AND/OR structure; here, showing |
| 1719 | * the result is either FALSE or NULL is good enough. top_level is false when |
| 1720 | * we have descended below a NOT or a strict function: now we must be able to |
| 1721 | * prove that the subexpression goes to NULL. |
| 1722 | * |
| 1723 | * We don't use expression_tree_walker here because we don't want to descend |
| 1724 | * through very many kinds of nodes; only the ones we can be sure are strict. |
| 1725 | */ |
| 1726 | List * |
| 1727 | find_nonnullable_vars(Node *clause) |
| 1728 | { |
| 1729 | return find_nonnullable_vars_walker(clause, true); |
| 1730 | } |
| 1731 | |
| 1732 | static List * |
| 1733 | find_nonnullable_vars_walker(Node *node, bool top_level) |
| 1734 | { |
| 1735 | List *result = NIL; |
| 1736 | ListCell *l; |
| 1737 | |
| 1738 | if (node == NULL) |
| 1739 | return NIL; |
| 1740 | if (IsA(node, Var)) |
| 1741 | { |
| 1742 | Var *var = (Var *) node; |
| 1743 | |
| 1744 | if (var->varlevelsup == 0) |
| 1745 | result = list_make1(var); |
| 1746 | } |
| 1747 | else if (IsA(node, List)) |
| 1748 | { |
| 1749 | /* |
| 1750 | * At top level, we are examining an implicit-AND list: if any of the |
| 1751 | * arms produces FALSE-or-NULL then the result is FALSE-or-NULL. If |
| 1752 | * not at top level, we are examining the arguments of a strict |
| 1753 | * function: if any of them produce NULL then the result of the |
| 1754 | * function must be NULL. So in both cases, the set of nonnullable |
| 1755 | * vars is the union of those found in the arms, and we pass down the |
| 1756 | * top_level flag unmodified. |
| 1757 | */ |
| 1758 | foreach(l, (List *) node) |
| 1759 | { |
| 1760 | result = list_concat(result, |
| 1761 | find_nonnullable_vars_walker(lfirst(l), |
| 1762 | top_level)); |
| 1763 | } |
| 1764 | } |
| 1765 | else if (IsA(node, FuncExpr)) |
| 1766 | { |
| 1767 | FuncExpr *expr = (FuncExpr *) node; |
| 1768 | |
| 1769 | if (func_strict(expr->funcid)) |
| 1770 | result = find_nonnullable_vars_walker((Node *) expr->args, false); |
| 1771 | } |
| 1772 | else if (IsA(node, OpExpr)) |
| 1773 | { |
| 1774 | OpExpr *expr = (OpExpr *) node; |
| 1775 | |
| 1776 | set_opfuncid(expr); |
| 1777 | if (func_strict(expr->opfuncid)) |
| 1778 | result = find_nonnullable_vars_walker((Node *) expr->args, false); |
| 1779 | } |
| 1780 | else if (IsA(node, ScalarArrayOpExpr)) |
| 1781 | { |
| 1782 | ScalarArrayOpExpr *expr = (ScalarArrayOpExpr *) node; |
| 1783 | |
| 1784 | if (is_strict_saop(expr, true)) |
| 1785 | result = find_nonnullable_vars_walker((Node *) expr->args, false); |
| 1786 | } |
| 1787 | else if (IsA(node, BoolExpr)) |
| 1788 | { |
| 1789 | BoolExpr *expr = (BoolExpr *) node; |
| 1790 | |
| 1791 | switch (expr->boolop) |
| 1792 | { |
| 1793 | case AND_EXPR: |
| 1794 | /* At top level we can just recurse (to the List case) */ |
| 1795 | if (top_level) |
| 1796 | { |
| 1797 | result = find_nonnullable_vars_walker((Node *) expr->args, |
| 1798 | top_level); |
| 1799 | break; |
| 1800 | } |
| 1801 | |
| 1802 | /* |
| 1803 | * Below top level, even if one arm produces NULL, the result |
| 1804 | * could be FALSE (hence not NULL). However, if *all* the |
| 1805 | * arms produce NULL then the result is NULL, so we can take |
| 1806 | * the intersection of the sets of nonnullable vars, just as |
| 1807 | * for OR. Fall through to share code. |
| 1808 | */ |
| 1809 | /* FALL THRU */ |
| 1810 | case OR_EXPR: |
| 1811 | |
| 1812 | /* |
| 1813 | * OR is strict if all of its arms are, so we can take the |
| 1814 | * intersection of the sets of nonnullable vars for each arm. |
| 1815 | * This works for both values of top_level. |
| 1816 | */ |
| 1817 | foreach(l, expr->args) |
| 1818 | { |
| 1819 | List *subresult; |
| 1820 | |
| 1821 | subresult = find_nonnullable_vars_walker(lfirst(l), |
| 1822 | top_level); |
| 1823 | if (result == NIL) /* first subresult? */ |
| 1824 | result = subresult; |
| 1825 | else |
| 1826 | result = list_intersection(result, subresult); |
| 1827 | |
| 1828 | /* |
| 1829 | * If the intersection is empty, we can stop looking. This |
| 1830 | * also justifies the test for first-subresult above. |
| 1831 | */ |
| 1832 | if (result == NIL) |
| 1833 | break; |
| 1834 | } |
| 1835 | break; |
| 1836 | case NOT_EXPR: |
| 1837 | /* NOT will return null if its arg is null */ |
| 1838 | result = find_nonnullable_vars_walker((Node *) expr->args, |
| 1839 | false); |
| 1840 | break; |
| 1841 | default: |
| 1842 | elog(ERROR, "unrecognized boolop: %d" , (int) expr->boolop); |
| 1843 | break; |
| 1844 | } |
| 1845 | } |
| 1846 | else if (IsA(node, RelabelType)) |
| 1847 | { |
| 1848 | RelabelType *expr = (RelabelType *) node; |
| 1849 | |
| 1850 | result = find_nonnullable_vars_walker((Node *) expr->arg, top_level); |
| 1851 | } |
| 1852 | else if (IsA(node, CoerceViaIO)) |
| 1853 | { |
| 1854 | /* not clear this is useful, but it can't hurt */ |
| 1855 | CoerceViaIO *expr = (CoerceViaIO *) node; |
| 1856 | |
| 1857 | result = find_nonnullable_vars_walker((Node *) expr->arg, false); |
| 1858 | } |
| 1859 | else if (IsA(node, ArrayCoerceExpr)) |
| 1860 | { |
| 1861 | /* ArrayCoerceExpr is strict at the array level; ignore elemexpr */ |
| 1862 | ArrayCoerceExpr *expr = (ArrayCoerceExpr *) node; |
| 1863 | |
| 1864 | result = find_nonnullable_vars_walker((Node *) expr->arg, top_level); |
| 1865 | } |
| 1866 | else if (IsA(node, ConvertRowtypeExpr)) |
| 1867 | { |
| 1868 | /* not clear this is useful, but it can't hurt */ |
| 1869 | ConvertRowtypeExpr *expr = (ConvertRowtypeExpr *) node; |
| 1870 | |
| 1871 | result = find_nonnullable_vars_walker((Node *) expr->arg, top_level); |
| 1872 | } |
| 1873 | else if (IsA(node, CollateExpr)) |
| 1874 | { |
| 1875 | CollateExpr *expr = (CollateExpr *) node; |
| 1876 | |
| 1877 | result = find_nonnullable_vars_walker((Node *) expr->arg, top_level); |
| 1878 | } |
| 1879 | else if (IsA(node, NullTest)) |
| 1880 | { |
| 1881 | /* IS NOT NULL can be considered strict, but only at top level */ |
| 1882 | NullTest *expr = (NullTest *) node; |
| 1883 | |
| 1884 | if (top_level && expr->nulltesttype == IS_NOT_NULL && !expr->argisrow) |
| 1885 | result = find_nonnullable_vars_walker((Node *) expr->arg, false); |
| 1886 | } |
| 1887 | else if (IsA(node, BooleanTest)) |
| 1888 | { |
| 1889 | /* Boolean tests that reject NULL are strict at top level */ |
| 1890 | BooleanTest *expr = (BooleanTest *) node; |
| 1891 | |
| 1892 | if (top_level && |
| 1893 | (expr->booltesttype == IS_TRUE || |
| 1894 | expr->booltesttype == IS_FALSE || |
| 1895 | expr->booltesttype == IS_NOT_UNKNOWN)) |
| 1896 | result = find_nonnullable_vars_walker((Node *) expr->arg, false); |
| 1897 | } |
| 1898 | else if (IsA(node, PlaceHolderVar)) |
| 1899 | { |
| 1900 | PlaceHolderVar *phv = (PlaceHolderVar *) node; |
| 1901 | |
| 1902 | result = find_nonnullable_vars_walker((Node *) phv->phexpr, top_level); |
| 1903 | } |
| 1904 | return result; |
| 1905 | } |
| 1906 | |
| 1907 | /* |
| 1908 | * find_forced_null_vars |
| 1909 | * Determine which Vars must be NULL for the given clause to return TRUE. |
| 1910 | * |
| 1911 | * This is the complement of find_nonnullable_vars: find the level-zero Vars |
| 1912 | * that must be NULL for the clause to return TRUE. (It is OK to err on the |
| 1913 | * side of conservatism; hence the analysis here is simplistic. In fact, |
| 1914 | * we only detect simple "var IS NULL" tests at the top level.) |
| 1915 | * |
| 1916 | * The result is a palloc'd List, but we have not copied the member Var nodes. |
| 1917 | * Also, we don't bother trying to eliminate duplicate entries. |
| 1918 | */ |
| 1919 | List * |
| 1920 | find_forced_null_vars(Node *node) |
| 1921 | { |
| 1922 | List *result = NIL; |
| 1923 | Var *var; |
| 1924 | ListCell *l; |
| 1925 | |
| 1926 | if (node == NULL) |
| 1927 | return NIL; |
| 1928 | /* Check single-clause cases using subroutine */ |
| 1929 | var = find_forced_null_var(node); |
| 1930 | if (var) |
| 1931 | { |
| 1932 | result = list_make1(var); |
| 1933 | } |
| 1934 | /* Otherwise, handle AND-conditions */ |
| 1935 | else if (IsA(node, List)) |
| 1936 | { |
| 1937 | /* |
| 1938 | * At top level, we are examining an implicit-AND list: if any of the |
| 1939 | * arms produces FALSE-or-NULL then the result is FALSE-or-NULL. |
| 1940 | */ |
| 1941 | foreach(l, (List *) node) |
| 1942 | { |
| 1943 | result = list_concat(result, |
| 1944 | find_forced_null_vars(lfirst(l))); |
| 1945 | } |
| 1946 | } |
| 1947 | else if (IsA(node, BoolExpr)) |
| 1948 | { |
| 1949 | BoolExpr *expr = (BoolExpr *) node; |
| 1950 | |
| 1951 | /* |
| 1952 | * We don't bother considering the OR case, because it's fairly |
| 1953 | * unlikely anyone would write "v1 IS NULL OR v1 IS NULL". Likewise, |
| 1954 | * the NOT case isn't worth expending code on. |
| 1955 | */ |
| 1956 | if (expr->boolop == AND_EXPR) |
| 1957 | { |
| 1958 | /* At top level we can just recurse (to the List case) */ |
| 1959 | result = find_forced_null_vars((Node *) expr->args); |
| 1960 | } |
| 1961 | } |
| 1962 | return result; |
| 1963 | } |
| 1964 | |
| 1965 | /* |
| 1966 | * find_forced_null_var |
| 1967 | * Return the Var forced null by the given clause, or NULL if it's |
| 1968 | * not an IS NULL-type clause. For success, the clause must enforce |
| 1969 | * *only* nullness of the particular Var, not any other conditions. |
| 1970 | * |
| 1971 | * This is just the single-clause case of find_forced_null_vars(), without |
| 1972 | * any allowance for AND conditions. It's used by initsplan.c on individual |
| 1973 | * qual clauses. The reason for not just applying find_forced_null_vars() |
| 1974 | * is that if an AND of an IS NULL clause with something else were to somehow |
| 1975 | * survive AND/OR flattening, initsplan.c might get fooled into discarding |
| 1976 | * the whole clause when only the IS NULL part of it had been proved redundant. |
| 1977 | */ |
| 1978 | Var * |
| 1979 | find_forced_null_var(Node *node) |
| 1980 | { |
| 1981 | if (node == NULL) |
| 1982 | return NULL; |
| 1983 | if (IsA(node, NullTest)) |
| 1984 | { |
| 1985 | /* check for var IS NULL */ |
| 1986 | NullTest *expr = (NullTest *) node; |
| 1987 | |
| 1988 | if (expr->nulltesttype == IS_NULL && !expr->argisrow) |
| 1989 | { |
| 1990 | Var *var = (Var *) expr->arg; |
| 1991 | |
| 1992 | if (var && IsA(var, Var) && |
| 1993 | var->varlevelsup == 0) |
| 1994 | return var; |
| 1995 | } |
| 1996 | } |
| 1997 | else if (IsA(node, BooleanTest)) |
| 1998 | { |
| 1999 | /* var IS UNKNOWN is equivalent to var IS NULL */ |
| 2000 | BooleanTest *expr = (BooleanTest *) node; |
| 2001 | |
| 2002 | if (expr->booltesttype == IS_UNKNOWN) |
| 2003 | { |
| 2004 | Var *var = (Var *) expr->arg; |
| 2005 | |
| 2006 | if (var && IsA(var, Var) && |
| 2007 | var->varlevelsup == 0) |
| 2008 | return var; |
| 2009 | } |
| 2010 | } |
| 2011 | return NULL; |
| 2012 | } |
| 2013 | |
| 2014 | /* |
| 2015 | * Can we treat a ScalarArrayOpExpr as strict? |
| 2016 | * |
| 2017 | * If "falseOK" is true, then a "false" result can be considered strict, |
| 2018 | * else we need to guarantee an actual NULL result for NULL input. |
| 2019 | * |
| 2020 | * "foo op ALL array" is strict if the op is strict *and* we can prove |
| 2021 | * that the array input isn't an empty array. We can check that |
| 2022 | * for the cases of an array constant and an ARRAY[] construct. |
| 2023 | * |
| 2024 | * "foo op ANY array" is strict in the falseOK sense if the op is strict. |
| 2025 | * If not falseOK, the test is the same as for "foo op ALL array". |
| 2026 | */ |
| 2027 | static bool |
| 2028 | is_strict_saop(ScalarArrayOpExpr *expr, bool falseOK) |
| 2029 | { |
| 2030 | Node *rightop; |
| 2031 | |
| 2032 | /* The contained operator must be strict. */ |
| 2033 | set_sa_opfuncid(expr); |
| 2034 | if (!func_strict(expr->opfuncid)) |
| 2035 | return false; |
| 2036 | /* If ANY and falseOK, that's all we need to check. */ |
| 2037 | if (expr->useOr && falseOK) |
| 2038 | return true; |
| 2039 | /* Else, we have to see if the array is provably non-empty. */ |
| 2040 | Assert(list_length(expr->args) == 2); |
| 2041 | rightop = (Node *) lsecond(expr->args); |
| 2042 | if (rightop && IsA(rightop, Const)) |
| 2043 | { |
| 2044 | Datum arraydatum = ((Const *) rightop)->constvalue; |
| 2045 | bool arrayisnull = ((Const *) rightop)->constisnull; |
| 2046 | ArrayType *arrayval; |
| 2047 | int nitems; |
| 2048 | |
| 2049 | if (arrayisnull) |
| 2050 | return false; |
| 2051 | arrayval = DatumGetArrayTypeP(arraydatum); |
| 2052 | nitems = ArrayGetNItems(ARR_NDIM(arrayval), ARR_DIMS(arrayval)); |
| 2053 | if (nitems > 0) |
| 2054 | return true; |
| 2055 | } |
| 2056 | else if (rightop && IsA(rightop, ArrayExpr)) |
| 2057 | { |
| 2058 | ArrayExpr *arrayexpr = (ArrayExpr *) rightop; |
| 2059 | |
| 2060 | if (arrayexpr->elements != NIL && !arrayexpr->multidims) |
| 2061 | return true; |
| 2062 | } |
| 2063 | return false; |
| 2064 | } |
| 2065 | |
| 2066 | |
| 2067 | /***************************************************************************** |
| 2068 | * Check for "pseudo-constant" clauses |
| 2069 | *****************************************************************************/ |
| 2070 | |
| 2071 | /* |
| 2072 | * is_pseudo_constant_clause |
| 2073 | * Detect whether an expression is "pseudo constant", ie, it contains no |
| 2074 | * variables of the current query level and no uses of volatile functions. |
| 2075 | * Such an expr is not necessarily a true constant: it can still contain |
| 2076 | * Params and outer-level Vars, not to mention functions whose results |
| 2077 | * may vary from one statement to the next. However, the expr's value |
| 2078 | * will be constant over any one scan of the current query, so it can be |
| 2079 | * used as, eg, an indexscan key. (Actually, the condition for indexscan |
| 2080 | * keys is weaker than this; see is_pseudo_constant_for_index().) |
| 2081 | * |
| 2082 | * CAUTION: this function omits to test for one very important class of |
| 2083 | * not-constant expressions, namely aggregates (Aggrefs). In current usage |
| 2084 | * this is only applied to WHERE clauses and so a check for Aggrefs would be |
| 2085 | * a waste of cycles; but be sure to also check contain_agg_clause() if you |
| 2086 | * want to know about pseudo-constness in other contexts. The same goes |
| 2087 | * for window functions (WindowFuncs). |
| 2088 | */ |
| 2089 | bool |
| 2090 | is_pseudo_constant_clause(Node *clause) |
| 2091 | { |
| 2092 | /* |
| 2093 | * We could implement this check in one recursive scan. But since the |
| 2094 | * check for volatile functions is both moderately expensive and unlikely |
| 2095 | * to fail, it seems better to look for Vars first and only check for |
| 2096 | * volatile functions if we find no Vars. |
| 2097 | */ |
| 2098 | if (!contain_var_clause(clause) && |
| 2099 | !contain_volatile_functions(clause)) |
| 2100 | return true; |
| 2101 | return false; |
| 2102 | } |
| 2103 | |
| 2104 | /* |
| 2105 | * is_pseudo_constant_clause_relids |
| 2106 | * Same as above, except caller already has available the var membership |
| 2107 | * of the expression; this lets us avoid the contain_var_clause() scan. |
| 2108 | */ |
| 2109 | bool |
| 2110 | is_pseudo_constant_clause_relids(Node *clause, Relids relids) |
| 2111 | { |
| 2112 | if (bms_is_empty(relids) && |
| 2113 | !contain_volatile_functions(clause)) |
| 2114 | return true; |
| 2115 | return false; |
| 2116 | } |
| 2117 | |
| 2118 | |
| 2119 | /***************************************************************************** |
| 2120 | * * |
| 2121 | * General clause-manipulating routines * |
| 2122 | * * |
| 2123 | *****************************************************************************/ |
| 2124 | |
| 2125 | /* |
| 2126 | * NumRelids |
| 2127 | * (formerly clause_relids) |
| 2128 | * |
| 2129 | * Returns the number of different relations referenced in 'clause'. |
| 2130 | */ |
| 2131 | int |
| 2132 | NumRelids(Node *clause) |
| 2133 | { |
| 2134 | Relids varnos = pull_varnos(clause); |
| 2135 | int result = bms_num_members(varnos); |
| 2136 | |
| 2137 | bms_free(varnos); |
| 2138 | return result; |
| 2139 | } |
| 2140 | |
| 2141 | /* |
| 2142 | * CommuteOpExpr: commute a binary operator clause |
| 2143 | * |
| 2144 | * XXX the clause is destructively modified! |
| 2145 | */ |
| 2146 | void |
| 2147 | CommuteOpExpr(OpExpr *clause) |
| 2148 | { |
| 2149 | Oid opoid; |
| 2150 | Node *temp; |
| 2151 | |
| 2152 | /* Sanity checks: caller is at fault if these fail */ |
| 2153 | if (!is_opclause(clause) || |
| 2154 | list_length(clause->args) != 2) |
| 2155 | elog(ERROR, "cannot commute non-binary-operator clause" ); |
| 2156 | |
| 2157 | opoid = get_commutator(clause->opno); |
| 2158 | |
| 2159 | if (!OidIsValid(opoid)) |
| 2160 | elog(ERROR, "could not find commutator for operator %u" , |
| 2161 | clause->opno); |
| 2162 | |
| 2163 | /* |
| 2164 | * modify the clause in-place! |
| 2165 | */ |
| 2166 | clause->opno = opoid; |
| 2167 | clause->opfuncid = InvalidOid; |
| 2168 | /* opresulttype, opretset, opcollid, inputcollid need not change */ |
| 2169 | |
| 2170 | temp = linitial(clause->args); |
| 2171 | linitial(clause->args) = lsecond(clause->args); |
| 2172 | lsecond(clause->args) = temp; |
| 2173 | } |
| 2174 | |
| 2175 | /* |
| 2176 | * Helper for eval_const_expressions: check that datatype of an attribute |
| 2177 | * is still what it was when the expression was parsed. This is needed to |
| 2178 | * guard against improper simplification after ALTER COLUMN TYPE. (XXX we |
| 2179 | * may well need to make similar checks elsewhere?) |
| 2180 | * |
| 2181 | * rowtypeid may come from a whole-row Var, and therefore it can be a domain |
| 2182 | * over composite, but for this purpose we only care about checking the type |
| 2183 | * of a contained field. |
| 2184 | */ |
| 2185 | static bool |
| 2186 | rowtype_field_matches(Oid rowtypeid, int fieldnum, |
| 2187 | Oid expectedtype, int32 expectedtypmod, |
| 2188 | Oid expectedcollation) |
| 2189 | { |
| 2190 | TupleDesc tupdesc; |
| 2191 | Form_pg_attribute attr; |
| 2192 | |
| 2193 | /* No issue for RECORD, since there is no way to ALTER such a type */ |
| 2194 | if (rowtypeid == RECORDOID) |
| 2195 | return true; |
| 2196 | tupdesc = lookup_rowtype_tupdesc_domain(rowtypeid, -1, false); |
| 2197 | if (fieldnum <= 0 || fieldnum > tupdesc->natts) |
| 2198 | { |
| 2199 | ReleaseTupleDesc(tupdesc); |
| 2200 | return false; |
| 2201 | } |
| 2202 | attr = TupleDescAttr(tupdesc, fieldnum - 1); |
| 2203 | if (attr->attisdropped || |
| 2204 | attr->atttypid != expectedtype || |
| 2205 | attr->atttypmod != expectedtypmod || |
| 2206 | attr->attcollation != expectedcollation) |
| 2207 | { |
| 2208 | ReleaseTupleDesc(tupdesc); |
| 2209 | return false; |
| 2210 | } |
| 2211 | ReleaseTupleDesc(tupdesc); |
| 2212 | return true; |
| 2213 | } |
| 2214 | |
| 2215 | |
| 2216 | /*-------------------- |
| 2217 | * eval_const_expressions |
| 2218 | * |
| 2219 | * Reduce any recognizably constant subexpressions of the given |
| 2220 | * expression tree, for example "2 + 2" => "4". More interestingly, |
| 2221 | * we can reduce certain boolean expressions even when they contain |
| 2222 | * non-constant subexpressions: "x OR true" => "true" no matter what |
| 2223 | * the subexpression x is. (XXX We assume that no such subexpression |
| 2224 | * will have important side-effects, which is not necessarily a good |
| 2225 | * assumption in the presence of user-defined functions; do we need a |
| 2226 | * pg_proc flag that prevents discarding the execution of a function?) |
| 2227 | * |
| 2228 | * We do understand that certain functions may deliver non-constant |
| 2229 | * results even with constant inputs, "nextval()" being the classic |
| 2230 | * example. Functions that are not marked "immutable" in pg_proc |
| 2231 | * will not be pre-evaluated here, although we will reduce their |
| 2232 | * arguments as far as possible. |
| 2233 | * |
| 2234 | * Whenever a function is eliminated from the expression by means of |
| 2235 | * constant-expression evaluation or inlining, we add the function to |
| 2236 | * root->glob->invalItems. This ensures the plan is known to depend on |
| 2237 | * such functions, even though they aren't referenced anymore. |
| 2238 | * |
| 2239 | * We assume that the tree has already been type-checked and contains |
| 2240 | * only operators and functions that are reasonable to try to execute. |
| 2241 | * |
| 2242 | * NOTE: "root" can be passed as NULL if the caller never wants to do any |
| 2243 | * Param substitutions nor receive info about inlined functions. |
| 2244 | * |
| 2245 | * NOTE: the planner assumes that this will always flatten nested AND and |
| 2246 | * OR clauses into N-argument form. See comments in prepqual.c. |
| 2247 | * |
| 2248 | * NOTE: another critical effect is that any function calls that require |
| 2249 | * default arguments will be expanded, and named-argument calls will be |
| 2250 | * converted to positional notation. The executor won't handle either. |
| 2251 | *-------------------- |
| 2252 | */ |
| 2253 | Node * |
| 2254 | eval_const_expressions(PlannerInfo *root, Node *node) |
| 2255 | { |
| 2256 | eval_const_expressions_context context; |
| 2257 | |
| 2258 | if (root) |
| 2259 | context.boundParams = root->glob->boundParams; /* bound Params */ |
| 2260 | else |
| 2261 | context.boundParams = NULL; |
| 2262 | context.root = root; /* for inlined-function dependencies */ |
| 2263 | context.active_fns = NIL; /* nothing being recursively simplified */ |
| 2264 | context.case_val = NULL; /* no CASE being examined */ |
| 2265 | context.estimate = false; /* safe transformations only */ |
| 2266 | return eval_const_expressions_mutator(node, &context); |
| 2267 | } |
| 2268 | |
| 2269 | /*-------------------- |
| 2270 | * estimate_expression_value |
| 2271 | * |
| 2272 | * This function attempts to estimate the value of an expression for |
| 2273 | * planning purposes. It is in essence a more aggressive version of |
| 2274 | * eval_const_expressions(): we will perform constant reductions that are |
| 2275 | * not necessarily 100% safe, but are reasonable for estimation purposes. |
| 2276 | * |
| 2277 | * Currently the extra steps that are taken in this mode are: |
| 2278 | * 1. Substitute values for Params, where a bound Param value has been made |
| 2279 | * available by the caller of planner(), even if the Param isn't marked |
| 2280 | * constant. This effectively means that we plan using the first supplied |
| 2281 | * value of the Param. |
| 2282 | * 2. Fold stable, as well as immutable, functions to constants. |
| 2283 | * 3. Reduce PlaceHolderVar nodes to their contained expressions. |
| 2284 | *-------------------- |
| 2285 | */ |
| 2286 | Node * |
| 2287 | estimate_expression_value(PlannerInfo *root, Node *node) |
| 2288 | { |
| 2289 | eval_const_expressions_context context; |
| 2290 | |
| 2291 | context.boundParams = root->glob->boundParams; /* bound Params */ |
| 2292 | /* we do not need to mark the plan as depending on inlined functions */ |
| 2293 | context.root = NULL; |
| 2294 | context.active_fns = NIL; /* nothing being recursively simplified */ |
| 2295 | context.case_val = NULL; /* no CASE being examined */ |
| 2296 | context.estimate = true; /* unsafe transformations OK */ |
| 2297 | return eval_const_expressions_mutator(node, &context); |
| 2298 | } |
| 2299 | |
| 2300 | /* |
| 2301 | * The generic case in eval_const_expressions_mutator is to recurse using |
| 2302 | * expression_tree_mutator, which will copy the given node unchanged but |
| 2303 | * const-simplify its arguments (if any) as far as possible. If the node |
| 2304 | * itself does immutable processing, and each of its arguments were reduced |
| 2305 | * to a Const, we can then reduce it to a Const using evaluate_expr. (Some |
| 2306 | * node types need more complicated logic; for example, a CASE expression |
| 2307 | * might be reducible to a constant even if not all its subtrees are.) |
| 2308 | */ |
| 2309 | #define ece_generic_processing(node) \ |
| 2310 | expression_tree_mutator((Node *) (node), eval_const_expressions_mutator, \ |
| 2311 | (void *) context) |
| 2312 | |
| 2313 | /* |
| 2314 | * Check whether all arguments of the given node were reduced to Consts. |
| 2315 | * By going directly to expression_tree_walker, contain_non_const_walker |
| 2316 | * is not applied to the node itself, only to its children. |
| 2317 | */ |
| 2318 | #define ece_all_arguments_const(node) \ |
| 2319 | (!expression_tree_walker((Node *) (node), contain_non_const_walker, NULL)) |
| 2320 | |
| 2321 | /* Generic macro for applying evaluate_expr */ |
| 2322 | #define ece_evaluate_expr(node) \ |
| 2323 | ((Node *) evaluate_expr((Expr *) (node), \ |
| 2324 | exprType((Node *) (node)), \ |
| 2325 | exprTypmod((Node *) (node)), \ |
| 2326 | exprCollation((Node *) (node)))) |
| 2327 | |
| 2328 | /* |
| 2329 | * Recursive guts of eval_const_expressions/estimate_expression_value |
| 2330 | */ |
| 2331 | static Node * |
| 2332 | eval_const_expressions_mutator(Node *node, |
| 2333 | eval_const_expressions_context *context) |
| 2334 | { |
| 2335 | if (node == NULL) |
| 2336 | return NULL; |
| 2337 | switch (nodeTag(node)) |
| 2338 | { |
| 2339 | case T_Param: |
| 2340 | { |
| 2341 | Param *param = (Param *) node; |
| 2342 | ParamListInfo paramLI = context->boundParams; |
| 2343 | |
| 2344 | /* Look to see if we've been given a value for this Param */ |
| 2345 | if (param->paramkind == PARAM_EXTERN && |
| 2346 | paramLI != NULL && |
| 2347 | param->paramid > 0 && |
| 2348 | param->paramid <= paramLI->numParams) |
| 2349 | { |
| 2350 | ParamExternData *prm; |
| 2351 | ParamExternData prmdata; |
| 2352 | |
| 2353 | /* |
| 2354 | * Give hook a chance in case parameter is dynamic. Tell |
| 2355 | * it that this fetch is speculative, so it should avoid |
| 2356 | * erroring out if parameter is unavailable. |
| 2357 | */ |
| 2358 | if (paramLI->paramFetch != NULL) |
| 2359 | prm = paramLI->paramFetch(paramLI, param->paramid, |
| 2360 | true, &prmdata); |
| 2361 | else |
| 2362 | prm = ¶mLI->params[param->paramid - 1]; |
| 2363 | |
| 2364 | /* |
| 2365 | * We don't just check OidIsValid, but insist that the |
| 2366 | * fetched type match the Param, just in case the hook did |
| 2367 | * something unexpected. No need to throw an error here |
| 2368 | * though; leave that for runtime. |
| 2369 | */ |
| 2370 | if (OidIsValid(prm->ptype) && |
| 2371 | prm->ptype == param->paramtype) |
| 2372 | { |
| 2373 | /* OK to substitute parameter value? */ |
| 2374 | if (context->estimate || |
| 2375 | (prm->pflags & PARAM_FLAG_CONST)) |
| 2376 | { |
| 2377 | /* |
| 2378 | * Return a Const representing the param value. |
| 2379 | * Must copy pass-by-ref datatypes, since the |
| 2380 | * Param might be in a memory context |
| 2381 | * shorter-lived than our output plan should be. |
| 2382 | */ |
| 2383 | int16 typLen; |
| 2384 | bool typByVal; |
| 2385 | Datum pval; |
| 2386 | |
| 2387 | get_typlenbyval(param->paramtype, |
| 2388 | &typLen, &typByVal); |
| 2389 | if (prm->isnull || typByVal) |
| 2390 | pval = prm->value; |
| 2391 | else |
| 2392 | pval = datumCopy(prm->value, typByVal, typLen); |
| 2393 | return (Node *) makeConst(param->paramtype, |
| 2394 | param->paramtypmod, |
| 2395 | param->paramcollid, |
| 2396 | (int) typLen, |
| 2397 | pval, |
| 2398 | prm->isnull, |
| 2399 | typByVal); |
| 2400 | } |
| 2401 | } |
| 2402 | } |
| 2403 | |
| 2404 | /* |
| 2405 | * Not replaceable, so just copy the Param (no need to |
| 2406 | * recurse) |
| 2407 | */ |
| 2408 | return (Node *) copyObject(param); |
| 2409 | } |
| 2410 | case T_WindowFunc: |
| 2411 | { |
| 2412 | WindowFunc *expr = (WindowFunc *) node; |
| 2413 | Oid funcid = expr->winfnoid; |
| 2414 | List *args; |
| 2415 | Expr *aggfilter; |
| 2416 | HeapTuple func_tuple; |
| 2417 | WindowFunc *newexpr; |
| 2418 | |
| 2419 | /* |
| 2420 | * We can't really simplify a WindowFunc node, but we mustn't |
| 2421 | * just fall through to the default processing, because we |
| 2422 | * have to apply expand_function_arguments to its argument |
| 2423 | * list. That takes care of inserting default arguments and |
| 2424 | * expanding named-argument notation. |
| 2425 | */ |
| 2426 | func_tuple = SearchSysCache1(PROCOID, ObjectIdGetDatum(funcid)); |
| 2427 | if (!HeapTupleIsValid(func_tuple)) |
| 2428 | elog(ERROR, "cache lookup failed for function %u" , funcid); |
| 2429 | |
| 2430 | args = expand_function_arguments(expr->args, expr->wintype, |
| 2431 | func_tuple); |
| 2432 | |
| 2433 | ReleaseSysCache(func_tuple); |
| 2434 | |
| 2435 | /* Now, recursively simplify the args (which are a List) */ |
| 2436 | args = (List *) |
| 2437 | expression_tree_mutator((Node *) args, |
| 2438 | eval_const_expressions_mutator, |
| 2439 | (void *) context); |
| 2440 | /* ... and the filter expression, which isn't */ |
| 2441 | aggfilter = (Expr *) |
| 2442 | eval_const_expressions_mutator((Node *) expr->aggfilter, |
| 2443 | context); |
| 2444 | |
| 2445 | /* And build the replacement WindowFunc node */ |
| 2446 | newexpr = makeNode(WindowFunc); |
| 2447 | newexpr->winfnoid = expr->winfnoid; |
| 2448 | newexpr->wintype = expr->wintype; |
| 2449 | newexpr->wincollid = expr->wincollid; |
| 2450 | newexpr->inputcollid = expr->inputcollid; |
| 2451 | newexpr->args = args; |
| 2452 | newexpr->aggfilter = aggfilter; |
| 2453 | newexpr->winref = expr->winref; |
| 2454 | newexpr->winstar = expr->winstar; |
| 2455 | newexpr->winagg = expr->winagg; |
| 2456 | newexpr->location = expr->location; |
| 2457 | |
| 2458 | return (Node *) newexpr; |
| 2459 | } |
| 2460 | case T_FuncExpr: |
| 2461 | { |
| 2462 | FuncExpr *expr = (FuncExpr *) node; |
| 2463 | List *args = expr->args; |
| 2464 | Expr *simple; |
| 2465 | FuncExpr *newexpr; |
| 2466 | |
| 2467 | /* |
| 2468 | * Code for op/func reduction is pretty bulky, so split it out |
| 2469 | * as a separate function. Note: exprTypmod normally returns |
| 2470 | * -1 for a FuncExpr, but not when the node is recognizably a |
| 2471 | * length coercion; we want to preserve the typmod in the |
| 2472 | * eventual Const if so. |
| 2473 | */ |
| 2474 | simple = simplify_function(expr->funcid, |
| 2475 | expr->funcresulttype, |
| 2476 | exprTypmod(node), |
| 2477 | expr->funccollid, |
| 2478 | expr->inputcollid, |
| 2479 | &args, |
| 2480 | expr->funcvariadic, |
| 2481 | true, |
| 2482 | true, |
| 2483 | context); |
| 2484 | if (simple) /* successfully simplified it */ |
| 2485 | return (Node *) simple; |
| 2486 | |
| 2487 | /* |
| 2488 | * The expression cannot be simplified any further, so build |
| 2489 | * and return a replacement FuncExpr node using the |
| 2490 | * possibly-simplified arguments. Note that we have also |
| 2491 | * converted the argument list to positional notation. |
| 2492 | */ |
| 2493 | newexpr = makeNode(FuncExpr); |
| 2494 | newexpr->funcid = expr->funcid; |
| 2495 | newexpr->funcresulttype = expr->funcresulttype; |
| 2496 | newexpr->funcretset = expr->funcretset; |
| 2497 | newexpr->funcvariadic = expr->funcvariadic; |
| 2498 | newexpr->funcformat = expr->funcformat; |
| 2499 | newexpr->funccollid = expr->funccollid; |
| 2500 | newexpr->inputcollid = expr->inputcollid; |
| 2501 | newexpr->args = args; |
| 2502 | newexpr->location = expr->location; |
| 2503 | return (Node *) newexpr; |
| 2504 | } |
| 2505 | case T_OpExpr: |
| 2506 | { |
| 2507 | OpExpr *expr = (OpExpr *) node; |
| 2508 | List *args = expr->args; |
| 2509 | Expr *simple; |
| 2510 | OpExpr *newexpr; |
| 2511 | |
| 2512 | /* |
| 2513 | * Need to get OID of underlying function. Okay to scribble |
| 2514 | * on input to this extent. |
| 2515 | */ |
| 2516 | set_opfuncid(expr); |
| 2517 | |
| 2518 | /* |
| 2519 | * Code for op/func reduction is pretty bulky, so split it out |
| 2520 | * as a separate function. |
| 2521 | */ |
| 2522 | simple = simplify_function(expr->opfuncid, |
| 2523 | expr->opresulttype, -1, |
| 2524 | expr->opcollid, |
| 2525 | expr->inputcollid, |
| 2526 | &args, |
| 2527 | false, |
| 2528 | true, |
| 2529 | true, |
| 2530 | context); |
| 2531 | if (simple) /* successfully simplified it */ |
| 2532 | return (Node *) simple; |
| 2533 | |
| 2534 | /* |
| 2535 | * If the operator is boolean equality or inequality, we know |
| 2536 | * how to simplify cases involving one constant and one |
| 2537 | * non-constant argument. |
| 2538 | */ |
| 2539 | if (expr->opno == BooleanEqualOperator || |
| 2540 | expr->opno == BooleanNotEqualOperator) |
| 2541 | { |
| 2542 | simple = (Expr *) simplify_boolean_equality(expr->opno, |
| 2543 | args); |
| 2544 | if (simple) /* successfully simplified it */ |
| 2545 | return (Node *) simple; |
| 2546 | } |
| 2547 | |
| 2548 | /* |
| 2549 | * The expression cannot be simplified any further, so build |
| 2550 | * and return a replacement OpExpr node using the |
| 2551 | * possibly-simplified arguments. |
| 2552 | */ |
| 2553 | newexpr = makeNode(OpExpr); |
| 2554 | newexpr->opno = expr->opno; |
| 2555 | newexpr->opfuncid = expr->opfuncid; |
| 2556 | newexpr->opresulttype = expr->opresulttype; |
| 2557 | newexpr->opretset = expr->opretset; |
| 2558 | newexpr->opcollid = expr->opcollid; |
| 2559 | newexpr->inputcollid = expr->inputcollid; |
| 2560 | newexpr->args = args; |
| 2561 | newexpr->location = expr->location; |
| 2562 | return (Node *) newexpr; |
| 2563 | } |
| 2564 | case T_DistinctExpr: |
| 2565 | { |
| 2566 | DistinctExpr *expr = (DistinctExpr *) node; |
| 2567 | List *args; |
| 2568 | ListCell *arg; |
| 2569 | bool has_null_input = false; |
| 2570 | bool all_null_input = true; |
| 2571 | bool has_nonconst_input = false; |
| 2572 | Expr *simple; |
| 2573 | DistinctExpr *newexpr; |
| 2574 | |
| 2575 | /* |
| 2576 | * Reduce constants in the DistinctExpr's arguments. We know |
| 2577 | * args is either NIL or a List node, so we can call |
| 2578 | * expression_tree_mutator directly rather than recursing to |
| 2579 | * self. |
| 2580 | */ |
| 2581 | args = (List *) expression_tree_mutator((Node *) expr->args, |
| 2582 | eval_const_expressions_mutator, |
| 2583 | (void *) context); |
| 2584 | |
| 2585 | /* |
| 2586 | * We must do our own check for NULLs because DistinctExpr has |
| 2587 | * different results for NULL input than the underlying |
| 2588 | * operator does. |
| 2589 | */ |
| 2590 | foreach(arg, args) |
| 2591 | { |
| 2592 | if (IsA(lfirst(arg), Const)) |
| 2593 | { |
| 2594 | has_null_input |= ((Const *) lfirst(arg))->constisnull; |
| 2595 | all_null_input &= ((Const *) lfirst(arg))->constisnull; |
| 2596 | } |
| 2597 | else |
| 2598 | has_nonconst_input = true; |
| 2599 | } |
| 2600 | |
| 2601 | /* all constants? then can optimize this out */ |
| 2602 | if (!has_nonconst_input) |
| 2603 | { |
| 2604 | /* all nulls? then not distinct */ |
| 2605 | if (all_null_input) |
| 2606 | return makeBoolConst(false, false); |
| 2607 | |
| 2608 | /* one null? then distinct */ |
| 2609 | if (has_null_input) |
| 2610 | return makeBoolConst(true, false); |
| 2611 | |
| 2612 | /* otherwise try to evaluate the '=' operator */ |
| 2613 | /* (NOT okay to try to inline it, though!) */ |
| 2614 | |
| 2615 | /* |
| 2616 | * Need to get OID of underlying function. Okay to |
| 2617 | * scribble on input to this extent. |
| 2618 | */ |
| 2619 | set_opfuncid((OpExpr *) expr); /* rely on struct |
| 2620 | * equivalence */ |
| 2621 | |
| 2622 | /* |
| 2623 | * Code for op/func reduction is pretty bulky, so split it |
| 2624 | * out as a separate function. |
| 2625 | */ |
| 2626 | simple = simplify_function(expr->opfuncid, |
| 2627 | expr->opresulttype, -1, |
| 2628 | expr->opcollid, |
| 2629 | expr->inputcollid, |
| 2630 | &args, |
| 2631 | false, |
| 2632 | false, |
| 2633 | false, |
| 2634 | context); |
| 2635 | if (simple) /* successfully simplified it */ |
| 2636 | { |
| 2637 | /* |
| 2638 | * Since the underlying operator is "=", must negate |
| 2639 | * its result |
| 2640 | */ |
| 2641 | Const *csimple = castNode(Const, simple); |
| 2642 | |
| 2643 | csimple->constvalue = |
| 2644 | BoolGetDatum(!DatumGetBool(csimple->constvalue)); |
| 2645 | return (Node *) csimple; |
| 2646 | } |
| 2647 | } |
| 2648 | |
| 2649 | /* |
| 2650 | * The expression cannot be simplified any further, so build |
| 2651 | * and return a replacement DistinctExpr node using the |
| 2652 | * possibly-simplified arguments. |
| 2653 | */ |
| 2654 | newexpr = makeNode(DistinctExpr); |
| 2655 | newexpr->opno = expr->opno; |
| 2656 | newexpr->opfuncid = expr->opfuncid; |
| 2657 | newexpr->opresulttype = expr->opresulttype; |
| 2658 | newexpr->opretset = expr->opretset; |
| 2659 | newexpr->opcollid = expr->opcollid; |
| 2660 | newexpr->inputcollid = expr->inputcollid; |
| 2661 | newexpr->args = args; |
| 2662 | newexpr->location = expr->location; |
| 2663 | return (Node *) newexpr; |
| 2664 | } |
| 2665 | case T_ScalarArrayOpExpr: |
| 2666 | { |
| 2667 | ScalarArrayOpExpr *saop; |
| 2668 | |
| 2669 | /* Copy the node and const-simplify its arguments */ |
| 2670 | saop = (ScalarArrayOpExpr *) ece_generic_processing(node); |
| 2671 | |
| 2672 | /* Make sure we know underlying function */ |
| 2673 | set_sa_opfuncid(saop); |
| 2674 | |
| 2675 | /* |
| 2676 | * If all arguments are Consts, and it's a safe function, we |
| 2677 | * can fold to a constant |
| 2678 | */ |
| 2679 | if (ece_all_arguments_const(saop) && |
| 2680 | ece_function_is_safe(saop->opfuncid, context)) |
| 2681 | return ece_evaluate_expr(saop); |
| 2682 | return (Node *) saop; |
| 2683 | } |
| 2684 | case T_BoolExpr: |
| 2685 | { |
| 2686 | BoolExpr *expr = (BoolExpr *) node; |
| 2687 | |
| 2688 | switch (expr->boolop) |
| 2689 | { |
| 2690 | case OR_EXPR: |
| 2691 | { |
| 2692 | List *newargs; |
| 2693 | bool haveNull = false; |
| 2694 | bool forceTrue = false; |
| 2695 | |
| 2696 | newargs = simplify_or_arguments(expr->args, |
| 2697 | context, |
| 2698 | &haveNull, |
| 2699 | &forceTrue); |
| 2700 | if (forceTrue) |
| 2701 | return makeBoolConst(true, false); |
| 2702 | if (haveNull) |
| 2703 | newargs = lappend(newargs, |
| 2704 | makeBoolConst(false, true)); |
| 2705 | /* If all the inputs are FALSE, result is FALSE */ |
| 2706 | if (newargs == NIL) |
| 2707 | return makeBoolConst(false, false); |
| 2708 | |
| 2709 | /* |
| 2710 | * If only one nonconst-or-NULL input, it's the |
| 2711 | * result |
| 2712 | */ |
| 2713 | if (list_length(newargs) == 1) |
| 2714 | return (Node *) linitial(newargs); |
| 2715 | /* Else we still need an OR node */ |
| 2716 | return (Node *) make_orclause(newargs); |
| 2717 | } |
| 2718 | case AND_EXPR: |
| 2719 | { |
| 2720 | List *newargs; |
| 2721 | bool haveNull = false; |
| 2722 | bool forceFalse = false; |
| 2723 | |
| 2724 | newargs = simplify_and_arguments(expr->args, |
| 2725 | context, |
| 2726 | &haveNull, |
| 2727 | &forceFalse); |
| 2728 | if (forceFalse) |
| 2729 | return makeBoolConst(false, false); |
| 2730 | if (haveNull) |
| 2731 | newargs = lappend(newargs, |
| 2732 | makeBoolConst(false, true)); |
| 2733 | /* If all the inputs are TRUE, result is TRUE */ |
| 2734 | if (newargs == NIL) |
| 2735 | return makeBoolConst(true, false); |
| 2736 | |
| 2737 | /* |
| 2738 | * If only one nonconst-or-NULL input, it's the |
| 2739 | * result |
| 2740 | */ |
| 2741 | if (list_length(newargs) == 1) |
| 2742 | return (Node *) linitial(newargs); |
| 2743 | /* Else we still need an AND node */ |
| 2744 | return (Node *) make_andclause(newargs); |
| 2745 | } |
| 2746 | case NOT_EXPR: |
| 2747 | { |
| 2748 | Node *arg; |
| 2749 | |
| 2750 | Assert(list_length(expr->args) == 1); |
| 2751 | arg = eval_const_expressions_mutator(linitial(expr->args), |
| 2752 | context); |
| 2753 | |
| 2754 | /* |
| 2755 | * Use negate_clause() to see if we can simplify |
| 2756 | * away the NOT. |
| 2757 | */ |
| 2758 | return negate_clause(arg); |
| 2759 | } |
| 2760 | default: |
| 2761 | elog(ERROR, "unrecognized boolop: %d" , |
| 2762 | (int) expr->boolop); |
| 2763 | break; |
| 2764 | } |
| 2765 | break; |
| 2766 | } |
| 2767 | case T_SubPlan: |
| 2768 | case T_AlternativeSubPlan: |
| 2769 | |
| 2770 | /* |
| 2771 | * Return a SubPlan unchanged --- too late to do anything with it. |
| 2772 | * |
| 2773 | * XXX should we ereport() here instead? Probably this routine |
| 2774 | * should never be invoked after SubPlan creation. |
| 2775 | */ |
| 2776 | return node; |
| 2777 | case T_RelabelType: |
| 2778 | { |
| 2779 | /* |
| 2780 | * If we can simplify the input to a constant, then we don't |
| 2781 | * need the RelabelType node anymore: just change the type |
| 2782 | * field of the Const node. Otherwise, must copy the |
| 2783 | * RelabelType node. |
| 2784 | */ |
| 2785 | RelabelType *relabel = (RelabelType *) node; |
| 2786 | Node *arg; |
| 2787 | |
| 2788 | arg = eval_const_expressions_mutator((Node *) relabel->arg, |
| 2789 | context); |
| 2790 | |
| 2791 | /* |
| 2792 | * If we find stacked RelabelTypes (eg, from foo :: int :: |
| 2793 | * oid) we can discard all but the top one. |
| 2794 | */ |
| 2795 | while (arg && IsA(arg, RelabelType)) |
| 2796 | arg = (Node *) ((RelabelType *) arg)->arg; |
| 2797 | |
| 2798 | if (arg && IsA(arg, Const)) |
| 2799 | { |
| 2800 | Const *con = (Const *) arg; |
| 2801 | |
| 2802 | con->consttype = relabel->resulttype; |
| 2803 | con->consttypmod = relabel->resulttypmod; |
| 2804 | con->constcollid = relabel->resultcollid; |
| 2805 | return (Node *) con; |
| 2806 | } |
| 2807 | else |
| 2808 | { |
| 2809 | RelabelType *newrelabel = makeNode(RelabelType); |
| 2810 | |
| 2811 | newrelabel->arg = (Expr *) arg; |
| 2812 | newrelabel->resulttype = relabel->resulttype; |
| 2813 | newrelabel->resulttypmod = relabel->resulttypmod; |
| 2814 | newrelabel->resultcollid = relabel->resultcollid; |
| 2815 | newrelabel->relabelformat = relabel->relabelformat; |
| 2816 | newrelabel->location = relabel->location; |
| 2817 | return (Node *) newrelabel; |
| 2818 | } |
| 2819 | } |
| 2820 | case T_CoerceViaIO: |
| 2821 | { |
| 2822 | CoerceViaIO *expr = (CoerceViaIO *) node; |
| 2823 | List *args; |
| 2824 | Oid outfunc; |
| 2825 | bool outtypisvarlena; |
| 2826 | Oid infunc; |
| 2827 | Oid intypioparam; |
| 2828 | Expr *simple; |
| 2829 | CoerceViaIO *newexpr; |
| 2830 | |
| 2831 | /* Make a List so we can use simplify_function */ |
| 2832 | args = list_make1(expr->arg); |
| 2833 | |
| 2834 | /* |
| 2835 | * CoerceViaIO represents calling the source type's output |
| 2836 | * function then the result type's input function. So, try to |
| 2837 | * simplify it as though it were a stack of two such function |
| 2838 | * calls. First we need to know what the functions are. |
| 2839 | * |
| 2840 | * Note that the coercion functions are assumed not to care |
| 2841 | * about input collation, so we just pass InvalidOid for that. |
| 2842 | */ |
| 2843 | getTypeOutputInfo(exprType((Node *) expr->arg), |
| 2844 | &outfunc, &outtypisvarlena); |
| 2845 | getTypeInputInfo(expr->resulttype, |
| 2846 | &infunc, &intypioparam); |
| 2847 | |
| 2848 | simple = simplify_function(outfunc, |
| 2849 | CSTRINGOID, -1, |
| 2850 | InvalidOid, |
| 2851 | InvalidOid, |
| 2852 | &args, |
| 2853 | false, |
| 2854 | true, |
| 2855 | true, |
| 2856 | context); |
| 2857 | if (simple) /* successfully simplified output fn */ |
| 2858 | { |
| 2859 | /* |
| 2860 | * Input functions may want 1 to 3 arguments. We always |
| 2861 | * supply all three, trusting that nothing downstream will |
| 2862 | * complain. |
| 2863 | */ |
| 2864 | args = list_make3(simple, |
| 2865 | makeConst(OIDOID, |
| 2866 | -1, |
| 2867 | InvalidOid, |
| 2868 | sizeof(Oid), |
| 2869 | ObjectIdGetDatum(intypioparam), |
| 2870 | false, |
| 2871 | true), |
| 2872 | makeConst(INT4OID, |
| 2873 | -1, |
| 2874 | InvalidOid, |
| 2875 | sizeof(int32), |
| 2876 | Int32GetDatum(-1), |
| 2877 | false, |
| 2878 | true)); |
| 2879 | |
| 2880 | simple = simplify_function(infunc, |
| 2881 | expr->resulttype, -1, |
| 2882 | expr->resultcollid, |
| 2883 | InvalidOid, |
| 2884 | &args, |
| 2885 | false, |
| 2886 | false, |
| 2887 | true, |
| 2888 | context); |
| 2889 | if (simple) /* successfully simplified input fn */ |
| 2890 | return (Node *) simple; |
| 2891 | } |
| 2892 | |
| 2893 | /* |
| 2894 | * The expression cannot be simplified any further, so build |
| 2895 | * and return a replacement CoerceViaIO node using the |
| 2896 | * possibly-simplified argument. |
| 2897 | */ |
| 2898 | newexpr = makeNode(CoerceViaIO); |
| 2899 | newexpr->arg = (Expr *) linitial(args); |
| 2900 | newexpr->resulttype = expr->resulttype; |
| 2901 | newexpr->resultcollid = expr->resultcollid; |
| 2902 | newexpr->coerceformat = expr->coerceformat; |
| 2903 | newexpr->location = expr->location; |
| 2904 | return (Node *) newexpr; |
| 2905 | } |
| 2906 | case T_ArrayCoerceExpr: |
| 2907 | { |
| 2908 | ArrayCoerceExpr *ac = makeNode(ArrayCoerceExpr); |
| 2909 | Node *save_case_val; |
| 2910 | |
| 2911 | /* |
| 2912 | * Copy the node and const-simplify its arguments. We can't |
| 2913 | * use ece_generic_processing() here because we need to mess |
| 2914 | * with case_val only while processing the elemexpr. |
| 2915 | */ |
| 2916 | memcpy(ac, node, sizeof(ArrayCoerceExpr)); |
| 2917 | ac->arg = (Expr *) |
| 2918 | eval_const_expressions_mutator((Node *) ac->arg, |
| 2919 | context); |
| 2920 | |
| 2921 | /* |
| 2922 | * Set up for the CaseTestExpr node contained in the elemexpr. |
| 2923 | * We must prevent it from absorbing any outer CASE value. |
| 2924 | */ |
| 2925 | save_case_val = context->case_val; |
| 2926 | context->case_val = NULL; |
| 2927 | |
| 2928 | ac->elemexpr = (Expr *) |
| 2929 | eval_const_expressions_mutator((Node *) ac->elemexpr, |
| 2930 | context); |
| 2931 | |
| 2932 | context->case_val = save_case_val; |
| 2933 | |
| 2934 | /* |
| 2935 | * If constant argument and the per-element expression is |
| 2936 | * immutable, we can simplify the whole thing to a constant. |
| 2937 | * Exception: although contain_mutable_functions considers |
| 2938 | * CoerceToDomain immutable for historical reasons, let's not |
| 2939 | * do so here; this ensures coercion to an array-over-domain |
| 2940 | * does not apply the domain's constraints until runtime. |
| 2941 | */ |
| 2942 | if (ac->arg && IsA(ac->arg, Const) && |
| 2943 | ac->elemexpr && !IsA(ac->elemexpr, CoerceToDomain) && |
| 2944 | !contain_mutable_functions((Node *) ac->elemexpr)) |
| 2945 | return ece_evaluate_expr(ac); |
| 2946 | |
| 2947 | return (Node *) ac; |
| 2948 | } |
| 2949 | case T_CollateExpr: |
| 2950 | { |
| 2951 | /* |
| 2952 | * If we can simplify the input to a constant, then we don't |
| 2953 | * need the CollateExpr node at all: just change the |
| 2954 | * constcollid field of the Const node. Otherwise, replace |
| 2955 | * the CollateExpr with a RelabelType. (We do that so as to |
| 2956 | * improve uniformity of expression representation and thus |
| 2957 | * simplify comparison of expressions.) |
| 2958 | */ |
| 2959 | CollateExpr *collate = (CollateExpr *) node; |
| 2960 | Node *arg; |
| 2961 | |
| 2962 | arg = eval_const_expressions_mutator((Node *) collate->arg, |
| 2963 | context); |
| 2964 | |
| 2965 | if (arg && IsA(arg, Const)) |
| 2966 | { |
| 2967 | Const *con = (Const *) arg; |
| 2968 | |
| 2969 | con->constcollid = collate->collOid; |
| 2970 | return (Node *) con; |
| 2971 | } |
| 2972 | else if (collate->collOid == exprCollation(arg)) |
| 2973 | { |
| 2974 | /* Don't need a RelabelType either... */ |
| 2975 | return arg; |
| 2976 | } |
| 2977 | else |
| 2978 | { |
| 2979 | RelabelType *relabel = makeNode(RelabelType); |
| 2980 | |
| 2981 | relabel->resulttype = exprType(arg); |
| 2982 | relabel->resulttypmod = exprTypmod(arg); |
| 2983 | relabel->resultcollid = collate->collOid; |
| 2984 | relabel->relabelformat = COERCE_IMPLICIT_CAST; |
| 2985 | relabel->location = collate->location; |
| 2986 | |
| 2987 | /* Don't create stacked RelabelTypes */ |
| 2988 | while (arg && IsA(arg, RelabelType)) |
| 2989 | arg = (Node *) ((RelabelType *) arg)->arg; |
| 2990 | relabel->arg = (Expr *) arg; |
| 2991 | |
| 2992 | return (Node *) relabel; |
| 2993 | } |
| 2994 | } |
| 2995 | case T_CaseExpr: |
| 2996 | { |
| 2997 | /*---------- |
| 2998 | * CASE expressions can be simplified if there are constant |
| 2999 | * condition clauses: |
| 3000 | * FALSE (or NULL): drop the alternative |
| 3001 | * TRUE: drop all remaining alternatives |
| 3002 | * If the first non-FALSE alternative is a constant TRUE, |
| 3003 | * we can simplify the entire CASE to that alternative's |
| 3004 | * expression. If there are no non-FALSE alternatives, |
| 3005 | * we simplify the entire CASE to the default result (ELSE). |
| 3006 | * |
| 3007 | * If we have a simple-form CASE with constant test |
| 3008 | * expression, we substitute the constant value for contained |
| 3009 | * CaseTestExpr placeholder nodes, so that we have the |
| 3010 | * opportunity to reduce constant test conditions. For |
| 3011 | * example this allows |
| 3012 | * CASE 0 WHEN 0 THEN 1 ELSE 1/0 END |
| 3013 | * to reduce to 1 rather than drawing a divide-by-0 error. |
| 3014 | * Note that when the test expression is constant, we don't |
| 3015 | * have to include it in the resulting CASE; for example |
| 3016 | * CASE 0 WHEN x THEN y ELSE z END |
| 3017 | * is transformed by the parser to |
| 3018 | * CASE 0 WHEN CaseTestExpr = x THEN y ELSE z END |
| 3019 | * which we can simplify to |
| 3020 | * CASE WHEN 0 = x THEN y ELSE z END |
| 3021 | * It is not necessary for the executor to evaluate the "arg" |
| 3022 | * expression when executing the CASE, since any contained |
| 3023 | * CaseTestExprs that might have referred to it will have been |
| 3024 | * replaced by the constant. |
| 3025 | *---------- |
| 3026 | */ |
| 3027 | CaseExpr *caseexpr = (CaseExpr *) node; |
| 3028 | CaseExpr *newcase; |
| 3029 | Node *save_case_val; |
| 3030 | Node *newarg; |
| 3031 | List *newargs; |
| 3032 | bool const_true_cond; |
| 3033 | Node *defresult = NULL; |
| 3034 | ListCell *arg; |
| 3035 | |
| 3036 | /* Simplify the test expression, if any */ |
| 3037 | newarg = eval_const_expressions_mutator((Node *) caseexpr->arg, |
| 3038 | context); |
| 3039 | |
| 3040 | /* Set up for contained CaseTestExpr nodes */ |
| 3041 | save_case_val = context->case_val; |
| 3042 | if (newarg && IsA(newarg, Const)) |
| 3043 | { |
| 3044 | context->case_val = newarg; |
| 3045 | newarg = NULL; /* not needed anymore, see above */ |
| 3046 | } |
| 3047 | else |
| 3048 | context->case_val = NULL; |
| 3049 | |
| 3050 | /* Simplify the WHEN clauses */ |
| 3051 | newargs = NIL; |
| 3052 | const_true_cond = false; |
| 3053 | foreach(arg, caseexpr->args) |
| 3054 | { |
| 3055 | CaseWhen *oldcasewhen = lfirst_node(CaseWhen, arg); |
| 3056 | Node *casecond; |
| 3057 | Node *caseresult; |
| 3058 | |
| 3059 | /* Simplify this alternative's test condition */ |
| 3060 | casecond = eval_const_expressions_mutator((Node *) oldcasewhen->expr, |
| 3061 | context); |
| 3062 | |
| 3063 | /* |
| 3064 | * If the test condition is constant FALSE (or NULL), then |
| 3065 | * drop this WHEN clause completely, without processing |
| 3066 | * the result. |
| 3067 | */ |
| 3068 | if (casecond && IsA(casecond, Const)) |
| 3069 | { |
| 3070 | Const *const_input = (Const *) casecond; |
| 3071 | |
| 3072 | if (const_input->constisnull || |
| 3073 | !DatumGetBool(const_input->constvalue)) |
| 3074 | continue; /* drop alternative with FALSE cond */ |
| 3075 | /* Else it's constant TRUE */ |
| 3076 | const_true_cond = true; |
| 3077 | } |
| 3078 | |
| 3079 | /* Simplify this alternative's result value */ |
| 3080 | caseresult = eval_const_expressions_mutator((Node *) oldcasewhen->result, |
| 3081 | context); |
| 3082 | |
| 3083 | /* If non-constant test condition, emit a new WHEN node */ |
| 3084 | if (!const_true_cond) |
| 3085 | { |
| 3086 | CaseWhen *newcasewhen = makeNode(CaseWhen); |
| 3087 | |
| 3088 | newcasewhen->expr = (Expr *) casecond; |
| 3089 | newcasewhen->result = (Expr *) caseresult; |
| 3090 | newcasewhen->location = oldcasewhen->location; |
| 3091 | newargs = lappend(newargs, newcasewhen); |
| 3092 | continue; |
| 3093 | } |
| 3094 | |
| 3095 | /* |
| 3096 | * Found a TRUE condition, so none of the remaining |
| 3097 | * alternatives can be reached. We treat the result as |
| 3098 | * the default result. |
| 3099 | */ |
| 3100 | defresult = caseresult; |
| 3101 | break; |
| 3102 | } |
| 3103 | |
| 3104 | /* Simplify the default result, unless we replaced it above */ |
| 3105 | if (!const_true_cond) |
| 3106 | defresult = eval_const_expressions_mutator((Node *) caseexpr->defresult, |
| 3107 | context); |
| 3108 | |
| 3109 | context->case_val = save_case_val; |
| 3110 | |
| 3111 | /* |
| 3112 | * If no non-FALSE alternatives, CASE reduces to the default |
| 3113 | * result |
| 3114 | */ |
| 3115 | if (newargs == NIL) |
| 3116 | return defresult; |
| 3117 | /* Otherwise we need a new CASE node */ |
| 3118 | newcase = makeNode(CaseExpr); |
| 3119 | newcase->casetype = caseexpr->casetype; |
| 3120 | newcase->casecollid = caseexpr->casecollid; |
| 3121 | newcase->arg = (Expr *) newarg; |
| 3122 | newcase->args = newargs; |
| 3123 | newcase->defresult = (Expr *) defresult; |
| 3124 | newcase->location = caseexpr->location; |
| 3125 | return (Node *) newcase; |
| 3126 | } |
| 3127 | case T_CaseTestExpr: |
| 3128 | { |
| 3129 | /* |
| 3130 | * If we know a constant test value for the current CASE |
| 3131 | * construct, substitute it for the placeholder. Else just |
| 3132 | * return the placeholder as-is. |
| 3133 | */ |
| 3134 | if (context->case_val) |
| 3135 | return copyObject(context->case_val); |
| 3136 | else |
| 3137 | return copyObject(node); |
| 3138 | } |
| 3139 | case T_SubscriptingRef: |
| 3140 | case T_ArrayExpr: |
| 3141 | case T_RowExpr: |
| 3142 | case T_MinMaxExpr: |
| 3143 | { |
| 3144 | /* |
| 3145 | * Generic handling for node types whose own processing is |
| 3146 | * known to be immutable, and for which we need no smarts |
| 3147 | * beyond "simplify if all inputs are constants". |
| 3148 | * |
| 3149 | * Treating MinMaxExpr this way amounts to assuming that the |
| 3150 | * btree comparison function it calls is immutable; see the |
| 3151 | * reasoning in contain_mutable_functions_walker. |
| 3152 | */ |
| 3153 | |
| 3154 | /* Copy the node and const-simplify its arguments */ |
| 3155 | node = ece_generic_processing(node); |
| 3156 | /* If all arguments are Consts, we can fold to a constant */ |
| 3157 | if (ece_all_arguments_const(node)) |
| 3158 | return ece_evaluate_expr(node); |
| 3159 | return node; |
| 3160 | } |
| 3161 | case T_CoalesceExpr: |
| 3162 | { |
| 3163 | CoalesceExpr *coalesceexpr = (CoalesceExpr *) node; |
| 3164 | CoalesceExpr *newcoalesce; |
| 3165 | List *newargs; |
| 3166 | ListCell *arg; |
| 3167 | |
| 3168 | newargs = NIL; |
| 3169 | foreach(arg, coalesceexpr->args) |
| 3170 | { |
| 3171 | Node *e; |
| 3172 | |
| 3173 | e = eval_const_expressions_mutator((Node *) lfirst(arg), |
| 3174 | context); |
| 3175 | |
| 3176 | /* |
| 3177 | * We can remove null constants from the list. For a |
| 3178 | * non-null constant, if it has not been preceded by any |
| 3179 | * other non-null-constant expressions then it is the |
| 3180 | * result. Otherwise, it's the next argument, but we can |
| 3181 | * drop following arguments since they will never be |
| 3182 | * reached. |
| 3183 | */ |
| 3184 | if (IsA(e, Const)) |
| 3185 | { |
| 3186 | if (((Const *) e)->constisnull) |
| 3187 | continue; /* drop null constant */ |
| 3188 | if (newargs == NIL) |
| 3189 | return e; /* first expr */ |
| 3190 | newargs = lappend(newargs, e); |
| 3191 | break; |
| 3192 | } |
| 3193 | newargs = lappend(newargs, e); |
| 3194 | } |
| 3195 | |
| 3196 | /* |
| 3197 | * If all the arguments were constant null, the result is just |
| 3198 | * null |
| 3199 | */ |
| 3200 | if (newargs == NIL) |
| 3201 | return (Node *) makeNullConst(coalesceexpr->coalescetype, |
| 3202 | -1, |
| 3203 | coalesceexpr->coalescecollid); |
| 3204 | |
| 3205 | newcoalesce = makeNode(CoalesceExpr); |
| 3206 | newcoalesce->coalescetype = coalesceexpr->coalescetype; |
| 3207 | newcoalesce->coalescecollid = coalesceexpr->coalescecollid; |
| 3208 | newcoalesce->args = newargs; |
| 3209 | newcoalesce->location = coalesceexpr->location; |
| 3210 | return (Node *) newcoalesce; |
| 3211 | } |
| 3212 | case T_SQLValueFunction: |
| 3213 | { |
| 3214 | /* |
| 3215 | * All variants of SQLValueFunction are stable, so if we are |
| 3216 | * estimating the expression's value, we should evaluate the |
| 3217 | * current function value. Otherwise just copy. |
| 3218 | */ |
| 3219 | SQLValueFunction *svf = (SQLValueFunction *) node; |
| 3220 | |
| 3221 | if (context->estimate) |
| 3222 | return (Node *) evaluate_expr((Expr *) svf, |
| 3223 | svf->type, |
| 3224 | svf->typmod, |
| 3225 | InvalidOid); |
| 3226 | else |
| 3227 | return copyObject((Node *) svf); |
| 3228 | } |
| 3229 | case T_FieldSelect: |
| 3230 | { |
| 3231 | /* |
| 3232 | * We can optimize field selection from a whole-row Var into a |
| 3233 | * simple Var. (This case won't be generated directly by the |
| 3234 | * parser, because ParseComplexProjection short-circuits it. |
| 3235 | * But it can arise while simplifying functions.) Also, we |
| 3236 | * can optimize field selection from a RowExpr construct, or |
| 3237 | * of course from a constant. |
| 3238 | * |
| 3239 | * However, replacing a whole-row Var in this way has a |
| 3240 | * pitfall: if we've already built the rel targetlist for the |
| 3241 | * source relation, then the whole-row Var is scheduled to be |
| 3242 | * produced by the relation scan, but the simple Var probably |
| 3243 | * isn't, which will lead to a failure in setrefs.c. This is |
| 3244 | * not a problem when handling simple single-level queries, in |
| 3245 | * which expression simplification always happens first. It |
| 3246 | * is a risk for lateral references from subqueries, though. |
| 3247 | * To avoid such failures, don't optimize uplevel references. |
| 3248 | * |
| 3249 | * We must also check that the declared type of the field is |
| 3250 | * still the same as when the FieldSelect was created --- this |
| 3251 | * can change if someone did ALTER COLUMN TYPE on the rowtype. |
| 3252 | * If it isn't, we skip the optimization; the case will |
| 3253 | * probably fail at runtime, but that's not our problem here. |
| 3254 | */ |
| 3255 | FieldSelect *fselect = (FieldSelect *) node; |
| 3256 | FieldSelect *newfselect; |
| 3257 | Node *arg; |
| 3258 | |
| 3259 | arg = eval_const_expressions_mutator((Node *) fselect->arg, |
| 3260 | context); |
| 3261 | if (arg && IsA(arg, Var) && |
| 3262 | ((Var *) arg)->varattno == InvalidAttrNumber && |
| 3263 | ((Var *) arg)->varlevelsup == 0) |
| 3264 | { |
| 3265 | if (rowtype_field_matches(((Var *) arg)->vartype, |
| 3266 | fselect->fieldnum, |
| 3267 | fselect->resulttype, |
| 3268 | fselect->resulttypmod, |
| 3269 | fselect->resultcollid)) |
| 3270 | return (Node *) makeVar(((Var *) arg)->varno, |
| 3271 | fselect->fieldnum, |
| 3272 | fselect->resulttype, |
| 3273 | fselect->resulttypmod, |
| 3274 | fselect->resultcollid, |
| 3275 | ((Var *) arg)->varlevelsup); |
| 3276 | } |
| 3277 | if (arg && IsA(arg, RowExpr)) |
| 3278 | { |
| 3279 | RowExpr *rowexpr = (RowExpr *) arg; |
| 3280 | |
| 3281 | if (fselect->fieldnum > 0 && |
| 3282 | fselect->fieldnum <= list_length(rowexpr->args)) |
| 3283 | { |
| 3284 | Node *fld = (Node *) list_nth(rowexpr->args, |
| 3285 | fselect->fieldnum - 1); |
| 3286 | |
| 3287 | if (rowtype_field_matches(rowexpr->row_typeid, |
| 3288 | fselect->fieldnum, |
| 3289 | fselect->resulttype, |
| 3290 | fselect->resulttypmod, |
| 3291 | fselect->resultcollid) && |
| 3292 | fselect->resulttype == exprType(fld) && |
| 3293 | fselect->resulttypmod == exprTypmod(fld) && |
| 3294 | fselect->resultcollid == exprCollation(fld)) |
| 3295 | return fld; |
| 3296 | } |
| 3297 | } |
| 3298 | newfselect = makeNode(FieldSelect); |
| 3299 | newfselect->arg = (Expr *) arg; |
| 3300 | newfselect->fieldnum = fselect->fieldnum; |
| 3301 | newfselect->resulttype = fselect->resulttype; |
| 3302 | newfselect->resulttypmod = fselect->resulttypmod; |
| 3303 | newfselect->resultcollid = fselect->resultcollid; |
| 3304 | if (arg && IsA(arg, Const)) |
| 3305 | { |
| 3306 | Const *con = (Const *) arg; |
| 3307 | |
| 3308 | if (rowtype_field_matches(con->consttype, |
| 3309 | newfselect->fieldnum, |
| 3310 | newfselect->resulttype, |
| 3311 | newfselect->resulttypmod, |
| 3312 | newfselect->resultcollid)) |
| 3313 | return ece_evaluate_expr(newfselect); |
| 3314 | } |
| 3315 | return (Node *) newfselect; |
| 3316 | } |
| 3317 | case T_NullTest: |
| 3318 | { |
| 3319 | NullTest *ntest = (NullTest *) node; |
| 3320 | NullTest *newntest; |
| 3321 | Node *arg; |
| 3322 | |
| 3323 | arg = eval_const_expressions_mutator((Node *) ntest->arg, |
| 3324 | context); |
| 3325 | if (ntest->argisrow && arg && IsA(arg, RowExpr)) |
| 3326 | { |
| 3327 | /* |
| 3328 | * We break ROW(...) IS [NOT] NULL into separate tests on |
| 3329 | * its component fields. This form is usually more |
| 3330 | * efficient to evaluate, as well as being more amenable |
| 3331 | * to optimization. |
| 3332 | */ |
| 3333 | RowExpr *rarg = (RowExpr *) arg; |
| 3334 | List *newargs = NIL; |
| 3335 | ListCell *l; |
| 3336 | |
| 3337 | foreach(l, rarg->args) |
| 3338 | { |
| 3339 | Node *relem = (Node *) lfirst(l); |
| 3340 | |
| 3341 | /* |
| 3342 | * A constant field refutes the whole NullTest if it's |
| 3343 | * of the wrong nullness; else we can discard it. |
| 3344 | */ |
| 3345 | if (relem && IsA(relem, Const)) |
| 3346 | { |
| 3347 | Const *carg = (Const *) relem; |
| 3348 | |
| 3349 | if (carg->constisnull ? |
| 3350 | (ntest->nulltesttype == IS_NOT_NULL) : |
| 3351 | (ntest->nulltesttype == IS_NULL)) |
| 3352 | return makeBoolConst(false, false); |
| 3353 | continue; |
| 3354 | } |
| 3355 | |
| 3356 | /* |
| 3357 | * Else, make a scalar (argisrow == false) NullTest |
| 3358 | * for this field. Scalar semantics are required |
| 3359 | * because IS [NOT] NULL doesn't recurse; see comments |
| 3360 | * in ExecEvalRowNullInt(). |
| 3361 | */ |
| 3362 | newntest = makeNode(NullTest); |
| 3363 | newntest->arg = (Expr *) relem; |
| 3364 | newntest->nulltesttype = ntest->nulltesttype; |
| 3365 | newntest->argisrow = false; |
| 3366 | newntest->location = ntest->location; |
| 3367 | newargs = lappend(newargs, newntest); |
| 3368 | } |
| 3369 | /* If all the inputs were constants, result is TRUE */ |
| 3370 | if (newargs == NIL) |
| 3371 | return makeBoolConst(true, false); |
| 3372 | /* If only one nonconst input, it's the result */ |
| 3373 | if (list_length(newargs) == 1) |
| 3374 | return (Node *) linitial(newargs); |
| 3375 | /* Else we need an AND node */ |
| 3376 | return (Node *) make_andclause(newargs); |
| 3377 | } |
| 3378 | if (!ntest->argisrow && arg && IsA(arg, Const)) |
| 3379 | { |
| 3380 | Const *carg = (Const *) arg; |
| 3381 | bool result; |
| 3382 | |
| 3383 | switch (ntest->nulltesttype) |
| 3384 | { |
| 3385 | case IS_NULL: |
| 3386 | result = carg->constisnull; |
| 3387 | break; |
| 3388 | case IS_NOT_NULL: |
| 3389 | result = !carg->constisnull; |
| 3390 | break; |
| 3391 | default: |
| 3392 | elog(ERROR, "unrecognized nulltesttype: %d" , |
| 3393 | (int) ntest->nulltesttype); |
| 3394 | result = false; /* keep compiler quiet */ |
| 3395 | break; |
| 3396 | } |
| 3397 | |
| 3398 | return makeBoolConst(result, false); |
| 3399 | } |
| 3400 | |
| 3401 | newntest = makeNode(NullTest); |
| 3402 | newntest->arg = (Expr *) arg; |
| 3403 | newntest->nulltesttype = ntest->nulltesttype; |
| 3404 | newntest->argisrow = ntest->argisrow; |
| 3405 | newntest->location = ntest->location; |
| 3406 | return (Node *) newntest; |
| 3407 | } |
| 3408 | case T_BooleanTest: |
| 3409 | { |
| 3410 | /* |
| 3411 | * This case could be folded into the generic handling used |
| 3412 | * for SubscriptingRef etc. But because the simplification |
| 3413 | * logic is so trivial, applying evaluate_expr() to perform it |
| 3414 | * would be a heavy overhead. BooleanTest is probably common |
| 3415 | * enough to justify keeping this bespoke implementation. |
| 3416 | */ |
| 3417 | BooleanTest *btest = (BooleanTest *) node; |
| 3418 | BooleanTest *newbtest; |
| 3419 | Node *arg; |
| 3420 | |
| 3421 | arg = eval_const_expressions_mutator((Node *) btest->arg, |
| 3422 | context); |
| 3423 | if (arg && IsA(arg, Const)) |
| 3424 | { |
| 3425 | Const *carg = (Const *) arg; |
| 3426 | bool result; |
| 3427 | |
| 3428 | switch (btest->booltesttype) |
| 3429 | { |
| 3430 | case IS_TRUE: |
| 3431 | result = (!carg->constisnull && |
| 3432 | DatumGetBool(carg->constvalue)); |
| 3433 | break; |
| 3434 | case IS_NOT_TRUE: |
| 3435 | result = (carg->constisnull || |
| 3436 | !DatumGetBool(carg->constvalue)); |
| 3437 | break; |
| 3438 | case IS_FALSE: |
| 3439 | result = (!carg->constisnull && |
| 3440 | !DatumGetBool(carg->constvalue)); |
| 3441 | break; |
| 3442 | case IS_NOT_FALSE: |
| 3443 | result = (carg->constisnull || |
| 3444 | DatumGetBool(carg->constvalue)); |
| 3445 | break; |
| 3446 | case IS_UNKNOWN: |
| 3447 | result = carg->constisnull; |
| 3448 | break; |
| 3449 | case IS_NOT_UNKNOWN: |
| 3450 | result = !carg->constisnull; |
| 3451 | break; |
| 3452 | default: |
| 3453 | elog(ERROR, "unrecognized booltesttype: %d" , |
| 3454 | (int) btest->booltesttype); |
| 3455 | result = false; /* keep compiler quiet */ |
| 3456 | break; |
| 3457 | } |
| 3458 | |
| 3459 | return makeBoolConst(result, false); |
| 3460 | } |
| 3461 | |
| 3462 | newbtest = makeNode(BooleanTest); |
| 3463 | newbtest->arg = (Expr *) arg; |
| 3464 | newbtest->booltesttype = btest->booltesttype; |
| 3465 | newbtest->location = btest->location; |
| 3466 | return (Node *) newbtest; |
| 3467 | } |
| 3468 | case T_CoerceToDomain: |
| 3469 | { |
| 3470 | /* |
| 3471 | * If the domain currently has no constraints, we replace the |
| 3472 | * CoerceToDomain node with a simple RelabelType, which is |
| 3473 | * both far faster to execute and more amenable to later |
| 3474 | * optimization. We must then mark the plan as needing to be |
| 3475 | * rebuilt if the domain's constraints change. |
| 3476 | * |
| 3477 | * Also, in estimation mode, always replace CoerceToDomain |
| 3478 | * nodes, effectively assuming that the coercion will succeed. |
| 3479 | */ |
| 3480 | CoerceToDomain *cdomain = (CoerceToDomain *) node; |
| 3481 | CoerceToDomain *newcdomain; |
| 3482 | Node *arg; |
| 3483 | |
| 3484 | arg = eval_const_expressions_mutator((Node *) cdomain->arg, |
| 3485 | context); |
| 3486 | if (context->estimate || |
| 3487 | !DomainHasConstraints(cdomain->resulttype)) |
| 3488 | { |
| 3489 | /* Record dependency, if this isn't estimation mode */ |
| 3490 | if (context->root && !context->estimate) |
| 3491 | record_plan_type_dependency(context->root, |
| 3492 | cdomain->resulttype); |
| 3493 | |
| 3494 | /* Generate RelabelType to substitute for CoerceToDomain */ |
| 3495 | /* This should match the RelabelType logic above */ |
| 3496 | |
| 3497 | while (arg && IsA(arg, RelabelType)) |
| 3498 | arg = (Node *) ((RelabelType *) arg)->arg; |
| 3499 | |
| 3500 | if (arg && IsA(arg, Const)) |
| 3501 | { |
| 3502 | Const *con = (Const *) arg; |
| 3503 | |
| 3504 | con->consttype = cdomain->resulttype; |
| 3505 | con->consttypmod = cdomain->resulttypmod; |
| 3506 | con->constcollid = cdomain->resultcollid; |
| 3507 | return (Node *) con; |
| 3508 | } |
| 3509 | else |
| 3510 | { |
| 3511 | RelabelType *newrelabel = makeNode(RelabelType); |
| 3512 | |
| 3513 | newrelabel->arg = (Expr *) arg; |
| 3514 | newrelabel->resulttype = cdomain->resulttype; |
| 3515 | newrelabel->resulttypmod = cdomain->resulttypmod; |
| 3516 | newrelabel->resultcollid = cdomain->resultcollid; |
| 3517 | newrelabel->relabelformat = cdomain->coercionformat; |
| 3518 | newrelabel->location = cdomain->location; |
| 3519 | return (Node *) newrelabel; |
| 3520 | } |
| 3521 | } |
| 3522 | |
| 3523 | newcdomain = makeNode(CoerceToDomain); |
| 3524 | newcdomain->arg = (Expr *) arg; |
| 3525 | newcdomain->resulttype = cdomain->resulttype; |
| 3526 | newcdomain->resulttypmod = cdomain->resulttypmod; |
| 3527 | newcdomain->resultcollid = cdomain->resultcollid; |
| 3528 | newcdomain->coercionformat = cdomain->coercionformat; |
| 3529 | newcdomain->location = cdomain->location; |
| 3530 | return (Node *) newcdomain; |
| 3531 | } |
| 3532 | case T_PlaceHolderVar: |
| 3533 | |
| 3534 | /* |
| 3535 | * In estimation mode, just strip the PlaceHolderVar node |
| 3536 | * altogether; this amounts to estimating that the contained value |
| 3537 | * won't be forced to null by an outer join. In regular mode we |
| 3538 | * just use the default behavior (ie, simplify the expression but |
| 3539 | * leave the PlaceHolderVar node intact). |
| 3540 | */ |
| 3541 | if (context->estimate) |
| 3542 | { |
| 3543 | PlaceHolderVar *phv = (PlaceHolderVar *) node; |
| 3544 | |
| 3545 | return eval_const_expressions_mutator((Node *) phv->phexpr, |
| 3546 | context); |
| 3547 | } |
| 3548 | break; |
| 3549 | case T_ConvertRowtypeExpr: |
| 3550 | { |
| 3551 | ConvertRowtypeExpr *cre = castNode(ConvertRowtypeExpr, node); |
| 3552 | Node *arg; |
| 3553 | ConvertRowtypeExpr *newcre; |
| 3554 | |
| 3555 | arg = eval_const_expressions_mutator((Node *) cre->arg, |
| 3556 | context); |
| 3557 | |
| 3558 | newcre = makeNode(ConvertRowtypeExpr); |
| 3559 | newcre->resulttype = cre->resulttype; |
| 3560 | newcre->convertformat = cre->convertformat; |
| 3561 | newcre->location = cre->location; |
| 3562 | |
| 3563 | /* |
| 3564 | * In case of a nested ConvertRowtypeExpr, we can convert the |
| 3565 | * leaf row directly to the topmost row format without any |
| 3566 | * intermediate conversions. (This works because |
| 3567 | * ConvertRowtypeExpr is used only for child->parent |
| 3568 | * conversion in inheritance trees, which works by exact match |
| 3569 | * of column name, and a column absent in an intermediate |
| 3570 | * result can't be present in the final result.) |
| 3571 | * |
| 3572 | * No need to check more than one level deep, because the |
| 3573 | * above recursion will have flattened anything else. |
| 3574 | */ |
| 3575 | if (arg != NULL && IsA(arg, ConvertRowtypeExpr)) |
| 3576 | { |
| 3577 | ConvertRowtypeExpr *argcre = (ConvertRowtypeExpr *) arg; |
| 3578 | |
| 3579 | arg = (Node *) argcre->arg; |
| 3580 | |
| 3581 | /* |
| 3582 | * Make sure an outer implicit conversion can't hide an |
| 3583 | * inner explicit one. |
| 3584 | */ |
| 3585 | if (newcre->convertformat == COERCE_IMPLICIT_CAST) |
| 3586 | newcre->convertformat = argcre->convertformat; |
| 3587 | } |
| 3588 | |
| 3589 | newcre->arg = (Expr *) arg; |
| 3590 | |
| 3591 | if (arg != NULL && IsA(arg, Const)) |
| 3592 | return ece_evaluate_expr((Node *) newcre); |
| 3593 | return (Node *) newcre; |
| 3594 | } |
| 3595 | default: |
| 3596 | break; |
| 3597 | } |
| 3598 | |
| 3599 | /* |
| 3600 | * For any node type not handled above, copy the node unchanged but |
| 3601 | * const-simplify its subexpressions. This is the correct thing for node |
| 3602 | * types whose behavior might change between planning and execution, such |
| 3603 | * as CurrentOfExpr. It's also a safe default for new node types not |
| 3604 | * known to this routine. |
| 3605 | */ |
| 3606 | return ece_generic_processing(node); |
| 3607 | } |
| 3608 | |
| 3609 | /* |
| 3610 | * Subroutine for eval_const_expressions: check for non-Const nodes. |
| 3611 | * |
| 3612 | * We can abort recursion immediately on finding a non-Const node. This is |
| 3613 | * critical for performance, else eval_const_expressions_mutator would take |
| 3614 | * O(N^2) time on non-simplifiable trees. However, we do need to descend |
| 3615 | * into List nodes since expression_tree_walker sometimes invokes the walker |
| 3616 | * function directly on List subtrees. |
| 3617 | */ |
| 3618 | static bool |
| 3619 | contain_non_const_walker(Node *node, void *context) |
| 3620 | { |
| 3621 | if (node == NULL) |
| 3622 | return false; |
| 3623 | if (IsA(node, Const)) |
| 3624 | return false; |
| 3625 | if (IsA(node, List)) |
| 3626 | return expression_tree_walker(node, contain_non_const_walker, context); |
| 3627 | /* Otherwise, abort the tree traversal and return true */ |
| 3628 | return true; |
| 3629 | } |
| 3630 | |
| 3631 | /* |
| 3632 | * Subroutine for eval_const_expressions: check if a function is OK to evaluate |
| 3633 | */ |
| 3634 | static bool |
| 3635 | ece_function_is_safe(Oid funcid, eval_const_expressions_context *context) |
| 3636 | { |
| 3637 | char provolatile = func_volatile(funcid); |
| 3638 | |
| 3639 | /* |
| 3640 | * Ordinarily we are only allowed to simplify immutable functions. But for |
| 3641 | * purposes of estimation, we consider it okay to simplify functions that |
| 3642 | * are merely stable; the risk that the result might change from planning |
| 3643 | * time to execution time is worth taking in preference to not being able |
| 3644 | * to estimate the value at all. |
| 3645 | */ |
| 3646 | if (provolatile == PROVOLATILE_IMMUTABLE) |
| 3647 | return true; |
| 3648 | if (context->estimate && provolatile == PROVOLATILE_STABLE) |
| 3649 | return true; |
| 3650 | return false; |
| 3651 | } |
| 3652 | |
| 3653 | /* |
| 3654 | * Subroutine for eval_const_expressions: process arguments of an OR clause |
| 3655 | * |
| 3656 | * This includes flattening of nested ORs as well as recursion to |
| 3657 | * eval_const_expressions to simplify the OR arguments. |
| 3658 | * |
| 3659 | * After simplification, OR arguments are handled as follows: |
| 3660 | * non constant: keep |
| 3661 | * FALSE: drop (does not affect result) |
| 3662 | * TRUE: force result to TRUE |
| 3663 | * NULL: keep only one |
| 3664 | * We must keep one NULL input because OR expressions evaluate to NULL when no |
| 3665 | * input is TRUE and at least one is NULL. We don't actually include the NULL |
| 3666 | * here, that's supposed to be done by the caller. |
| 3667 | * |
| 3668 | * The output arguments *haveNull and *forceTrue must be initialized false |
| 3669 | * by the caller. They will be set true if a NULL constant or TRUE constant, |
| 3670 | * respectively, is detected anywhere in the argument list. |
| 3671 | */ |
| 3672 | static List * |
| 3673 | simplify_or_arguments(List *args, |
| 3674 | eval_const_expressions_context *context, |
| 3675 | bool *haveNull, bool *forceTrue) |
| 3676 | { |
| 3677 | List *newargs = NIL; |
| 3678 | List *unprocessed_args; |
| 3679 | |
| 3680 | /* |
| 3681 | * We want to ensure that any OR immediately beneath another OR gets |
| 3682 | * flattened into a single OR-list, so as to simplify later reasoning. |
| 3683 | * |
| 3684 | * To avoid stack overflow from recursion of eval_const_expressions, we |
| 3685 | * resort to some tenseness here: we keep a list of not-yet-processed |
| 3686 | * inputs, and handle flattening of nested ORs by prepending to the to-do |
| 3687 | * list instead of recursing. Now that the parser generates N-argument |
| 3688 | * ORs from simple lists, this complexity is probably less necessary than |
| 3689 | * it once was, but we might as well keep the logic. |
| 3690 | */ |
| 3691 | unprocessed_args = list_copy(args); |
| 3692 | while (unprocessed_args) |
| 3693 | { |
| 3694 | Node *arg = (Node *) linitial(unprocessed_args); |
| 3695 | |
| 3696 | unprocessed_args = list_delete_first(unprocessed_args); |
| 3697 | |
| 3698 | /* flatten nested ORs as per above comment */ |
| 3699 | if (is_orclause(arg)) |
| 3700 | { |
| 3701 | List *subargs = list_copy(((BoolExpr *) arg)->args); |
| 3702 | |
| 3703 | /* overly tense code to avoid leaking unused list header */ |
| 3704 | if (!unprocessed_args) |
| 3705 | unprocessed_args = subargs; |
| 3706 | else |
| 3707 | { |
| 3708 | List *oldhdr = unprocessed_args; |
| 3709 | |
| 3710 | unprocessed_args = list_concat(subargs, unprocessed_args); |
| 3711 | pfree(oldhdr); |
| 3712 | } |
| 3713 | continue; |
| 3714 | } |
| 3715 | |
| 3716 | /* If it's not an OR, simplify it */ |
| 3717 | arg = eval_const_expressions_mutator(arg, context); |
| 3718 | |
| 3719 | /* |
| 3720 | * It is unlikely but not impossible for simplification of a non-OR |
| 3721 | * clause to produce an OR. Recheck, but don't be too tense about it |
| 3722 | * since it's not a mainstream case. In particular we don't worry |
| 3723 | * about const-simplifying the input twice. |
| 3724 | */ |
| 3725 | if (is_orclause(arg)) |
| 3726 | { |
| 3727 | List *subargs = list_copy(((BoolExpr *) arg)->args); |
| 3728 | |
| 3729 | unprocessed_args = list_concat(subargs, unprocessed_args); |
| 3730 | continue; |
| 3731 | } |
| 3732 | |
| 3733 | /* |
| 3734 | * OK, we have a const-simplified non-OR argument. Process it per |
| 3735 | * comments above. |
| 3736 | */ |
| 3737 | if (IsA(arg, Const)) |
| 3738 | { |
| 3739 | Const *const_input = (Const *) arg; |
| 3740 | |
| 3741 | if (const_input->constisnull) |
| 3742 | *haveNull = true; |
| 3743 | else if (DatumGetBool(const_input->constvalue)) |
| 3744 | { |
| 3745 | *forceTrue = true; |
| 3746 | |
| 3747 | /* |
| 3748 | * Once we detect a TRUE result we can just exit the loop |
| 3749 | * immediately. However, if we ever add a notion of |
| 3750 | * non-removable functions, we'd need to keep scanning. |
| 3751 | */ |
| 3752 | return NIL; |
| 3753 | } |
| 3754 | /* otherwise, we can drop the constant-false input */ |
| 3755 | continue; |
| 3756 | } |
| 3757 | |
| 3758 | /* else emit the simplified arg into the result list */ |
| 3759 | newargs = lappend(newargs, arg); |
| 3760 | } |
| 3761 | |
| 3762 | return newargs; |
| 3763 | } |
| 3764 | |
| 3765 | /* |
| 3766 | * Subroutine for eval_const_expressions: process arguments of an AND clause |
| 3767 | * |
| 3768 | * This includes flattening of nested ANDs as well as recursion to |
| 3769 | * eval_const_expressions to simplify the AND arguments. |
| 3770 | * |
| 3771 | * After simplification, AND arguments are handled as follows: |
| 3772 | * non constant: keep |
| 3773 | * TRUE: drop (does not affect result) |
| 3774 | * FALSE: force result to FALSE |
| 3775 | * NULL: keep only one |
| 3776 | * We must keep one NULL input because AND expressions evaluate to NULL when |
| 3777 | * no input is FALSE and at least one is NULL. We don't actually include the |
| 3778 | * NULL here, that's supposed to be done by the caller. |
| 3779 | * |
| 3780 | * The output arguments *haveNull and *forceFalse must be initialized false |
| 3781 | * by the caller. They will be set true if a null constant or false constant, |
| 3782 | * respectively, is detected anywhere in the argument list. |
| 3783 | */ |
| 3784 | static List * |
| 3785 | simplify_and_arguments(List *args, |
| 3786 | eval_const_expressions_context *context, |
| 3787 | bool *haveNull, bool *forceFalse) |
| 3788 | { |
| 3789 | List *newargs = NIL; |
| 3790 | List *unprocessed_args; |
| 3791 | |
| 3792 | /* See comments in simplify_or_arguments */ |
| 3793 | unprocessed_args = list_copy(args); |
| 3794 | while (unprocessed_args) |
| 3795 | { |
| 3796 | Node *arg = (Node *) linitial(unprocessed_args); |
| 3797 | |
| 3798 | unprocessed_args = list_delete_first(unprocessed_args); |
| 3799 | |
| 3800 | /* flatten nested ANDs as per above comment */ |
| 3801 | if (is_andclause(arg)) |
| 3802 | { |
| 3803 | List *subargs = list_copy(((BoolExpr *) arg)->args); |
| 3804 | |
| 3805 | /* overly tense code to avoid leaking unused list header */ |
| 3806 | if (!unprocessed_args) |
| 3807 | unprocessed_args = subargs; |
| 3808 | else |
| 3809 | { |
| 3810 | List *oldhdr = unprocessed_args; |
| 3811 | |
| 3812 | unprocessed_args = list_concat(subargs, unprocessed_args); |
| 3813 | pfree(oldhdr); |
| 3814 | } |
| 3815 | continue; |
| 3816 | } |
| 3817 | |
| 3818 | /* If it's not an AND, simplify it */ |
| 3819 | arg = eval_const_expressions_mutator(arg, context); |
| 3820 | |
| 3821 | /* |
| 3822 | * It is unlikely but not impossible for simplification of a non-AND |
| 3823 | * clause to produce an AND. Recheck, but don't be too tense about it |
| 3824 | * since it's not a mainstream case. In particular we don't worry |
| 3825 | * about const-simplifying the input twice. |
| 3826 | */ |
| 3827 | if (is_andclause(arg)) |
| 3828 | { |
| 3829 | List *subargs = list_copy(((BoolExpr *) arg)->args); |
| 3830 | |
| 3831 | unprocessed_args = list_concat(subargs, unprocessed_args); |
| 3832 | continue; |
| 3833 | } |
| 3834 | |
| 3835 | /* |
| 3836 | * OK, we have a const-simplified non-AND argument. Process it per |
| 3837 | * comments above. |
| 3838 | */ |
| 3839 | if (IsA(arg, Const)) |
| 3840 | { |
| 3841 | Const *const_input = (Const *) arg; |
| 3842 | |
| 3843 | if (const_input->constisnull) |
| 3844 | *haveNull = true; |
| 3845 | else if (!DatumGetBool(const_input->constvalue)) |
| 3846 | { |
| 3847 | *forceFalse = true; |
| 3848 | |
| 3849 | /* |
| 3850 | * Once we detect a FALSE result we can just exit the loop |
| 3851 | * immediately. However, if we ever add a notion of |
| 3852 | * non-removable functions, we'd need to keep scanning. |
| 3853 | */ |
| 3854 | return NIL; |
| 3855 | } |
| 3856 | /* otherwise, we can drop the constant-true input */ |
| 3857 | continue; |
| 3858 | } |
| 3859 | |
| 3860 | /* else emit the simplified arg into the result list */ |
| 3861 | newargs = lappend(newargs, arg); |
| 3862 | } |
| 3863 | |
| 3864 | return newargs; |
| 3865 | } |
| 3866 | |
| 3867 | /* |
| 3868 | * Subroutine for eval_const_expressions: try to simplify boolean equality |
| 3869 | * or inequality condition |
| 3870 | * |
| 3871 | * Inputs are the operator OID and the simplified arguments to the operator. |
| 3872 | * Returns a simplified expression if successful, or NULL if cannot |
| 3873 | * simplify the expression. |
| 3874 | * |
| 3875 | * The idea here is to reduce "x = true" to "x" and "x = false" to "NOT x", |
| 3876 | * or similarly "x <> true" to "NOT x" and "x <> false" to "x". |
| 3877 | * This is only marginally useful in itself, but doing it in constant folding |
| 3878 | * ensures that we will recognize these forms as being equivalent in, for |
| 3879 | * example, partial index matching. |
| 3880 | * |
| 3881 | * We come here only if simplify_function has failed; therefore we cannot |
| 3882 | * see two constant inputs, nor a constant-NULL input. |
| 3883 | */ |
| 3884 | static Node * |
| 3885 | simplify_boolean_equality(Oid opno, List *args) |
| 3886 | { |
| 3887 | Node *leftop; |
| 3888 | Node *rightop; |
| 3889 | |
| 3890 | Assert(list_length(args) == 2); |
| 3891 | leftop = linitial(args); |
| 3892 | rightop = lsecond(args); |
| 3893 | if (leftop && IsA(leftop, Const)) |
| 3894 | { |
| 3895 | Assert(!((Const *) leftop)->constisnull); |
| 3896 | if (opno == BooleanEqualOperator) |
| 3897 | { |
| 3898 | if (DatumGetBool(((Const *) leftop)->constvalue)) |
| 3899 | return rightop; /* true = foo */ |
| 3900 | else |
| 3901 | return negate_clause(rightop); /* false = foo */ |
| 3902 | } |
| 3903 | else |
| 3904 | { |
| 3905 | if (DatumGetBool(((Const *) leftop)->constvalue)) |
| 3906 | return negate_clause(rightop); /* true <> foo */ |
| 3907 | else |
| 3908 | return rightop; /* false <> foo */ |
| 3909 | } |
| 3910 | } |
| 3911 | if (rightop && IsA(rightop, Const)) |
| 3912 | { |
| 3913 | Assert(!((Const *) rightop)->constisnull); |
| 3914 | if (opno == BooleanEqualOperator) |
| 3915 | { |
| 3916 | if (DatumGetBool(((Const *) rightop)->constvalue)) |
| 3917 | return leftop; /* foo = true */ |
| 3918 | else |
| 3919 | return negate_clause(leftop); /* foo = false */ |
| 3920 | } |
| 3921 | else |
| 3922 | { |
| 3923 | if (DatumGetBool(((Const *) rightop)->constvalue)) |
| 3924 | return negate_clause(leftop); /* foo <> true */ |
| 3925 | else |
| 3926 | return leftop; /* foo <> false */ |
| 3927 | } |
| 3928 | } |
| 3929 | return NULL; |
| 3930 | } |
| 3931 | |
| 3932 | /* |
| 3933 | * Subroutine for eval_const_expressions: try to simplify a function call |
| 3934 | * (which might originally have been an operator; we don't care) |
| 3935 | * |
| 3936 | * Inputs are the function OID, actual result type OID (which is needed for |
| 3937 | * polymorphic functions), result typmod, result collation, the input |
| 3938 | * collation to use for the function, the original argument list (not |
| 3939 | * const-simplified yet, unless process_args is false), and some flags; |
| 3940 | * also the context data for eval_const_expressions. |
| 3941 | * |
| 3942 | * Returns a simplified expression if successful, or NULL if cannot |
| 3943 | * simplify the function call. |
| 3944 | * |
| 3945 | * This function is also responsible for converting named-notation argument |
| 3946 | * lists into positional notation and/or adding any needed default argument |
| 3947 | * expressions; which is a bit grotty, but it avoids extra fetches of the |
| 3948 | * function's pg_proc tuple. For this reason, the args list is |
| 3949 | * pass-by-reference. Conversion and const-simplification of the args list |
| 3950 | * will be done even if simplification of the function call itself is not |
| 3951 | * possible. |
| 3952 | */ |
| 3953 | static Expr * |
| 3954 | simplify_function(Oid funcid, Oid result_type, int32 result_typmod, |
| 3955 | Oid result_collid, Oid input_collid, List **args_p, |
| 3956 | bool funcvariadic, bool process_args, bool allow_non_const, |
| 3957 | eval_const_expressions_context *context) |
| 3958 | { |
| 3959 | List *args = *args_p; |
| 3960 | HeapTuple func_tuple; |
| 3961 | Form_pg_proc func_form; |
| 3962 | Expr *newexpr; |
| 3963 | |
| 3964 | /* |
| 3965 | * We have three strategies for simplification: execute the function to |
| 3966 | * deliver a constant result, use a transform function to generate a |
| 3967 | * substitute node tree, or expand in-line the body of the function |
| 3968 | * definition (which only works for simple SQL-language functions, but |
| 3969 | * that is a common case). Each case needs access to the function's |
| 3970 | * pg_proc tuple, so fetch it just once. |
| 3971 | * |
| 3972 | * Note: the allow_non_const flag suppresses both the second and third |
| 3973 | * strategies; so if !allow_non_const, simplify_function can only return a |
| 3974 | * Const or NULL. Argument-list rewriting happens anyway, though. |
| 3975 | */ |
| 3976 | func_tuple = SearchSysCache1(PROCOID, ObjectIdGetDatum(funcid)); |
| 3977 | if (!HeapTupleIsValid(func_tuple)) |
| 3978 | elog(ERROR, "cache lookup failed for function %u" , funcid); |
| 3979 | func_form = (Form_pg_proc) GETSTRUCT(func_tuple); |
| 3980 | |
| 3981 | /* |
| 3982 | * Process the function arguments, unless the caller did it already. |
| 3983 | * |
| 3984 | * Here we must deal with named or defaulted arguments, and then |
| 3985 | * recursively apply eval_const_expressions to the whole argument list. |
| 3986 | */ |
| 3987 | if (process_args) |
| 3988 | { |
| 3989 | args = expand_function_arguments(args, result_type, func_tuple); |
| 3990 | args = (List *) expression_tree_mutator((Node *) args, |
| 3991 | eval_const_expressions_mutator, |
| 3992 | (void *) context); |
| 3993 | /* Argument processing done, give it back to the caller */ |
| 3994 | *args_p = args; |
| 3995 | } |
| 3996 | |
| 3997 | /* Now attempt simplification of the function call proper. */ |
| 3998 | |
| 3999 | newexpr = evaluate_function(funcid, result_type, result_typmod, |
| 4000 | result_collid, input_collid, |
| 4001 | args, funcvariadic, |
| 4002 | func_tuple, context); |
| 4003 | |
| 4004 | if (!newexpr && allow_non_const && OidIsValid(func_form->prosupport)) |
| 4005 | { |
| 4006 | /* |
| 4007 | * Build a SupportRequestSimplify node to pass to the support |
| 4008 | * function, pointing to a dummy FuncExpr node containing the |
| 4009 | * simplified arg list. We use this approach to present a uniform |
| 4010 | * interface to the support function regardless of how the target |
| 4011 | * function is actually being invoked. |
| 4012 | */ |
| 4013 | SupportRequestSimplify req; |
| 4014 | FuncExpr fexpr; |
| 4015 | |
| 4016 | fexpr.xpr.type = T_FuncExpr; |
| 4017 | fexpr.funcid = funcid; |
| 4018 | fexpr.funcresulttype = result_type; |
| 4019 | fexpr.funcretset = func_form->proretset; |
| 4020 | fexpr.funcvariadic = funcvariadic; |
| 4021 | fexpr.funcformat = COERCE_EXPLICIT_CALL; |
| 4022 | fexpr.funccollid = result_collid; |
| 4023 | fexpr.inputcollid = input_collid; |
| 4024 | fexpr.args = args; |
| 4025 | fexpr.location = -1; |
| 4026 | |
| 4027 | req.type = T_SupportRequestSimplify; |
| 4028 | req.root = context->root; |
| 4029 | req.fcall = &fexpr; |
| 4030 | |
| 4031 | newexpr = (Expr *) |
| 4032 | DatumGetPointer(OidFunctionCall1(func_form->prosupport, |
| 4033 | PointerGetDatum(&req))); |
| 4034 | |
| 4035 | /* catch a possible API misunderstanding */ |
| 4036 | Assert(newexpr != (Expr *) &fexpr); |
| 4037 | } |
| 4038 | |
| 4039 | if (!newexpr && allow_non_const) |
| 4040 | newexpr = inline_function(funcid, result_type, result_collid, |
| 4041 | input_collid, args, funcvariadic, |
| 4042 | func_tuple, context); |
| 4043 | |
| 4044 | ReleaseSysCache(func_tuple); |
| 4045 | |
| 4046 | return newexpr; |
| 4047 | } |
| 4048 | |
| 4049 | /* |
| 4050 | * expand_function_arguments: convert named-notation args to positional args |
| 4051 | * and/or insert default args, as needed |
| 4052 | * |
| 4053 | * If we need to change anything, the input argument list is copied, not |
| 4054 | * modified. |
| 4055 | * |
| 4056 | * Note: this gets applied to operator argument lists too, even though the |
| 4057 | * cases it handles should never occur there. This should be OK since it |
| 4058 | * will fall through very quickly if there's nothing to do. |
| 4059 | */ |
| 4060 | List * |
| 4061 | expand_function_arguments(List *args, Oid result_type, HeapTuple func_tuple) |
| 4062 | { |
| 4063 | Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple); |
| 4064 | bool has_named_args = false; |
| 4065 | ListCell *lc; |
| 4066 | |
| 4067 | /* Do we have any named arguments? */ |
| 4068 | foreach(lc, args) |
| 4069 | { |
| 4070 | Node *arg = (Node *) lfirst(lc); |
| 4071 | |
| 4072 | if (IsA(arg, NamedArgExpr)) |
| 4073 | { |
| 4074 | has_named_args = true; |
| 4075 | break; |
| 4076 | } |
| 4077 | } |
| 4078 | |
| 4079 | /* If so, we must apply reorder_function_arguments */ |
| 4080 | if (has_named_args) |
| 4081 | { |
| 4082 | args = reorder_function_arguments(args, func_tuple); |
| 4083 | /* Recheck argument types and add casts if needed */ |
| 4084 | recheck_cast_function_args(args, result_type, func_tuple); |
| 4085 | } |
| 4086 | else if (list_length(args) < funcform->pronargs) |
| 4087 | { |
| 4088 | /* No named args, but we seem to be short some defaults */ |
| 4089 | args = add_function_defaults(args, func_tuple); |
| 4090 | /* Recheck argument types and add casts if needed */ |
| 4091 | recheck_cast_function_args(args, result_type, func_tuple); |
| 4092 | } |
| 4093 | |
| 4094 | return args; |
| 4095 | } |
| 4096 | |
| 4097 | /* |
| 4098 | * reorder_function_arguments: convert named-notation args to positional args |
| 4099 | * |
| 4100 | * This function also inserts default argument values as needed, since it's |
| 4101 | * impossible to form a truly valid positional call without that. |
| 4102 | */ |
| 4103 | static List * |
| 4104 | reorder_function_arguments(List *args, HeapTuple func_tuple) |
| 4105 | { |
| 4106 | Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple); |
| 4107 | int pronargs = funcform->pronargs; |
| 4108 | int nargsprovided = list_length(args); |
| 4109 | Node *argarray[FUNC_MAX_ARGS]; |
| 4110 | ListCell *lc; |
| 4111 | int i; |
| 4112 | |
| 4113 | Assert(nargsprovided <= pronargs); |
| 4114 | if (pronargs > FUNC_MAX_ARGS) |
| 4115 | elog(ERROR, "too many function arguments" ); |
| 4116 | MemSet(argarray, 0, pronargs * sizeof(Node *)); |
| 4117 | |
| 4118 | /* Deconstruct the argument list into an array indexed by argnumber */ |
| 4119 | i = 0; |
| 4120 | foreach(lc, args) |
| 4121 | { |
| 4122 | Node *arg = (Node *) lfirst(lc); |
| 4123 | |
| 4124 | if (!IsA(arg, NamedArgExpr)) |
| 4125 | { |
| 4126 | /* positional argument, assumed to precede all named args */ |
| 4127 | Assert(argarray[i] == NULL); |
| 4128 | argarray[i++] = arg; |
| 4129 | } |
| 4130 | else |
| 4131 | { |
| 4132 | NamedArgExpr *na = (NamedArgExpr *) arg; |
| 4133 | |
| 4134 | Assert(argarray[na->argnumber] == NULL); |
| 4135 | argarray[na->argnumber] = (Node *) na->arg; |
| 4136 | } |
| 4137 | } |
| 4138 | |
| 4139 | /* |
| 4140 | * Fetch default expressions, if needed, and insert into array at proper |
| 4141 | * locations (they aren't necessarily consecutive or all used) |
| 4142 | */ |
| 4143 | if (nargsprovided < pronargs) |
| 4144 | { |
| 4145 | List *defaults = fetch_function_defaults(func_tuple); |
| 4146 | |
| 4147 | i = pronargs - funcform->pronargdefaults; |
| 4148 | foreach(lc, defaults) |
| 4149 | { |
| 4150 | if (argarray[i] == NULL) |
| 4151 | argarray[i] = (Node *) lfirst(lc); |
| 4152 | i++; |
| 4153 | } |
| 4154 | } |
| 4155 | |
| 4156 | /* Now reconstruct the args list in proper order */ |
| 4157 | args = NIL; |
| 4158 | for (i = 0; i < pronargs; i++) |
| 4159 | { |
| 4160 | Assert(argarray[i] != NULL); |
| 4161 | args = lappend(args, argarray[i]); |
| 4162 | } |
| 4163 | |
| 4164 | return args; |
| 4165 | } |
| 4166 | |
| 4167 | /* |
| 4168 | * add_function_defaults: add missing function arguments from its defaults |
| 4169 | * |
| 4170 | * This is used only when the argument list was positional to begin with, |
| 4171 | * and so we know we just need to add defaults at the end. |
| 4172 | */ |
| 4173 | static List * |
| 4174 | add_function_defaults(List *args, HeapTuple func_tuple) |
| 4175 | { |
| 4176 | Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple); |
| 4177 | int nargsprovided = list_length(args); |
| 4178 | List *defaults; |
| 4179 | int ndelete; |
| 4180 | |
| 4181 | /* Get all the default expressions from the pg_proc tuple */ |
| 4182 | defaults = fetch_function_defaults(func_tuple); |
| 4183 | |
| 4184 | /* Delete any unused defaults from the list */ |
| 4185 | ndelete = nargsprovided + list_length(defaults) - funcform->pronargs; |
| 4186 | if (ndelete < 0) |
| 4187 | elog(ERROR, "not enough default arguments" ); |
| 4188 | while (ndelete-- > 0) |
| 4189 | defaults = list_delete_first(defaults); |
| 4190 | |
| 4191 | /* And form the combined argument list, not modifying the input list */ |
| 4192 | return list_concat(list_copy(args), defaults); |
| 4193 | } |
| 4194 | |
| 4195 | /* |
| 4196 | * fetch_function_defaults: get function's default arguments as expression list |
| 4197 | */ |
| 4198 | static List * |
| 4199 | fetch_function_defaults(HeapTuple func_tuple) |
| 4200 | { |
| 4201 | List *defaults; |
| 4202 | Datum proargdefaults; |
| 4203 | bool isnull; |
| 4204 | char *str; |
| 4205 | |
| 4206 | /* The error cases here shouldn't happen, but check anyway */ |
| 4207 | proargdefaults = SysCacheGetAttr(PROCOID, func_tuple, |
| 4208 | Anum_pg_proc_proargdefaults, |
| 4209 | &isnull); |
| 4210 | if (isnull) |
| 4211 | elog(ERROR, "not enough default arguments" ); |
| 4212 | str = TextDatumGetCString(proargdefaults); |
| 4213 | defaults = castNode(List, stringToNode(str)); |
| 4214 | pfree(str); |
| 4215 | return defaults; |
| 4216 | } |
| 4217 | |
| 4218 | /* |
| 4219 | * recheck_cast_function_args: recheck function args and typecast as needed |
| 4220 | * after adding defaults. |
| 4221 | * |
| 4222 | * It is possible for some of the defaulted arguments to be polymorphic; |
| 4223 | * therefore we can't assume that the default expressions have the correct |
| 4224 | * data types already. We have to re-resolve polymorphics and do coercion |
| 4225 | * just like the parser did. |
| 4226 | * |
| 4227 | * This should be a no-op if there are no polymorphic arguments, |
| 4228 | * but we do it anyway to be sure. |
| 4229 | * |
| 4230 | * Note: if any casts are needed, the args list is modified in-place; |
| 4231 | * caller should have already copied the list structure. |
| 4232 | */ |
| 4233 | static void |
| 4234 | recheck_cast_function_args(List *args, Oid result_type, HeapTuple func_tuple) |
| 4235 | { |
| 4236 | Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple); |
| 4237 | int nargs; |
| 4238 | Oid actual_arg_types[FUNC_MAX_ARGS]; |
| 4239 | Oid declared_arg_types[FUNC_MAX_ARGS]; |
| 4240 | Oid rettype; |
| 4241 | ListCell *lc; |
| 4242 | |
| 4243 | if (list_length(args) > FUNC_MAX_ARGS) |
| 4244 | elog(ERROR, "too many function arguments" ); |
| 4245 | nargs = 0; |
| 4246 | foreach(lc, args) |
| 4247 | { |
| 4248 | actual_arg_types[nargs++] = exprType((Node *) lfirst(lc)); |
| 4249 | } |
| 4250 | Assert(nargs == funcform->pronargs); |
| 4251 | memcpy(declared_arg_types, funcform->proargtypes.values, |
| 4252 | funcform->pronargs * sizeof(Oid)); |
| 4253 | rettype = enforce_generic_type_consistency(actual_arg_types, |
| 4254 | declared_arg_types, |
| 4255 | nargs, |
| 4256 | funcform->prorettype, |
| 4257 | false); |
| 4258 | /* let's just check we got the same answer as the parser did ... */ |
| 4259 | if (rettype != result_type) |
| 4260 | elog(ERROR, "function's resolved result type changed during planning" ); |
| 4261 | |
| 4262 | /* perform any necessary typecasting of arguments */ |
| 4263 | make_fn_arguments(NULL, args, actual_arg_types, declared_arg_types); |
| 4264 | } |
| 4265 | |
| 4266 | /* |
| 4267 | * evaluate_function: try to pre-evaluate a function call |
| 4268 | * |
| 4269 | * We can do this if the function is strict and has any constant-null inputs |
| 4270 | * (just return a null constant), or if the function is immutable and has all |
| 4271 | * constant inputs (call it and return the result as a Const node). In |
| 4272 | * estimation mode we are willing to pre-evaluate stable functions too. |
| 4273 | * |
| 4274 | * Returns a simplified expression if successful, or NULL if cannot |
| 4275 | * simplify the function. |
| 4276 | */ |
| 4277 | static Expr * |
| 4278 | evaluate_function(Oid funcid, Oid result_type, int32 result_typmod, |
| 4279 | Oid result_collid, Oid input_collid, List *args, |
| 4280 | bool funcvariadic, |
| 4281 | HeapTuple func_tuple, |
| 4282 | eval_const_expressions_context *context) |
| 4283 | { |
| 4284 | Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple); |
| 4285 | bool has_nonconst_input = false; |
| 4286 | bool has_null_input = false; |
| 4287 | ListCell *arg; |
| 4288 | FuncExpr *newexpr; |
| 4289 | |
| 4290 | /* |
| 4291 | * Can't simplify if it returns a set. |
| 4292 | */ |
| 4293 | if (funcform->proretset) |
| 4294 | return NULL; |
| 4295 | |
| 4296 | /* |
| 4297 | * Can't simplify if it returns RECORD. The immediate problem is that it |
| 4298 | * will be needing an expected tupdesc which we can't supply here. |
| 4299 | * |
| 4300 | * In the case where it has OUT parameters, it could get by without an |
| 4301 | * expected tupdesc, but we still have issues: get_expr_result_type() |
| 4302 | * doesn't know how to extract type info from a RECORD constant, and in |
| 4303 | * the case of a NULL function result there doesn't seem to be any clean |
| 4304 | * way to fix that. In view of the likelihood of there being still other |
| 4305 | * gotchas, seems best to leave the function call unreduced. |
| 4306 | */ |
| 4307 | if (funcform->prorettype == RECORDOID) |
| 4308 | return NULL; |
| 4309 | |
| 4310 | /* |
| 4311 | * Check for constant inputs and especially constant-NULL inputs. |
| 4312 | */ |
| 4313 | foreach(arg, args) |
| 4314 | { |
| 4315 | if (IsA(lfirst(arg), Const)) |
| 4316 | has_null_input |= ((Const *) lfirst(arg))->constisnull; |
| 4317 | else |
| 4318 | has_nonconst_input = true; |
| 4319 | } |
| 4320 | |
| 4321 | /* |
| 4322 | * If the function is strict and has a constant-NULL input, it will never |
| 4323 | * be called at all, so we can replace the call by a NULL constant, even |
| 4324 | * if there are other inputs that aren't constant, and even if the |
| 4325 | * function is not otherwise immutable. |
| 4326 | */ |
| 4327 | if (funcform->proisstrict && has_null_input) |
| 4328 | return (Expr *) makeNullConst(result_type, result_typmod, |
| 4329 | result_collid); |
| 4330 | |
| 4331 | /* |
| 4332 | * Otherwise, can simplify only if all inputs are constants. (For a |
| 4333 | * non-strict function, constant NULL inputs are treated the same as |
| 4334 | * constant non-NULL inputs.) |
| 4335 | */ |
| 4336 | if (has_nonconst_input) |
| 4337 | return NULL; |
| 4338 | |
| 4339 | /* |
| 4340 | * Ordinarily we are only allowed to simplify immutable functions. But for |
| 4341 | * purposes of estimation, we consider it okay to simplify functions that |
| 4342 | * are merely stable; the risk that the result might change from planning |
| 4343 | * time to execution time is worth taking in preference to not being able |
| 4344 | * to estimate the value at all. |
| 4345 | */ |
| 4346 | if (funcform->provolatile == PROVOLATILE_IMMUTABLE) |
| 4347 | /* okay */ ; |
| 4348 | else if (context->estimate && funcform->provolatile == PROVOLATILE_STABLE) |
| 4349 | /* okay */ ; |
| 4350 | else |
| 4351 | return NULL; |
| 4352 | |
| 4353 | /* |
| 4354 | * OK, looks like we can simplify this operator/function. |
| 4355 | * |
| 4356 | * Build a new FuncExpr node containing the already-simplified arguments. |
| 4357 | */ |
| 4358 | newexpr = makeNode(FuncExpr); |
| 4359 | newexpr->funcid = funcid; |
| 4360 | newexpr->funcresulttype = result_type; |
| 4361 | newexpr->funcretset = false; |
| 4362 | newexpr->funcvariadic = funcvariadic; |
| 4363 | newexpr->funcformat = COERCE_EXPLICIT_CALL; /* doesn't matter */ |
| 4364 | newexpr->funccollid = result_collid; /* doesn't matter */ |
| 4365 | newexpr->inputcollid = input_collid; |
| 4366 | newexpr->args = args; |
| 4367 | newexpr->location = -1; |
| 4368 | |
| 4369 | return evaluate_expr((Expr *) newexpr, result_type, result_typmod, |
| 4370 | result_collid); |
| 4371 | } |
| 4372 | |
| 4373 | /* |
| 4374 | * inline_function: try to expand a function call inline |
| 4375 | * |
| 4376 | * If the function is a sufficiently simple SQL-language function |
| 4377 | * (just "SELECT expression"), then we can inline it and avoid the rather |
| 4378 | * high per-call overhead of SQL functions. Furthermore, this can expose |
| 4379 | * opportunities for constant-folding within the function expression. |
| 4380 | * |
| 4381 | * We have to beware of some special cases however. A directly or |
| 4382 | * indirectly recursive function would cause us to recurse forever, |
| 4383 | * so we keep track of which functions we are already expanding and |
| 4384 | * do not re-expand them. Also, if a parameter is used more than once |
| 4385 | * in the SQL-function body, we require it not to contain any volatile |
| 4386 | * functions (volatiles might deliver inconsistent answers) nor to be |
| 4387 | * unreasonably expensive to evaluate. The expensiveness check not only |
| 4388 | * prevents us from doing multiple evaluations of an expensive parameter |
| 4389 | * at runtime, but is a safety value to limit growth of an expression due |
| 4390 | * to repeated inlining. |
| 4391 | * |
| 4392 | * We must also beware of changing the volatility or strictness status of |
| 4393 | * functions by inlining them. |
| 4394 | * |
| 4395 | * Also, at the moment we can't inline functions returning RECORD. This |
| 4396 | * doesn't work in the general case because it discards information such |
| 4397 | * as OUT-parameter declarations. |
| 4398 | * |
| 4399 | * Also, context-dependent expression nodes in the argument list are trouble. |
| 4400 | * |
| 4401 | * Returns a simplified expression if successful, or NULL if cannot |
| 4402 | * simplify the function. |
| 4403 | */ |
| 4404 | static Expr * |
| 4405 | inline_function(Oid funcid, Oid result_type, Oid result_collid, |
| 4406 | Oid input_collid, List *args, |
| 4407 | bool funcvariadic, |
| 4408 | HeapTuple func_tuple, |
| 4409 | eval_const_expressions_context *context) |
| 4410 | { |
| 4411 | Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple); |
| 4412 | char *src; |
| 4413 | Datum tmp; |
| 4414 | bool isNull; |
| 4415 | bool modifyTargetList; |
| 4416 | MemoryContext oldcxt; |
| 4417 | MemoryContext mycxt; |
| 4418 | inline_error_callback_arg callback_arg; |
| 4419 | ErrorContextCallback sqlerrcontext; |
| 4420 | FuncExpr *fexpr; |
| 4421 | SQLFunctionParseInfoPtr pinfo; |
| 4422 | ParseState *pstate; |
| 4423 | List *raw_parsetree_list; |
| 4424 | Query *querytree; |
| 4425 | Node *newexpr; |
| 4426 | int *usecounts; |
| 4427 | ListCell *arg; |
| 4428 | int i; |
| 4429 | |
| 4430 | /* |
| 4431 | * Forget it if the function is not SQL-language or has other showstopper |
| 4432 | * properties. (The prokind and nargs checks are just paranoia.) |
| 4433 | */ |
| 4434 | if (funcform->prolang != SQLlanguageId || |
| 4435 | funcform->prokind != PROKIND_FUNCTION || |
| 4436 | funcform->prosecdef || |
| 4437 | funcform->proretset || |
| 4438 | funcform->prorettype == RECORDOID || |
| 4439 | !heap_attisnull(func_tuple, Anum_pg_proc_proconfig, NULL) || |
| 4440 | funcform->pronargs != list_length(args)) |
| 4441 | return NULL; |
| 4442 | |
| 4443 | /* Check for recursive function, and give up trying to expand if so */ |
| 4444 | if (list_member_oid(context->active_fns, funcid)) |
| 4445 | return NULL; |
| 4446 | |
| 4447 | /* Check permission to call function (fail later, if not) */ |
| 4448 | if (pg_proc_aclcheck(funcid, GetUserId(), ACL_EXECUTE) != ACLCHECK_OK) |
| 4449 | return NULL; |
| 4450 | |
| 4451 | /* Check whether a plugin wants to hook function entry/exit */ |
| 4452 | if (FmgrHookIsNeeded(funcid)) |
| 4453 | return NULL; |
| 4454 | |
| 4455 | /* |
| 4456 | * Make a temporary memory context, so that we don't leak all the stuff |
| 4457 | * that parsing might create. |
| 4458 | */ |
| 4459 | mycxt = AllocSetContextCreate(CurrentMemoryContext, |
| 4460 | "inline_function" , |
| 4461 | ALLOCSET_DEFAULT_SIZES); |
| 4462 | oldcxt = MemoryContextSwitchTo(mycxt); |
| 4463 | |
| 4464 | /* Fetch the function body */ |
| 4465 | tmp = SysCacheGetAttr(PROCOID, |
| 4466 | func_tuple, |
| 4467 | Anum_pg_proc_prosrc, |
| 4468 | &isNull); |
| 4469 | if (isNull) |
| 4470 | elog(ERROR, "null prosrc for function %u" , funcid); |
| 4471 | src = TextDatumGetCString(tmp); |
| 4472 | |
| 4473 | /* |
| 4474 | * Setup error traceback support for ereport(). This is so that we can |
| 4475 | * finger the function that bad information came from. |
| 4476 | */ |
| 4477 | callback_arg.proname = NameStr(funcform->proname); |
| 4478 | callback_arg.prosrc = src; |
| 4479 | |
| 4480 | sqlerrcontext.callback = sql_inline_error_callback; |
| 4481 | sqlerrcontext.arg = (void *) &callback_arg; |
| 4482 | sqlerrcontext.previous = error_context_stack; |
| 4483 | error_context_stack = &sqlerrcontext; |
| 4484 | |
| 4485 | /* |
| 4486 | * Set up to handle parameters while parsing the function body. We need a |
| 4487 | * dummy FuncExpr node containing the already-simplified arguments to pass |
| 4488 | * to prepare_sql_fn_parse_info. (It is really only needed if there are |
| 4489 | * some polymorphic arguments, but for simplicity we always build it.) |
| 4490 | */ |
| 4491 | fexpr = makeNode(FuncExpr); |
| 4492 | fexpr->funcid = funcid; |
| 4493 | fexpr->funcresulttype = result_type; |
| 4494 | fexpr->funcretset = false; |
| 4495 | fexpr->funcvariadic = funcvariadic; |
| 4496 | fexpr->funcformat = COERCE_EXPLICIT_CALL; /* doesn't matter */ |
| 4497 | fexpr->funccollid = result_collid; /* doesn't matter */ |
| 4498 | fexpr->inputcollid = input_collid; |
| 4499 | fexpr->args = args; |
| 4500 | fexpr->location = -1; |
| 4501 | |
| 4502 | pinfo = prepare_sql_fn_parse_info(func_tuple, |
| 4503 | (Node *) fexpr, |
| 4504 | input_collid); |
| 4505 | |
| 4506 | /* |
| 4507 | * We just do parsing and parse analysis, not rewriting, because rewriting |
| 4508 | * will not affect table-free-SELECT-only queries, which is all that we |
| 4509 | * care about. Also, we can punt as soon as we detect more than one |
| 4510 | * command in the function body. |
| 4511 | */ |
| 4512 | raw_parsetree_list = pg_parse_query(src); |
| 4513 | if (list_length(raw_parsetree_list) != 1) |
| 4514 | goto fail; |
| 4515 | |
| 4516 | pstate = make_parsestate(NULL); |
| 4517 | pstate->p_sourcetext = src; |
| 4518 | sql_fn_parser_setup(pstate, pinfo); |
| 4519 | |
| 4520 | querytree = transformTopLevelStmt(pstate, linitial(raw_parsetree_list)); |
| 4521 | |
| 4522 | free_parsestate(pstate); |
| 4523 | |
| 4524 | /* |
| 4525 | * The single command must be a simple "SELECT expression". |
| 4526 | * |
| 4527 | * Note: if you change the tests involved in this, see also plpgsql's |
| 4528 | * exec_simple_check_plan(). That generally needs to have the same idea |
| 4529 | * of what's a "simple expression", so that inlining a function that |
| 4530 | * previously wasn't inlined won't change plpgsql's conclusion. |
| 4531 | */ |
| 4532 | if (!IsA(querytree, Query) || |
| 4533 | querytree->commandType != CMD_SELECT || |
| 4534 | querytree->hasAggs || |
| 4535 | querytree->hasWindowFuncs || |
| 4536 | querytree->hasTargetSRFs || |
| 4537 | querytree->hasSubLinks || |
| 4538 | querytree->cteList || |
| 4539 | querytree->rtable || |
| 4540 | querytree->jointree->fromlist || |
| 4541 | querytree->jointree->quals || |
| 4542 | querytree->groupClause || |
| 4543 | querytree->groupingSets || |
| 4544 | querytree->havingQual || |
| 4545 | querytree->windowClause || |
| 4546 | querytree->distinctClause || |
| 4547 | querytree->sortClause || |
| 4548 | querytree->limitOffset || |
| 4549 | querytree->limitCount || |
| 4550 | querytree->setOperations || |
| 4551 | list_length(querytree->targetList) != 1) |
| 4552 | goto fail; |
| 4553 | |
| 4554 | /* |
| 4555 | * Make sure the function (still) returns what it's declared to. This |
| 4556 | * will raise an error if wrong, but that's okay since the function would |
| 4557 | * fail at runtime anyway. Note that check_sql_fn_retval will also insert |
| 4558 | * a RelabelType if needed to make the tlist expression match the declared |
| 4559 | * type of the function. |
| 4560 | * |
| 4561 | * Note: we do not try this until we have verified that no rewriting was |
| 4562 | * needed; that's probably not important, but let's be careful. |
| 4563 | */ |
| 4564 | if (check_sql_fn_retval(funcid, result_type, list_make1(querytree), |
| 4565 | &modifyTargetList, NULL)) |
| 4566 | goto fail; /* reject whole-tuple-result cases */ |
| 4567 | |
| 4568 | /* Now we can grab the tlist expression */ |
| 4569 | newexpr = (Node *) ((TargetEntry *) linitial(querytree->targetList))->expr; |
| 4570 | |
| 4571 | /* |
| 4572 | * If the SQL function returns VOID, we can only inline it if it is a |
| 4573 | * SELECT of an expression returning VOID (ie, it's just a redirection to |
| 4574 | * another VOID-returning function). In all non-VOID-returning cases, |
| 4575 | * check_sql_fn_retval should ensure that newexpr returns the function's |
| 4576 | * declared result type, so this test shouldn't fail otherwise; but we may |
| 4577 | * as well cope gracefully if it does. |
| 4578 | */ |
| 4579 | if (exprType(newexpr) != result_type) |
| 4580 | goto fail; |
| 4581 | |
| 4582 | /* check_sql_fn_retval couldn't have made any dangerous tlist changes */ |
| 4583 | Assert(!modifyTargetList); |
| 4584 | |
| 4585 | /* |
| 4586 | * Additional validity checks on the expression. It mustn't be more |
| 4587 | * volatile than the surrounding function (this is to avoid breaking hacks |
| 4588 | * that involve pretending a function is immutable when it really ain't). |
| 4589 | * If the surrounding function is declared strict, then the expression |
| 4590 | * must contain only strict constructs and must use all of the function |
| 4591 | * parameters (this is overkill, but an exact analysis is hard). |
| 4592 | */ |
| 4593 | if (funcform->provolatile == PROVOLATILE_IMMUTABLE && |
| 4594 | contain_mutable_functions(newexpr)) |
| 4595 | goto fail; |
| 4596 | else if (funcform->provolatile == PROVOLATILE_STABLE && |
| 4597 | contain_volatile_functions(newexpr)) |
| 4598 | goto fail; |
| 4599 | |
| 4600 | if (funcform->proisstrict && |
| 4601 | contain_nonstrict_functions(newexpr)) |
| 4602 | goto fail; |
| 4603 | |
| 4604 | /* |
| 4605 | * If any parameter expression contains a context-dependent node, we can't |
| 4606 | * inline, for fear of putting such a node into the wrong context. |
| 4607 | */ |
| 4608 | if (contain_context_dependent_node((Node *) args)) |
| 4609 | goto fail; |
| 4610 | |
| 4611 | /* |
| 4612 | * We may be able to do it; there are still checks on parameter usage to |
| 4613 | * make, but those are most easily done in combination with the actual |
| 4614 | * substitution of the inputs. So start building expression with inputs |
| 4615 | * substituted. |
| 4616 | */ |
| 4617 | usecounts = (int *) palloc0(funcform->pronargs * sizeof(int)); |
| 4618 | newexpr = substitute_actual_parameters(newexpr, funcform->pronargs, |
| 4619 | args, usecounts); |
| 4620 | |
| 4621 | /* Now check for parameter usage */ |
| 4622 | i = 0; |
| 4623 | foreach(arg, args) |
| 4624 | { |
| 4625 | Node *param = lfirst(arg); |
| 4626 | |
| 4627 | if (usecounts[i] == 0) |
| 4628 | { |
| 4629 | /* Param not used at all: uncool if func is strict */ |
| 4630 | if (funcform->proisstrict) |
| 4631 | goto fail; |
| 4632 | } |
| 4633 | else if (usecounts[i] != 1) |
| 4634 | { |
| 4635 | /* Param used multiple times: uncool if expensive or volatile */ |
| 4636 | QualCost eval_cost; |
| 4637 | |
| 4638 | /* |
| 4639 | * We define "expensive" as "contains any subplan or more than 10 |
| 4640 | * operators". Note that the subplan search has to be done |
| 4641 | * explicitly, since cost_qual_eval() will barf on unplanned |
| 4642 | * subselects. |
| 4643 | */ |
| 4644 | if (contain_subplans(param)) |
| 4645 | goto fail; |
| 4646 | cost_qual_eval(&eval_cost, list_make1(param), NULL); |
| 4647 | if (eval_cost.startup + eval_cost.per_tuple > |
| 4648 | 10 * cpu_operator_cost) |
| 4649 | goto fail; |
| 4650 | |
| 4651 | /* |
| 4652 | * Check volatility last since this is more expensive than the |
| 4653 | * above tests |
| 4654 | */ |
| 4655 | if (contain_volatile_functions(param)) |
| 4656 | goto fail; |
| 4657 | } |
| 4658 | i++; |
| 4659 | } |
| 4660 | |
| 4661 | /* |
| 4662 | * Whew --- we can make the substitution. Copy the modified expression |
| 4663 | * out of the temporary memory context, and clean up. |
| 4664 | */ |
| 4665 | MemoryContextSwitchTo(oldcxt); |
| 4666 | |
| 4667 | newexpr = copyObject(newexpr); |
| 4668 | |
| 4669 | MemoryContextDelete(mycxt); |
| 4670 | |
| 4671 | /* |
| 4672 | * If the result is of a collatable type, force the result to expose the |
| 4673 | * correct collation. In most cases this does not matter, but it's |
| 4674 | * possible that the function result is used directly as a sort key or in |
| 4675 | * other places where we expect exprCollation() to tell the truth. |
| 4676 | */ |
| 4677 | if (OidIsValid(result_collid)) |
| 4678 | { |
| 4679 | Oid exprcoll = exprCollation(newexpr); |
| 4680 | |
| 4681 | if (OidIsValid(exprcoll) && exprcoll != result_collid) |
| 4682 | { |
| 4683 | CollateExpr *newnode = makeNode(CollateExpr); |
| 4684 | |
| 4685 | newnode->arg = (Expr *) newexpr; |
| 4686 | newnode->collOid = result_collid; |
| 4687 | newnode->location = -1; |
| 4688 | |
| 4689 | newexpr = (Node *) newnode; |
| 4690 | } |
| 4691 | } |
| 4692 | |
| 4693 | /* |
| 4694 | * Since there is now no trace of the function in the plan tree, we must |
| 4695 | * explicitly record the plan's dependency on the function. |
| 4696 | */ |
| 4697 | if (context->root) |
| 4698 | record_plan_function_dependency(context->root, funcid); |
| 4699 | |
| 4700 | /* |
| 4701 | * Recursively try to simplify the modified expression. Here we must add |
| 4702 | * the current function to the context list of active functions. |
| 4703 | */ |
| 4704 | context->active_fns = lcons_oid(funcid, context->active_fns); |
| 4705 | newexpr = eval_const_expressions_mutator(newexpr, context); |
| 4706 | context->active_fns = list_delete_first(context->active_fns); |
| 4707 | |
| 4708 | error_context_stack = sqlerrcontext.previous; |
| 4709 | |
| 4710 | return (Expr *) newexpr; |
| 4711 | |
| 4712 | /* Here if func is not inlinable: release temp memory and return NULL */ |
| 4713 | fail: |
| 4714 | MemoryContextSwitchTo(oldcxt); |
| 4715 | MemoryContextDelete(mycxt); |
| 4716 | error_context_stack = sqlerrcontext.previous; |
| 4717 | |
| 4718 | return NULL; |
| 4719 | } |
| 4720 | |
| 4721 | /* |
| 4722 | * Replace Param nodes by appropriate actual parameters |
| 4723 | */ |
| 4724 | static Node * |
| 4725 | substitute_actual_parameters(Node *expr, int nargs, List *args, |
| 4726 | int *usecounts) |
| 4727 | { |
| 4728 | substitute_actual_parameters_context context; |
| 4729 | |
| 4730 | context.nargs = nargs; |
| 4731 | context.args = args; |
| 4732 | context.usecounts = usecounts; |
| 4733 | |
| 4734 | return substitute_actual_parameters_mutator(expr, &context); |
| 4735 | } |
| 4736 | |
| 4737 | static Node * |
| 4738 | substitute_actual_parameters_mutator(Node *node, |
| 4739 | substitute_actual_parameters_context *context) |
| 4740 | { |
| 4741 | if (node == NULL) |
| 4742 | return NULL; |
| 4743 | if (IsA(node, Param)) |
| 4744 | { |
| 4745 | Param *param = (Param *) node; |
| 4746 | |
| 4747 | if (param->paramkind != PARAM_EXTERN) |
| 4748 | elog(ERROR, "unexpected paramkind: %d" , (int) param->paramkind); |
| 4749 | if (param->paramid <= 0 || param->paramid > context->nargs) |
| 4750 | elog(ERROR, "invalid paramid: %d" , param->paramid); |
| 4751 | |
| 4752 | /* Count usage of parameter */ |
| 4753 | context->usecounts[param->paramid - 1]++; |
| 4754 | |
| 4755 | /* Select the appropriate actual arg and replace the Param with it */ |
| 4756 | /* We don't need to copy at this time (it'll get done later) */ |
| 4757 | return list_nth(context->args, param->paramid - 1); |
| 4758 | } |
| 4759 | return expression_tree_mutator(node, substitute_actual_parameters_mutator, |
| 4760 | (void *) context); |
| 4761 | } |
| 4762 | |
| 4763 | /* |
| 4764 | * error context callback to let us supply a call-stack traceback |
| 4765 | */ |
| 4766 | static void |
| 4767 | sql_inline_error_callback(void *arg) |
| 4768 | { |
| 4769 | inline_error_callback_arg *callback_arg = (inline_error_callback_arg *) arg; |
| 4770 | int syntaxerrposition; |
| 4771 | |
| 4772 | /* If it's a syntax error, convert to internal syntax error report */ |
| 4773 | syntaxerrposition = geterrposition(); |
| 4774 | if (syntaxerrposition > 0) |
| 4775 | { |
| 4776 | errposition(0); |
| 4777 | internalerrposition(syntaxerrposition); |
| 4778 | internalerrquery(callback_arg->prosrc); |
| 4779 | } |
| 4780 | |
| 4781 | errcontext("SQL function \"%s\" during inlining" , callback_arg->proname); |
| 4782 | } |
| 4783 | |
| 4784 | /* |
| 4785 | * evaluate_expr: pre-evaluate a constant expression |
| 4786 | * |
| 4787 | * We use the executor's routine ExecEvalExpr() to avoid duplication of |
| 4788 | * code and ensure we get the same result as the executor would get. |
| 4789 | */ |
| 4790 | Expr * |
| 4791 | evaluate_expr(Expr *expr, Oid result_type, int32 result_typmod, |
| 4792 | Oid result_collation) |
| 4793 | { |
| 4794 | EState *estate; |
| 4795 | ExprState *exprstate; |
| 4796 | MemoryContext oldcontext; |
| 4797 | Datum const_val; |
| 4798 | bool const_is_null; |
| 4799 | int16 resultTypLen; |
| 4800 | bool resultTypByVal; |
| 4801 | |
| 4802 | /* |
| 4803 | * To use the executor, we need an EState. |
| 4804 | */ |
| 4805 | estate = CreateExecutorState(); |
| 4806 | |
| 4807 | /* We can use the estate's working context to avoid memory leaks. */ |
| 4808 | oldcontext = MemoryContextSwitchTo(estate->es_query_cxt); |
| 4809 | |
| 4810 | /* Make sure any opfuncids are filled in. */ |
| 4811 | fix_opfuncids((Node *) expr); |
| 4812 | |
| 4813 | /* |
| 4814 | * Prepare expr for execution. (Note: we can't use ExecPrepareExpr |
| 4815 | * because it'd result in recursively invoking eval_const_expressions.) |
| 4816 | */ |
| 4817 | exprstate = ExecInitExpr(expr, NULL); |
| 4818 | |
| 4819 | /* |
| 4820 | * And evaluate it. |
| 4821 | * |
| 4822 | * It is OK to use a default econtext because none of the ExecEvalExpr() |
| 4823 | * code used in this situation will use econtext. That might seem |
| 4824 | * fortuitous, but it's not so unreasonable --- a constant expression does |
| 4825 | * not depend on context, by definition, n'est ce pas? |
| 4826 | */ |
| 4827 | const_val = ExecEvalExprSwitchContext(exprstate, |
| 4828 | GetPerTupleExprContext(estate), |
| 4829 | &const_is_null); |
| 4830 | |
| 4831 | /* Get info needed about result datatype */ |
| 4832 | get_typlenbyval(result_type, &resultTypLen, &resultTypByVal); |
| 4833 | |
| 4834 | /* Get back to outer memory context */ |
| 4835 | MemoryContextSwitchTo(oldcontext); |
| 4836 | |
| 4837 | /* |
| 4838 | * Must copy result out of sub-context used by expression eval. |
| 4839 | * |
| 4840 | * Also, if it's varlena, forcibly detoast it. This protects us against |
| 4841 | * storing TOAST pointers into plans that might outlive the referenced |
| 4842 | * data. (makeConst would handle detoasting anyway, but it's worth a few |
| 4843 | * extra lines here so that we can do the copy and detoast in one step.) |
| 4844 | */ |
| 4845 | if (!const_is_null) |
| 4846 | { |
| 4847 | if (resultTypLen == -1) |
| 4848 | const_val = PointerGetDatum(PG_DETOAST_DATUM_COPY(const_val)); |
| 4849 | else |
| 4850 | const_val = datumCopy(const_val, resultTypByVal, resultTypLen); |
| 4851 | } |
| 4852 | |
| 4853 | /* Release all the junk we just created */ |
| 4854 | FreeExecutorState(estate); |
| 4855 | |
| 4856 | /* |
| 4857 | * Make the constant result node. |
| 4858 | */ |
| 4859 | return (Expr *) makeConst(result_type, result_typmod, result_collation, |
| 4860 | resultTypLen, |
| 4861 | const_val, const_is_null, |
| 4862 | resultTypByVal); |
| 4863 | } |
| 4864 | |
| 4865 | |
| 4866 | /* |
| 4867 | * inline_set_returning_function |
| 4868 | * Attempt to "inline" a set-returning function in the FROM clause. |
| 4869 | * |
| 4870 | * "rte" is an RTE_FUNCTION rangetable entry. If it represents a call of a |
| 4871 | * set-returning SQL function that can safely be inlined, expand the function |
| 4872 | * and return the substitute Query structure. Otherwise, return NULL. |
| 4873 | * |
| 4874 | * This has a good deal of similarity to inline_function(), but that's |
| 4875 | * for the non-set-returning case, and there are enough differences to |
| 4876 | * justify separate functions. |
| 4877 | */ |
| 4878 | Query * |
| 4879 | inline_set_returning_function(PlannerInfo *root, RangeTblEntry *rte) |
| 4880 | { |
| 4881 | RangeTblFunction *rtfunc; |
| 4882 | FuncExpr *fexpr; |
| 4883 | Oid func_oid; |
| 4884 | HeapTuple func_tuple; |
| 4885 | Form_pg_proc funcform; |
| 4886 | char *src; |
| 4887 | Datum tmp; |
| 4888 | bool isNull; |
| 4889 | bool modifyTargetList; |
| 4890 | MemoryContext oldcxt; |
| 4891 | MemoryContext mycxt; |
| 4892 | List *saveInvalItems; |
| 4893 | inline_error_callback_arg callback_arg; |
| 4894 | ErrorContextCallback sqlerrcontext; |
| 4895 | SQLFunctionParseInfoPtr pinfo; |
| 4896 | List *raw_parsetree_list; |
| 4897 | List *querytree_list; |
| 4898 | Query *querytree; |
| 4899 | |
| 4900 | Assert(rte->rtekind == RTE_FUNCTION); |
| 4901 | |
| 4902 | /* |
| 4903 | * It doesn't make a lot of sense for a SQL SRF to refer to itself in its |
| 4904 | * own FROM clause, since that must cause infinite recursion at runtime. |
| 4905 | * It will cause this code to recurse too, so check for stack overflow. |
| 4906 | * (There's no need to do more.) |
| 4907 | */ |
| 4908 | check_stack_depth(); |
| 4909 | |
| 4910 | /* Fail if the RTE has ORDINALITY - we don't implement that here. */ |
| 4911 | if (rte->funcordinality) |
| 4912 | return NULL; |
| 4913 | |
| 4914 | /* Fail if RTE isn't a single, simple FuncExpr */ |
| 4915 | if (list_length(rte->functions) != 1) |
| 4916 | return NULL; |
| 4917 | rtfunc = (RangeTblFunction *) linitial(rte->functions); |
| 4918 | |
| 4919 | if (!IsA(rtfunc->funcexpr, FuncExpr)) |
| 4920 | return NULL; |
| 4921 | fexpr = (FuncExpr *) rtfunc->funcexpr; |
| 4922 | |
| 4923 | func_oid = fexpr->funcid; |
| 4924 | |
| 4925 | /* |
| 4926 | * The function must be declared to return a set, else inlining would |
| 4927 | * change the results if the contained SELECT didn't return exactly one |
| 4928 | * row. |
| 4929 | */ |
| 4930 | if (!fexpr->funcretset) |
| 4931 | return NULL; |
| 4932 | |
| 4933 | /* |
| 4934 | * Refuse to inline if the arguments contain any volatile functions or |
| 4935 | * sub-selects. Volatile functions are rejected because inlining may |
| 4936 | * result in the arguments being evaluated multiple times, risking a |
| 4937 | * change in behavior. Sub-selects are rejected partly for implementation |
| 4938 | * reasons (pushing them down another level might change their behavior) |
| 4939 | * and partly because they're likely to be expensive and so multiple |
| 4940 | * evaluation would be bad. |
| 4941 | */ |
| 4942 | if (contain_volatile_functions((Node *) fexpr->args) || |
| 4943 | contain_subplans((Node *) fexpr->args)) |
| 4944 | return NULL; |
| 4945 | |
| 4946 | /* Check permission to call function (fail later, if not) */ |
| 4947 | if (pg_proc_aclcheck(func_oid, GetUserId(), ACL_EXECUTE) != ACLCHECK_OK) |
| 4948 | return NULL; |
| 4949 | |
| 4950 | /* Check whether a plugin wants to hook function entry/exit */ |
| 4951 | if (FmgrHookIsNeeded(func_oid)) |
| 4952 | return NULL; |
| 4953 | |
| 4954 | /* |
| 4955 | * OK, let's take a look at the function's pg_proc entry. |
| 4956 | */ |
| 4957 | func_tuple = SearchSysCache1(PROCOID, ObjectIdGetDatum(func_oid)); |
| 4958 | if (!HeapTupleIsValid(func_tuple)) |
| 4959 | elog(ERROR, "cache lookup failed for function %u" , func_oid); |
| 4960 | funcform = (Form_pg_proc) GETSTRUCT(func_tuple); |
| 4961 | |
| 4962 | /* |
| 4963 | * Forget it if the function is not SQL-language or has other showstopper |
| 4964 | * properties. In particular it mustn't be declared STRICT, since we |
| 4965 | * couldn't enforce that. It also mustn't be VOLATILE, because that is |
| 4966 | * supposed to cause it to be executed with its own snapshot, rather than |
| 4967 | * sharing the snapshot of the calling query. We also disallow returning |
| 4968 | * SETOF VOID, because inlining would result in exposing the actual result |
| 4969 | * of the function's last SELECT, which should not happen in that case. |
| 4970 | * (Rechecking prokind and proretset is just paranoia.) |
| 4971 | */ |
| 4972 | if (funcform->prolang != SQLlanguageId || |
| 4973 | funcform->prokind != PROKIND_FUNCTION || |
| 4974 | funcform->proisstrict || |
| 4975 | funcform->provolatile == PROVOLATILE_VOLATILE || |
| 4976 | funcform->prorettype == VOIDOID || |
| 4977 | funcform->prosecdef || |
| 4978 | !funcform->proretset || |
| 4979 | !heap_attisnull(func_tuple, Anum_pg_proc_proconfig, NULL)) |
| 4980 | { |
| 4981 | ReleaseSysCache(func_tuple); |
| 4982 | return NULL; |
| 4983 | } |
| 4984 | |
| 4985 | /* |
| 4986 | * Make a temporary memory context, so that we don't leak all the stuff |
| 4987 | * that parsing might create. |
| 4988 | */ |
| 4989 | mycxt = AllocSetContextCreate(CurrentMemoryContext, |
| 4990 | "inline_set_returning_function" , |
| 4991 | ALLOCSET_DEFAULT_SIZES); |
| 4992 | oldcxt = MemoryContextSwitchTo(mycxt); |
| 4993 | |
| 4994 | /* |
| 4995 | * When we call eval_const_expressions below, it might try to add items to |
| 4996 | * root->glob->invalItems. Since it is running in the temp context, those |
| 4997 | * items will be in that context, and will need to be copied out if we're |
| 4998 | * successful. Temporarily reset the list so that we can keep those items |
| 4999 | * separate from the pre-existing list contents. |
| 5000 | */ |
| 5001 | saveInvalItems = root->glob->invalItems; |
| 5002 | root->glob->invalItems = NIL; |
| 5003 | |
| 5004 | /* Fetch the function body */ |
| 5005 | tmp = SysCacheGetAttr(PROCOID, |
| 5006 | func_tuple, |
| 5007 | Anum_pg_proc_prosrc, |
| 5008 | &isNull); |
| 5009 | if (isNull) |
| 5010 | elog(ERROR, "null prosrc for function %u" , func_oid); |
| 5011 | src = TextDatumGetCString(tmp); |
| 5012 | |
| 5013 | /* |
| 5014 | * Setup error traceback support for ereport(). This is so that we can |
| 5015 | * finger the function that bad information came from. |
| 5016 | */ |
| 5017 | callback_arg.proname = NameStr(funcform->proname); |
| 5018 | callback_arg.prosrc = src; |
| 5019 | |
| 5020 | sqlerrcontext.callback = sql_inline_error_callback; |
| 5021 | sqlerrcontext.arg = (void *) &callback_arg; |
| 5022 | sqlerrcontext.previous = error_context_stack; |
| 5023 | error_context_stack = &sqlerrcontext; |
| 5024 | |
| 5025 | /* |
| 5026 | * Run eval_const_expressions on the function call. This is necessary to |
| 5027 | * ensure that named-argument notation is converted to positional notation |
| 5028 | * and any default arguments are inserted. It's a bit of overkill for the |
| 5029 | * arguments, since they'll get processed again later, but no harm will be |
| 5030 | * done. |
| 5031 | */ |
| 5032 | fexpr = (FuncExpr *) eval_const_expressions(root, (Node *) fexpr); |
| 5033 | |
| 5034 | /* It should still be a call of the same function, but let's check */ |
| 5035 | if (!IsA(fexpr, FuncExpr) || |
| 5036 | fexpr->funcid != func_oid) |
| 5037 | goto fail; |
| 5038 | |
| 5039 | /* Arg list length should now match the function */ |
| 5040 | if (list_length(fexpr->args) != funcform->pronargs) |
| 5041 | goto fail; |
| 5042 | |
| 5043 | /* |
| 5044 | * Set up to handle parameters while parsing the function body. We can |
| 5045 | * use the FuncExpr just created as the input for |
| 5046 | * prepare_sql_fn_parse_info. |
| 5047 | */ |
| 5048 | pinfo = prepare_sql_fn_parse_info(func_tuple, |
| 5049 | (Node *) fexpr, |
| 5050 | fexpr->inputcollid); |
| 5051 | |
| 5052 | /* |
| 5053 | * Parse, analyze, and rewrite (unlike inline_function(), we can't skip |
| 5054 | * rewriting here). We can fail as soon as we find more than one query, |
| 5055 | * though. |
| 5056 | */ |
| 5057 | raw_parsetree_list = pg_parse_query(src); |
| 5058 | if (list_length(raw_parsetree_list) != 1) |
| 5059 | goto fail; |
| 5060 | |
| 5061 | querytree_list = pg_analyze_and_rewrite_params(linitial(raw_parsetree_list), |
| 5062 | src, |
| 5063 | (ParserSetupHook) sql_fn_parser_setup, |
| 5064 | pinfo, NULL); |
| 5065 | if (list_length(querytree_list) != 1) |
| 5066 | goto fail; |
| 5067 | querytree = linitial(querytree_list); |
| 5068 | |
| 5069 | /* |
| 5070 | * The single command must be a plain SELECT. |
| 5071 | */ |
| 5072 | if (!IsA(querytree, Query) || |
| 5073 | querytree->commandType != CMD_SELECT) |
| 5074 | goto fail; |
| 5075 | |
| 5076 | /* |
| 5077 | * Make sure the function (still) returns what it's declared to. This |
| 5078 | * will raise an error if wrong, but that's okay since the function would |
| 5079 | * fail at runtime anyway. Note that check_sql_fn_retval will also insert |
| 5080 | * RelabelType(s) and/or NULL columns if needed to make the tlist |
| 5081 | * expression(s) match the declared type of the function. |
| 5082 | * |
| 5083 | * If the function returns a composite type, don't inline unless the check |
| 5084 | * shows it's returning a whole tuple result; otherwise what it's |
| 5085 | * returning is a single composite column which is not what we need. (Like |
| 5086 | * check_sql_fn_retval, we deliberately exclude domains over composite |
| 5087 | * here.) |
| 5088 | */ |
| 5089 | if (!check_sql_fn_retval(func_oid, fexpr->funcresulttype, |
| 5090 | querytree_list, |
| 5091 | &modifyTargetList, NULL) && |
| 5092 | (get_typtype(fexpr->funcresulttype) == TYPTYPE_COMPOSITE || |
| 5093 | fexpr->funcresulttype == RECORDOID)) |
| 5094 | goto fail; /* reject not-whole-tuple-result cases */ |
| 5095 | |
| 5096 | /* |
| 5097 | * If we had to modify the tlist to make it match, and the statement is |
| 5098 | * one in which changing the tlist contents could change semantics, we |
| 5099 | * have to punt and not inline. |
| 5100 | */ |
| 5101 | if (modifyTargetList) |
| 5102 | goto fail; |
| 5103 | |
| 5104 | /* |
| 5105 | * If it returns RECORD, we have to check against the column type list |
| 5106 | * provided in the RTE; check_sql_fn_retval can't do that. (If no match, |
| 5107 | * we just fail to inline, rather than complaining; see notes for |
| 5108 | * tlist_matches_coltypelist.) We don't have to do this for functions |
| 5109 | * with declared OUT parameters, even though their funcresulttype is |
| 5110 | * RECORDOID, so check get_func_result_type too. |
| 5111 | */ |
| 5112 | if (fexpr->funcresulttype == RECORDOID && |
| 5113 | get_func_result_type(func_oid, NULL, NULL) == TYPEFUNC_RECORD && |
| 5114 | !tlist_matches_coltypelist(querytree->targetList, |
| 5115 | rtfunc->funccoltypes)) |
| 5116 | goto fail; |
| 5117 | |
| 5118 | /* |
| 5119 | * Looks good --- substitute parameters into the query. |
| 5120 | */ |
| 5121 | querytree = substitute_actual_srf_parameters(querytree, |
| 5122 | funcform->pronargs, |
| 5123 | fexpr->args); |
| 5124 | |
| 5125 | /* |
| 5126 | * Copy the modified query out of the temporary memory context, and clean |
| 5127 | * up. |
| 5128 | */ |
| 5129 | MemoryContextSwitchTo(oldcxt); |
| 5130 | |
| 5131 | querytree = copyObject(querytree); |
| 5132 | |
| 5133 | /* copy up any new invalItems, too */ |
| 5134 | root->glob->invalItems = list_concat(saveInvalItems, |
| 5135 | copyObject(root->glob->invalItems)); |
| 5136 | |
| 5137 | MemoryContextDelete(mycxt); |
| 5138 | error_context_stack = sqlerrcontext.previous; |
| 5139 | ReleaseSysCache(func_tuple); |
| 5140 | |
| 5141 | /* |
| 5142 | * We don't have to fix collations here because the upper query is already |
| 5143 | * parsed, ie, the collations in the RTE are what count. |
| 5144 | */ |
| 5145 | |
| 5146 | /* |
| 5147 | * Since there is now no trace of the function in the plan tree, we must |
| 5148 | * explicitly record the plan's dependency on the function. |
| 5149 | */ |
| 5150 | record_plan_function_dependency(root, func_oid); |
| 5151 | |
| 5152 | return querytree; |
| 5153 | |
| 5154 | /* Here if func is not inlinable: release temp memory and return NULL */ |
| 5155 | fail: |
| 5156 | MemoryContextSwitchTo(oldcxt); |
| 5157 | root->glob->invalItems = saveInvalItems; |
| 5158 | MemoryContextDelete(mycxt); |
| 5159 | error_context_stack = sqlerrcontext.previous; |
| 5160 | ReleaseSysCache(func_tuple); |
| 5161 | |
| 5162 | return NULL; |
| 5163 | } |
| 5164 | |
| 5165 | /* |
| 5166 | * Replace Param nodes by appropriate actual parameters |
| 5167 | * |
| 5168 | * This is just enough different from substitute_actual_parameters() |
| 5169 | * that it needs its own code. |
| 5170 | */ |
| 5171 | static Query * |
| 5172 | substitute_actual_srf_parameters(Query *expr, int nargs, List *args) |
| 5173 | { |
| 5174 | substitute_actual_srf_parameters_context context; |
| 5175 | |
| 5176 | context.nargs = nargs; |
| 5177 | context.args = args; |
| 5178 | context.sublevels_up = 1; |
| 5179 | |
| 5180 | return query_tree_mutator(expr, |
| 5181 | substitute_actual_srf_parameters_mutator, |
| 5182 | &context, |
| 5183 | 0); |
| 5184 | } |
| 5185 | |
| 5186 | static Node * |
| 5187 | substitute_actual_srf_parameters_mutator(Node *node, |
| 5188 | substitute_actual_srf_parameters_context *context) |
| 5189 | { |
| 5190 | Node *result; |
| 5191 | |
| 5192 | if (node == NULL) |
| 5193 | return NULL; |
| 5194 | if (IsA(node, Query)) |
| 5195 | { |
| 5196 | context->sublevels_up++; |
| 5197 | result = (Node *) query_tree_mutator((Query *) node, |
| 5198 | substitute_actual_srf_parameters_mutator, |
| 5199 | (void *) context, |
| 5200 | 0); |
| 5201 | context->sublevels_up--; |
| 5202 | return result; |
| 5203 | } |
| 5204 | if (IsA(node, Param)) |
| 5205 | { |
| 5206 | Param *param = (Param *) node; |
| 5207 | |
| 5208 | if (param->paramkind == PARAM_EXTERN) |
| 5209 | { |
| 5210 | if (param->paramid <= 0 || param->paramid > context->nargs) |
| 5211 | elog(ERROR, "invalid paramid: %d" , param->paramid); |
| 5212 | |
| 5213 | /* |
| 5214 | * Since the parameter is being inserted into a subquery, we must |
| 5215 | * adjust levels. |
| 5216 | */ |
| 5217 | result = copyObject(list_nth(context->args, param->paramid - 1)); |
| 5218 | IncrementVarSublevelsUp(result, context->sublevels_up, 0); |
| 5219 | return result; |
| 5220 | } |
| 5221 | } |
| 5222 | return expression_tree_mutator(node, |
| 5223 | substitute_actual_srf_parameters_mutator, |
| 5224 | (void *) context); |
| 5225 | } |
| 5226 | |
| 5227 | /* |
| 5228 | * Check whether a SELECT targetlist emits the specified column types, |
| 5229 | * to see if it's safe to inline a function returning record. |
| 5230 | * |
| 5231 | * We insist on exact match here. The executor allows binary-coercible |
| 5232 | * cases too, but we don't have a way to preserve the correct column types |
| 5233 | * in the correct places if we inline the function in such a case. |
| 5234 | * |
| 5235 | * Note that we only check type OIDs not typmods; this agrees with what the |
| 5236 | * executor would do at runtime, and attributing a specific typmod to a |
| 5237 | * function result is largely wishful thinking anyway. |
| 5238 | */ |
| 5239 | static bool |
| 5240 | tlist_matches_coltypelist(List *tlist, List *coltypelist) |
| 5241 | { |
| 5242 | ListCell *tlistitem; |
| 5243 | ListCell *clistitem; |
| 5244 | |
| 5245 | clistitem = list_head(coltypelist); |
| 5246 | foreach(tlistitem, tlist) |
| 5247 | { |
| 5248 | TargetEntry *tle = (TargetEntry *) lfirst(tlistitem); |
| 5249 | Oid coltype; |
| 5250 | |
| 5251 | if (tle->resjunk) |
| 5252 | continue; /* ignore junk columns */ |
| 5253 | |
| 5254 | if (clistitem == NULL) |
| 5255 | return false; /* too many tlist items */ |
| 5256 | |
| 5257 | coltype = lfirst_oid(clistitem); |
| 5258 | clistitem = lnext(clistitem); |
| 5259 | |
| 5260 | if (exprType((Node *) tle->expr) != coltype) |
| 5261 | return false; /* column type mismatch */ |
| 5262 | } |
| 5263 | |
| 5264 | if (clistitem != NULL) |
| 5265 | return false; /* too few tlist items */ |
| 5266 | |
| 5267 | return true; |
| 5268 | } |
| 5269 | |