| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* clauses.c |
| 4 |
* routines to manipulate qualification clauses |
| 5 |
* |
| 6 |
* Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
| 7 |
* Portions Copyright (c) 1994, Regents of the University of California |
| 8 |
* |
| 9 |
* |
| 10 |
* IDENTIFICATION |
| 11 |
* src/backend/optimizer/util/clauses.c |
| 12 |
* |
| 13 |
* HISTORY |
| 14 |
* AUTHOR DATE MAJOR EVENT |
| 15 |
* Andrew Yu Nov 3, 1994 clause.c and clauses.c combined |
| 16 |
* |
| 17 |
*------------------------------------------------------------------------- |
| 18 |
*/ |
| 19 |
|
| 20 |
#include "postgres.h" |
| 21 |
|
| 22 |
#include "access/htup_details.h" |
| 23 |
#include "catalog/pg_language.h" |
| 24 |
#include "catalog/pg_operator.h" |
| 25 |
#include "catalog/pg_proc.h" |
| 26 |
#include "catalog/pg_type.h" |
| 27 |
#include "executor/executor.h" |
| 28 |
#include "executor/functions.h" |
| 29 |
#include "funcapi.h" |
| 30 |
#include "miscadmin.h" |
| 31 |
#include "nodes/makefuncs.h" |
| 32 |
#include "nodes/multibitmapset.h" |
| 33 |
#include "nodes/nodeFuncs.h" |
| 34 |
#include "nodes/subscripting.h" |
| 35 |
#include "nodes/supportnodes.h" |
| 36 |
#include "optimizer/clauses.h" |
| 37 |
#include "optimizer/cost.h" |
| 38 |
#include "optimizer/optimizer.h" |
| 39 |
#include "optimizer/plancat.h" |
| 40 |
#include "optimizer/planmain.h" |
| 41 |
#include "parser/analyze.h" |
| 42 |
#include "parser/parse_coerce.h" |
| 43 |
#include "parser/parse_collate.h" |
| 44 |
#include "parser/parse_func.h" |
| 45 |
#include "parser/parse_oper.h" |
| 46 |
#include "rewrite/rewriteHandler.h" |
| 47 |
#include "rewrite/rewriteManip.h" |
| 48 |
#include "tcop/tcopprot.h" |
| 49 |
#include "utils/acl.h" |
| 50 |
#include "utils/builtins.h" |
| 51 |
#include "utils/datum.h" |
| 52 |
#include "utils/fmgroids.h" |
| 53 |
#include "utils/json.h" |
| 54 |
#include "utils/jsonb.h" |
| 55 |
#include "utils/jsonpath.h" |
| 56 |
#include "utils/lsyscache.h" |
| 57 |
#include "utils/memutils.h" |
| 58 |
#include "utils/syscache.h" |
| 59 |
#include "utils/typcache.h" |
| 60 |
|
| 61 |
typedef struct |
| 62 |
{ |
| 63 |
ParamListInfo boundParams; |
| 64 |
PlannerInfo *root; |
| 65 |
List *active_fns; |
| 66 |
Node *case_val; |
| 67 |
bool estimate; |
| 68 |
} eval_const_expressions_context; |
| 69 |
|
| 70 |
typedef struct |
| 71 |
{ |
| 72 |
int nargs; |
| 73 |
List *args; |
| 74 |
int *usecounts; |
| 75 |
} substitute_actual_parameters_context; |
| 76 |
|
| 77 |
typedef struct |
| 78 |
{ |
| 79 |
int nargs; |
| 80 |
List *args; |
| 81 |
int sublevels_up; |
| 82 |
} substitute_actual_srf_parameters_context; |
| 83 |
|
| 84 |
typedef struct |
| 85 |
{ |
| 86 |
char *proname; |
| 87 |
char *prosrc; |
| 88 |
} inline_error_callback_arg; |
| 89 |
|
| 90 |
typedef struct |
| 91 |
{ |
| 92 |
char max_hazard; /* worst proparallel hazard found so far */ |
| 93 |
char max_interesting; /* worst proparallel hazard of interest */ |
| 94 |
List *safe_param_ids; /* PARAM_EXEC Param IDs to treat as safe */ |
| 95 |
} max_parallel_hazard_context; |
| 96 |
|
| 97 |
/* |
| 98 |
* Walker context for expression_has_grouping_conflict. get_eqop is a callback |
| 99 |
* that returns the equality operator used for grouping. cb_context is opaque |
| 100 |
* to the walker and is forwarded to get_eqop unchanged. |
| 101 |
*/ |
| 102 |
typedef struct |
| 103 |
{ |
| 104 |
grouping_eqop_callback get_eqop; |
| 105 |
void *cb_context; |
| 106 |
} grouping_walker_ctx; |
| 107 |
|
| 108 |
static bool contain_agg_clause_walker(Node *node, void *context); |
| 109 |
static bool find_window_functions_walker(Node *node, WindowFuncLists *lists); |
| 110 |
static bool contain_subplans_walker(Node *node, void *context); |
| 111 |
static bool contain_mutable_functions_walker(Node *node, void *context); |
| 112 |
static bool contain_volatile_functions_walker(Node *node, void *context); |
| 113 |
static bool contain_volatile_functions_not_nextval_walker(Node *node, void *context); |
| 114 |
static bool max_parallel_hazard_walker(Node *node, |
| 115 |
max_parallel_hazard_context *context); |
| 116 |
static bool contain_nonstrict_functions_walker(Node *node, void *context); |
| 117 |
static bool contain_exec_param_walker(Node *node, List *param_ids); |
| 118 |
static bool contain_context_dependent_node(Node *clause); |
| 119 |
static bool contain_context_dependent_node_walker(Node *node, int *flags); |
| 120 |
static bool contain_leaked_vars_walker(Node *node, void *context); |
| 121 |
static Relids find_nonnullable_rels_walker(Node *node, bool top_level); |
| 122 |
static List *find_nonnullable_vars_walker(Node *node, bool top_level); |
| 123 |
static bool is_strict_saop(ScalarArrayOpExpr *expr, bool falseOK); |
| 124 |
static bool convert_saop_to_hashed_saop_walker(Node *node, void *context); |
| 125 |
static bool grouping_conflict_walker(Node *node, grouping_walker_ctx *ctx); |
| 126 |
static bool grouping_check_operands(Oid opno, Oid inputcollid, |
| 127 |
List *args, grouping_walker_ctx *ctx); |
| 128 |
static bool grouping_check_operand(Node *arg, Oid opno, Oid inputcollid, |
| 129 |
grouping_walker_ctx *ctx); |
| 130 |
static Node *eval_const_expressions_mutator(Node *node, |
| 131 |
eval_const_expressions_context *context); |
| 132 |
static bool contain_non_const_walker(Node *node, void *context); |
| 133 |
static bool ece_function_is_safe(Oid funcid, |
| 134 |
eval_const_expressions_context *context); |
| 135 |
static List *simplify_or_arguments(List *args, |
| 136 |
eval_const_expressions_context *context, |
| 137 |
bool *haveNull, bool *forceTrue); |
| 138 |
static List *simplify_and_arguments(List *args, |
| 139 |
eval_const_expressions_context *context, |
| 140 |
bool *haveNull, bool *forceFalse); |
| 141 |
static Node *simplify_boolean_equality(Oid opno, List *args); |
| 142 |
static Expr *simplify_function(Oid funcid, |
| 143 |
Oid result_type, int32 result_typmod, |
| 144 |
Oid result_collid, Oid input_collid, List **args_p, |
| 145 |
bool funcvariadic, bool process_args, bool allow_non_const, |
| 146 |
eval_const_expressions_context *context); |
| 147 |
static List *reorder_function_arguments(List *args, int pronargs, |
| 148 |
HeapTuple func_tuple); |
| 149 |
static List *add_function_defaults(List *args, int pronargs, |
| 150 |
HeapTuple func_tuple); |
| 151 |
static List *fetch_function_defaults(HeapTuple func_tuple); |
| 152 |
static void recheck_cast_function_args(List *args, Oid result_type, |
| 153 |
Oid *proargtypes, int pronargs, |
| 154 |
HeapTuple func_tuple); |
| 155 |
static Expr *evaluate_function(Oid funcid, Oid result_type, int32 result_typmod, |
| 156 |
Oid result_collid, Oid input_collid, List *args, |
| 157 |
bool funcvariadic, |
| 158 |
HeapTuple func_tuple, |
| 159 |
eval_const_expressions_context *context); |
| 160 |
static Expr *inline_function(Oid funcid, Oid result_type, Oid result_collid, |
| 161 |
Oid input_collid, List *args, |
| 162 |
bool funcvariadic, |
| 163 |
HeapTuple func_tuple, |
| 164 |
eval_const_expressions_context *context); |
| 165 |
static Node *substitute_actual_parameters(Node *expr, int nargs, List *args, |
| 166 |
int *usecounts); |
| 167 |
static Node *substitute_actual_parameters_mutator(Node *node, |
| 168 |
substitute_actual_parameters_context *context); |
| 169 |
static void sql_inline_error_callback(void *arg); |
| 170 |
static Query *substitute_actual_srf_parameters(Query *expr, |
| 171 |
int nargs, List *args); |
| 172 |
static Node *substitute_actual_srf_parameters_mutator(Node *node, |
| 173 |
substitute_actual_srf_parameters_context *context); |
| 174 |
static bool pull_paramids_walker(Node *node, Bitmapset **context); |
| 175 |
|
| 176 |
|
| 177 |
/***************************************************************************** |
| 178 |
* Aggregate-function clause manipulation |
| 179 |
*****************************************************************************/ |
| 180 |
|
| 181 |
/* |
| 182 |
* contain_agg_clause |
| 183 |
* Recursively search for Aggref/GroupingFunc nodes within a clause. |
| 184 |
* |
| 185 |
* Returns true if any aggregate found. |
| 186 |
* |
| 187 |
* This does not descend into subqueries, and so should be used only after |
| 188 |
* reduction of sublinks to subplans, or in contexts where it's known there |
| 189 |
* are no subqueries. There mustn't be outer-aggregate references either. |
| 190 |
* |
| 191 |
* (If you want something like this but able to deal with subqueries, |
| 192 |
* see rewriteManip.c's contain_aggs_of_level().) |
| 193 |
*/ |
| 194 |
bool |
| 195 |
contain_agg_clause(Node *clause) |
| 196 |
{ |
| 197 |
return contain_agg_clause_walker(clause, NULL); |
| 198 |
} |
| 199 |
|
| 200 |
static bool |
| 201 |
contain_agg_clause_walker(Node *node, void *context) |
| 202 |
{ |
| 203 |
if (node == NULL) |
| 204 |
return false; |
| 205 |
if (IsA(node, Aggref)) |
| 206 |
{ |
| 207 |
Assert(((Aggref *) node)->agglevelsup == 0); |
| 208 |
return true; /* abort the tree traversal and return true */ |
| 209 |
} |
| 210 |
if (IsA(node, GroupingFunc)) |
| 211 |
{ |
| 212 |
Assert(((GroupingFunc *) node)->agglevelsup == 0); |
| 213 |
return true; /* abort the tree traversal and return true */ |
| 214 |
} |
| 215 |
Assert(!IsA(node, SubLink)); |
| 216 |
return expression_tree_walker(node, contain_agg_clause_walker, context); |
| 217 |
} |
| 218 |
|
| 219 |
/***************************************************************************** |
| 220 |
* Window-function clause manipulation |
| 221 |
*****************************************************************************/ |
| 222 |
|
| 223 |
/* |
| 224 |
* contain_window_function |
| 225 |
* Recursively search for WindowFunc nodes within a clause. |
| 226 |
* |
| 227 |
* Since window functions don't have level fields, but are hard-wired to |
| 228 |
* be associated with the current query level, this is just the same as |
| 229 |
* rewriteManip.c's function. |
| 230 |
*/ |
| 231 |
bool |
| 232 |
contain_window_function(Node *clause) |
| 233 |
{ |
| 234 |
return contain_windowfuncs(clause); |
| 235 |
} |
| 236 |
|
| 237 |
/* |
| 238 |
* find_window_functions |
| 239 |
* Locate all the WindowFunc nodes in an expression tree, and organize |
| 240 |
* them by winref ID number. |
| 241 |
* |
| 242 |
* Caller must provide an upper bound on the winref IDs expected in the tree. |
| 243 |
*/ |
| 244 |
WindowFuncLists * |
| 245 |
find_window_functions(Node *clause, Index maxWinRef) |
| 246 |
{ |
| 247 |
WindowFuncLists *lists = palloc(sizeof(WindowFuncLists)); |
| 248 |
|
| 249 |
lists->numWindowFuncs = 0; |
| 250 |
lists->maxWinRef = maxWinRef; |
| 251 |
lists->windowFuncs = (List **) palloc0((maxWinRef + 1) * sizeof(List *)); |
| 252 |
(void) find_window_functions_walker(clause, lists); |
| 253 |
return lists; |
| 254 |
} |
| 255 |
|
| 256 |
static bool |
| 257 |
find_window_functions_walker(Node *node, WindowFuncLists *lists) |
| 258 |
{ |
| 259 |
if (node == NULL) |
| 260 |
return false; |
| 261 |
if (IsA(node, WindowFunc)) |
| 262 |
{ |
| 263 |
WindowFunc *wfunc = (WindowFunc *) node; |
| 264 |
|
| 265 |
/* winref is unsigned, so one-sided test is OK */ |
| 266 |
if (wfunc->winref > lists->maxWinRef) |
| 267 |
elog(ERROR, "WindowFunc contains out-of-range winref %u", |
| 268 |
wfunc->winref); |
| 269 |
|
| 270 |
lists->windowFuncs[wfunc->winref] = |
| 271 |
lappend(lists->windowFuncs[wfunc->winref], wfunc); |
| 272 |
lists->numWindowFuncs++; |
| 273 |
|
| 274 |
/* |
| 275 |
* We assume that the parser checked that there are no window |
| 276 |
* functions in the arguments or filter clause. Hence, we need not |
| 277 |
* recurse into them. (If either the parser or the planner screws up |
| 278 |
* on this point, the executor will still catch it; see ExecInitExpr.) |
| 279 |
*/ |
| 280 |
return false; |
| 281 |
} |
| 282 |
Assert(!IsA(node, SubLink)); |
| 283 |
return expression_tree_walker(node, find_window_functions_walker, lists); |
| 284 |
} |
| 285 |
|
| 286 |
|
| 287 |
/***************************************************************************** |
| 288 |
* Support for expressions returning sets |
| 289 |
*****************************************************************************/ |
| 290 |
|
| 291 |
/* |
| 292 |
* expression_returns_set_rows |
| 293 |
* Estimate the number of rows returned by a set-returning expression. |
| 294 |
* The result is 1 if it's not a set-returning expression. |
| 295 |
* |
| 296 |
* We should only examine the top-level function or operator; it used to be |
| 297 |
* appropriate to recurse, but not anymore. (Even if there are more SRFs in |
| 298 |
* the function's inputs, their multipliers are accounted for separately.) |
| 299 |
* |
| 300 |
* Note: keep this in sync with expression_returns_set() in nodes/nodeFuncs.c. |
| 301 |
*/ |
| 302 |
double |
| 303 |
expression_returns_set_rows(PlannerInfo *root, Node *clause) |
| 304 |
{ |
| 305 |
if (clause == NULL) |
| 306 |
return 1.0; |
| 307 |
if (IsA(clause, FuncExpr)) |
| 308 |
{ |
| 309 |
FuncExpr *expr = (FuncExpr *) clause; |
| 310 |
|
| 311 |
if (expr->funcretset) |
| 312 |
return clamp_row_est(get_function_rows(root, expr->funcid, clause)); |
| 313 |
} |
| 314 |
if (IsA(clause, OpExpr)) |
| 315 |
{ |
| 316 |
OpExpr *expr = (OpExpr *) clause; |
| 317 |
|
| 318 |
if (expr->opretset) |
| 319 |
{ |
| 320 |
set_opfuncid(expr); |
| 321 |
return clamp_row_est(get_function_rows(root, expr->opfuncid, clause)); |
| 322 |
} |
| 323 |
} |
| 324 |
return 1.0; |
| 325 |
} |
| 326 |
|
| 327 |
|
| 328 |
/***************************************************************************** |
| 329 |
* Subplan clause manipulation |
| 330 |
*****************************************************************************/ |
| 331 |
|
| 332 |
/* |
| 333 |
* contain_subplans |
| 334 |
* Recursively search for subplan nodes within a clause. |
| 335 |
* |
| 336 |
* If we see a SubLink node, we will return true. This is only possible if |
| 337 |
* the expression tree hasn't yet been transformed by subselect.c. We do not |
| 338 |
* know whether the node will produce a true subplan or just an initplan, |
| 339 |
* but we make the conservative assumption that it will be a subplan. |
| 340 |
* |
| 341 |
* Returns true if any subplan found. |
| 342 |
*/ |
| 343 |
bool |
| 344 |
contain_subplans(Node *clause) |
| 345 |
{ |
| 346 |
return contain_subplans_walker(clause, NULL); |
| 347 |
} |
| 348 |
|
| 349 |
static bool |
| 350 |
contain_subplans_walker(Node *node, void *context) |
| 351 |
{ |
| 352 |
if (node == NULL) |
| 353 |
return false; |
| 354 |
if (IsA(node, SubPlan) || |
| 355 |
IsA(node, AlternativeSubPlan) || |
| 356 |
IsA(node, SubLink)) |
| 357 |
return true; /* abort the tree traversal and return true */ |
| 358 |
return expression_tree_walker(node, contain_subplans_walker, context); |
| 359 |
} |
| 360 |
|
| 361 |
|
| 362 |
/***************************************************************************** |
| 363 |
* Check clauses for mutable functions |
| 364 |
*****************************************************************************/ |
| 365 |
|
| 366 |
/* |
| 367 |
* contain_mutable_functions |
| 368 |
* Recursively search for mutable functions within a clause. |
| 369 |
* |
| 370 |
* Returns true if any mutable function (or operator implemented by a |
| 371 |
* mutable function) is found. This test is needed so that we don't |
| 372 |
* mistakenly think that something like "WHERE random() < 0.5" can be treated |
| 373 |
* as a constant qualification. |
| 374 |
* |
| 375 |
* This will give the right answer only for clauses that have been put |
| 376 |
* through expression preprocessing. Callers outside the planner typically |
| 377 |
* should use contain_mutable_functions_after_planning() instead, for the |
| 378 |
* reasons given there. |
| 379 |
* |
| 380 |
* We will recursively look into Query nodes (i.e., SubLink sub-selects) |
| 381 |
* but not into SubPlans. See comments for contain_volatile_functions(). |
| 382 |
*/ |
| 383 |
bool |
| 384 |
contain_mutable_functions(Node *clause) |
| 385 |
{ |
| 386 |
return contain_mutable_functions_walker(clause, NULL); |
| 387 |
} |
| 388 |
|
| 389 |
static bool |
| 390 |
contain_mutable_functions_checker(Oid func_id, void *context) |
| 391 |
{ |
| 392 |
return (func_volatile(func_id) != PROVOLATILE_IMMUTABLE); |
| 393 |
} |
| 394 |
|
| 395 |
static bool |
| 396 |
contain_mutable_functions_walker(Node *node, void *context) |
| 397 |
{ |
| 398 |
if (node == NULL) |
| 399 |
return false; |
| 400 |
/* Check for mutable functions in node itself */ |
| 401 |
if (check_functions_in_node(node, contain_mutable_functions_checker, |
| 402 |
context)) |
| 403 |
return true; |
| 404 |
|
| 405 |
if (IsA(node, JsonConstructorExpr)) |
| 406 |
{ |
| 407 |
const JsonConstructorExpr *ctor = (JsonConstructorExpr *) node; |
| 408 |
ListCell *lc; |
| 409 |
bool is_jsonb; |
| 410 |
|
| 411 |
is_jsonb = ctor->returning->format->format_type == JS_FORMAT_JSONB; |
| 412 |
|
| 413 |
/* |
| 414 |
* Check argument_type => json[b] conversions specifically. We still |
| 415 |
* recurse to check 'args' below, but here we want to specifically |
| 416 |
* check whether or not the emitted clause would fail to be immutable |
| 417 |
* because of TimeZone, for example. |
| 418 |
*/ |
| 419 |
foreach(lc, ctor->args) |
| 420 |
{ |
| 421 |
Oid typid = exprType(lfirst(lc)); |
| 422 |
|
| 423 |
if (is_jsonb ? |
| 424 |
!to_jsonb_is_immutable(typid) : |
| 425 |
!to_json_is_immutable(typid)) |
| 426 |
return true; |
| 427 |
} |
| 428 |
|
| 429 |
/* Check all subnodes */ |
| 430 |
} |
| 431 |
|
| 432 |
if (IsA(node, JsonExpr)) |
| 433 |
{ |
| 434 |
JsonExpr *jexpr = castNode(JsonExpr, node); |
| 435 |
Const *cnst; |
| 436 |
|
| 437 |
if (!IsA(jexpr->path_spec, Const)) |
| 438 |
return true; |
| 439 |
|
| 440 |
cnst = castNode(Const, jexpr->path_spec); |
| 441 |
|
| 442 |
Assert(cnst->consttype == JSONPATHOID); |
| 443 |
if (cnst->constisnull) |
| 444 |
return false; |
| 445 |
|
| 446 |
if (jspIsMutable(DatumGetJsonPathP(cnst->constvalue), |
| 447 |
jexpr->passing_names, jexpr->passing_values)) |
| 448 |
return true; |
| 449 |
} |
| 450 |
|
| 451 |
if (IsA(node, SQLValueFunction)) |
| 452 |
{ |
| 453 |
/* all variants of SQLValueFunction are stable */ |
| 454 |
return true; |
| 455 |
} |
| 456 |
|
| 457 |
if (IsA(node, NextValueExpr)) |
| 458 |
{ |
| 459 |
/* NextValueExpr is volatile */ |
| 460 |
return true; |
| 461 |
} |
| 462 |
|
| 463 |
/* |
| 464 |
* It should be safe to treat MinMaxExpr as immutable, because it will |
| 465 |
* depend on a non-cross-type btree comparison function, and those should |
| 466 |
* always be immutable. Treating XmlExpr as immutable is more dubious, |
| 467 |
* and treating CoerceToDomain as immutable is outright dangerous. But we |
| 468 |
* have done so historically, and changing this would probably cause more |
| 469 |
* problems than it would fix. In practice, if you have a non-immutable |
| 470 |
* domain constraint you are in for pain anyhow. |
| 471 |
*/ |
| 472 |
|
| 473 |
/* Recurse to check arguments */ |
| 474 |
if (IsA(node, Query)) |
| 475 |
{ |
| 476 |
/* Recurse into subselects */ |
| 477 |
return query_tree_walker((Query *) node, |
| 478 |
contain_mutable_functions_walker, |
| 479 |
context, 0); |
| 480 |
} |
| 481 |
return expression_tree_walker(node, contain_mutable_functions_walker, |
| 482 |
context); |
| 483 |
} |
| 484 |
|
| 485 |
/* |
| 486 |
* contain_mutable_functions_after_planning |
| 487 |
* Test whether given expression contains mutable functions. |
| 488 |
* |
| 489 |
* This is a wrapper for contain_mutable_functions() that is safe to use from |
| 490 |
* outside the planner. The difference is that it first runs the expression |
| 491 |
* through expression_planner(). There are two key reasons why we need that: |
| 492 |
* |
| 493 |
* First, function default arguments will get inserted, which may affect |
| 494 |
* volatility (consider "default now()"). |
| 495 |
* |
| 496 |
* Second, inline-able functions will get inlined, which may allow us to |
| 497 |
* conclude that the function is really less volatile than it's marked. |
| 498 |
* As an example, polymorphic functions must be marked with the most volatile |
| 499 |
* behavior that they have for any input type, but once we inline the |
| 500 |
* function we may be able to conclude that it's not so volatile for the |
| 501 |
* particular input type we're dealing with. |
| 502 |
*/ |
| 503 |
bool |
| 504 |
contain_mutable_functions_after_planning(Expr *expr) |
| 505 |
{ |
| 506 |
/* We assume here that expression_planner() won't scribble on its input */ |
| 507 |
expr = expression_planner(expr); |
| 508 |
|
| 509 |
/* Now we can search for non-immutable functions */ |
| 510 |
return contain_mutable_functions((Node *) expr); |
| 511 |
} |
| 512 |
|
| 513 |
|
| 514 |
/***************************************************************************** |
| 515 |
* Check clauses for volatile functions |
| 516 |
*****************************************************************************/ |
| 517 |
|
| 518 |
/* |
| 519 |
* contain_volatile_functions |
| 520 |
* Recursively search for volatile functions within a clause. |
| 521 |
* |
| 522 |
* Returns true if any volatile function (or operator implemented by a |
| 523 |
* volatile function) is found. This test prevents, for example, |
| 524 |
* invalid conversions of volatile expressions into indexscan quals. |
| 525 |
* |
| 526 |
* This will give the right answer only for clauses that have been put |
| 527 |
* through expression preprocessing. Callers outside the planner typically |
| 528 |
* should use contain_volatile_functions_after_planning() instead, for the |
| 529 |
* reasons given there. |
| 530 |
* |
| 531 |
* We will recursively look into Query nodes (i.e., SubLink sub-selects) |
| 532 |
* but not into SubPlans. This is a bit odd, but intentional. If we are |
| 533 |
* looking at a SubLink, we are probably deciding whether a query tree |
| 534 |
* transformation is safe, and a contained sub-select should affect that; |
| 535 |
* for example, duplicating a sub-select containing a volatile function |
| 536 |
* would be bad. However, once we've got to the stage of having SubPlans, |
| 537 |
* subsequent planning need not consider volatility within those, since |
| 538 |
* the executor won't change its evaluation rules for a SubPlan based on |
| 539 |
* volatility. |
| 540 |
* |
| 541 |
* For some node types, for example, RestrictInfo and PathTarget, we cache |
| 542 |
* whether we found any volatile functions or not and reuse that value in any |
| 543 |
* future checks for that node. All of the logic for determining if the |
| 544 |
* cached value should be set to VOLATILITY_NOVOLATILE or VOLATILITY_VOLATILE |
| 545 |
* belongs in this function. Any code which makes changes to these nodes |
| 546 |
* which could change the outcome this function must set the cached value back |
| 547 |
* to VOLATILITY_UNKNOWN. That allows this function to redetermine the |
| 548 |
* correct value during the next call, should we need to redetermine if the |
| 549 |
* node contains any volatile functions again in the future. |
| 550 |
*/ |
| 551 |
bool |
| 552 |
contain_volatile_functions(Node *clause) |
| 553 |
{ |
| 554 |
return contain_volatile_functions_walker(clause, NULL); |
| 555 |
} |
| 556 |
|
| 557 |
static bool |
| 558 |
contain_volatile_functions_checker(Oid func_id, void *context) |
| 559 |
{ |
| 560 |
return (func_volatile(func_id) == PROVOLATILE_VOLATILE); |
| 561 |
} |
| 562 |
|
| 563 |
static bool |
| 564 |
contain_volatile_functions_walker(Node *node, void *context) |
| 565 |
{ |
| 566 |
if (node == NULL) |
| 567 |
return false; |
| 568 |
/* Check for volatile functions in node itself */ |
| 569 |
if (check_functions_in_node(node, contain_volatile_functions_checker, |
| 570 |
context)) |
| 571 |
return true; |
| 572 |
|
| 573 |
if (IsA(node, NextValueExpr)) |
| 574 |
{ |
| 575 |
/* NextValueExpr is volatile */ |
| 576 |
return true; |
| 577 |
} |
| 578 |
|
| 579 |
if (IsA(node, RestrictInfo)) |
| 580 |
{ |
| 581 |
RestrictInfo *rinfo = (RestrictInfo *) node; |
| 582 |
|
| 583 |
/* |
| 584 |
* For RestrictInfo, check if we've checked the volatility of it |
| 585 |
* before. If so, we can just use the cached value and not bother |
| 586 |
* checking it again. Otherwise, check it and cache if whether we |
| 587 |
* found any volatile functions. |
| 588 |
*/ |
| 589 |
if (rinfo->has_volatile == VOLATILITY_NOVOLATILE) |
| 590 |
return false; |
| 591 |
else if (rinfo->has_volatile == VOLATILITY_VOLATILE) |
| 592 |
return true; |
| 593 |
else |
| 594 |
{ |
| 595 |
bool hasvolatile; |
| 596 |
|
| 597 |
hasvolatile = contain_volatile_functions_walker((Node *) rinfo->clause, |
| 598 |
context); |
| 599 |
if (hasvolatile) |
| 600 |
rinfo->has_volatile = VOLATILITY_VOLATILE; |
| 601 |
else |
| 602 |
rinfo->has_volatile = VOLATILITY_NOVOLATILE; |
| 603 |
|
| 604 |
return hasvolatile; |
| 605 |
} |
| 606 |
} |
| 607 |
|
| 608 |
if (IsA(node, PathTarget)) |
| 609 |
{ |
| 610 |
PathTarget *target = (PathTarget *) node; |
| 611 |
|
| 612 |
/* |
| 613 |
* We also do caching for PathTarget the same as we do above for |
| 614 |
* RestrictInfos. |
| 615 |
*/ |
| 616 |
if (target->has_volatile_expr == VOLATILITY_NOVOLATILE) |
| 617 |
return false; |
| 618 |
else if (target->has_volatile_expr == VOLATILITY_VOLATILE) |
| 619 |
return true; |
| 620 |
else |
| 621 |
{ |
| 622 |
bool hasvolatile; |
| 623 |
|
| 624 |
hasvolatile = contain_volatile_functions_walker((Node *) target->exprs, |
| 625 |
context); |
| 626 |
|
| 627 |
if (hasvolatile) |
| 628 |
target->has_volatile_expr = VOLATILITY_VOLATILE; |
| 629 |
else |
| 630 |
target->has_volatile_expr = VOLATILITY_NOVOLATILE; |
| 631 |
|
| 632 |
return hasvolatile; |
| 633 |
} |
| 634 |
} |
| 635 |
|
| 636 |
/* |
| 637 |
* See notes in contain_mutable_functions_walker about why we treat |
| 638 |
* MinMaxExpr, XmlExpr, and CoerceToDomain as immutable, while |
| 639 |
* SQLValueFunction is stable. Hence, none of them are of interest here. |
| 640 |
*/ |
| 641 |
|
| 642 |
/* Recurse to check arguments */ |
| 643 |
if (IsA(node, Query)) |
| 644 |
{ |
| 645 |
/* Recurse into subselects */ |
| 646 |
return query_tree_walker((Query *) node, |
| 647 |
contain_volatile_functions_walker, |
| 648 |
context, 0); |
| 649 |
} |
| 650 |
return expression_tree_walker(node, contain_volatile_functions_walker, |
| 651 |
context); |
| 652 |
} |
| 653 |
|
| 654 |
/* |
| 655 |
* contain_volatile_functions_after_planning |
| 656 |
* Test whether given expression contains volatile functions. |
| 657 |
* |
| 658 |
* This is a wrapper for contain_volatile_functions() that is safe to use from |
| 659 |
* outside the planner. The difference is that it first runs the expression |
| 660 |
* through expression_planner(). There are two key reasons why we need that: |
| 661 |
* |
| 662 |
* First, function default arguments will get inserted, which may affect |
| 663 |
* volatility (consider "default random()"). |
| 664 |
* |
| 665 |
* Second, inline-able functions will get inlined, which may allow us to |
| 666 |
* conclude that the function is really less volatile than it's marked. |
| 667 |
* As an example, polymorphic functions must be marked with the most volatile |
| 668 |
* behavior that they have for any input type, but once we inline the |
| 669 |
* function we may be able to conclude that it's not so volatile for the |
| 670 |
* particular input type we're dealing with. |
| 671 |
*/ |
| 672 |
bool |
| 673 |
contain_volatile_functions_after_planning(Expr *expr) |
| 674 |
{ |
| 675 |
/* We assume here that expression_planner() won't scribble on its input */ |
| 676 |
expr = expression_planner(expr); |
| 677 |
|
| 678 |
/* Now we can search for volatile functions */ |
| 679 |
return contain_volatile_functions((Node *) expr); |
| 680 |
} |
| 681 |
|
| 682 |
/* |
| 683 |
* Special purpose version of contain_volatile_functions() for use in COPY: |
| 684 |
* ignore nextval(), but treat all other functions normally. |
| 685 |
*/ |
| 686 |
bool |
| 687 |
contain_volatile_functions_not_nextval(Node *clause) |
| 688 |
{ |
| 689 |
return contain_volatile_functions_not_nextval_walker(clause, NULL); |
| 690 |
} |
| 691 |
|
| 692 |
static bool |
| 693 |
contain_volatile_functions_not_nextval_checker(Oid func_id, void *context) |
| 694 |
{ |
| 695 |
return (func_id != F_NEXTVAL && |
| 696 |
func_volatile(func_id) == PROVOLATILE_VOLATILE); |
| 697 |
} |
| 698 |
|
| 699 |
static bool |
| 700 |
contain_volatile_functions_not_nextval_walker(Node *node, void *context) |
| 701 |
{ |
| 702 |
if (node == NULL) |
| 703 |
return false; |
| 704 |
/* Check for volatile functions in node itself */ |
| 705 |
if (check_functions_in_node(node, |
| 706 |
contain_volatile_functions_not_nextval_checker, |
| 707 |
context)) |
| 708 |
return true; |
| 709 |
|
| 710 |
/* |
| 711 |
* See notes in contain_mutable_functions_walker about why we treat |
| 712 |
* MinMaxExpr, XmlExpr, and CoerceToDomain as immutable, while |
| 713 |
* SQLValueFunction is stable. Hence, none of them are of interest here. |
| 714 |
* Also, since we're intentionally ignoring nextval(), presumably we |
| 715 |
* should ignore NextValueExpr. |
| 716 |
*/ |
| 717 |
|
| 718 |
/* Recurse to check arguments */ |
| 719 |
if (IsA(node, Query)) |
| 720 |
{ |
| 721 |
/* Recurse into subselects */ |
| 722 |
return query_tree_walker((Query *) node, |
| 723 |
contain_volatile_functions_not_nextval_walker, |
| 724 |
context, 0); |
| 725 |
} |
| 726 |
return expression_tree_walker(node, |
| 727 |
contain_volatile_functions_not_nextval_walker, |
| 728 |
context); |
| 729 |
} |
| 730 |
|
| 731 |
|
| 732 |
/***************************************************************************** |
| 733 |
* Check queries for parallel unsafe and/or restricted constructs |
| 734 |
*****************************************************************************/ |
| 735 |
|
| 736 |
/* |
| 737 |
* max_parallel_hazard |
| 738 |
* Find the worst parallel-hazard level in the given query |
| 739 |
* |
| 740 |
* Returns the worst function hazard property (the earliest in this list: |
| 741 |
* PROPARALLEL_UNSAFE, PROPARALLEL_RESTRICTED, PROPARALLEL_SAFE) that can |
| 742 |
* be found in the given parsetree. We use this to find out whether the query |
| 743 |
* can be parallelized at all. The caller will also save the result in |
| 744 |
* PlannerGlobal so as to short-circuit checks of portions of the querytree |
| 745 |
* later, in the common case where everything is SAFE. |
| 746 |
*/ |
| 747 |
char |
| 748 |
max_parallel_hazard(Query *parse) |
| 749 |
{ |
| 750 |
max_parallel_hazard_context context; |
| 751 |
|
| 752 |
context.max_hazard = PROPARALLEL_SAFE; |
| 753 |
context.max_interesting = PROPARALLEL_UNSAFE; |
| 754 |
context.safe_param_ids = NIL; |
| 755 |
(void) max_parallel_hazard_walker((Node *) parse, &context); |
| 756 |
return context.max_hazard; |
| 757 |
} |
| 758 |
|
| 759 |
/* |
| 760 |
* is_parallel_safe |
| 761 |
* Detect whether the given expr contains only parallel-safe functions |
| 762 |
* |
| 763 |
* root->glob->maxParallelHazard must previously have been set to the |
| 764 |
* result of max_parallel_hazard() on the whole query. |
| 765 |
*/ |
| 766 |
bool |
| 767 |
is_parallel_safe(PlannerInfo *root, Node *node) |
| 768 |
{ |
| 769 |
max_parallel_hazard_context context; |
| 770 |
PlannerInfo *proot; |
| 771 |
ListCell *l; |
| 772 |
|
| 773 |
/* |
| 774 |
* Even if the original querytree contained nothing unsafe, we need to |
| 775 |
* search the expression if we have generated any PARAM_EXEC Params while |
| 776 |
* planning, because those are parallel-restricted and there might be one |
| 777 |
* in this expression. But otherwise we don't need to look. |
| 778 |
*/ |
| 779 |
if (root->glob->maxParallelHazard == PROPARALLEL_SAFE && |
| 780 |
root->glob->paramExecTypes == NIL) |
| 781 |
return true; |
| 782 |
/* Else use max_parallel_hazard's search logic, but stop on RESTRICTED */ |
| 783 |
context.max_hazard = PROPARALLEL_SAFE; |
| 784 |
context.max_interesting = PROPARALLEL_RESTRICTED; |
| 785 |
context.safe_param_ids = NIL; |
| 786 |
|
| 787 |
/* |
| 788 |
* The params that refer to the same or parent query level are considered |
| 789 |
* parallel-safe. The idea is that we compute such params at Gather or |
| 790 |
* Gather Merge node and pass their value to workers. |
| 791 |
*/ |
| 792 |
for (proot = root; proot != NULL; proot = proot->parent_root) |
| 793 |
{ |
| 794 |
foreach(l, proot->init_plans) |
| 795 |
{ |
| 796 |
SubPlan *initsubplan = (SubPlan *) lfirst(l); |
| 797 |
|
| 798 |
context.safe_param_ids = list_concat(context.safe_param_ids, |
| 799 |
initsubplan->setParam); |
| 800 |
} |
| 801 |
} |
| 802 |
|
| 803 |
return !max_parallel_hazard_walker(node, &context); |
| 804 |
} |
| 805 |
|
| 806 |
/* core logic for all parallel-hazard checks */ |
| 807 |
static bool |
| 808 |
max_parallel_hazard_test(char proparallel, max_parallel_hazard_context *context) |
| 809 |
{ |
| 810 |
switch (proparallel) |
| 811 |
{ |
| 812 |
case PROPARALLEL_SAFE: |
| 813 |
/* nothing to see here, move along */ |
| 814 |
break; |
| 815 |
case PROPARALLEL_RESTRICTED: |
| 816 |
/* increase max_hazard to RESTRICTED */ |
| 817 |
Assert(context->max_hazard != PROPARALLEL_UNSAFE); |
| 818 |
context->max_hazard = proparallel; |
| 819 |
/* done if we are not expecting any unsafe functions */ |
| 820 |
if (context->max_interesting == proparallel) |
| 821 |
return true; |
| 822 |
break; |
| 823 |
case PROPARALLEL_UNSAFE: |
| 824 |
context->max_hazard = proparallel; |
| 825 |
/* we're always done at the first unsafe construct */ |
| 826 |
return true; |
| 827 |
default: |
| 828 |
elog(ERROR, "unrecognized proparallel value \"%c\"", proparallel); |
| 829 |
break; |
| 830 |
} |
| 831 |
return false; |
| 832 |
} |
| 833 |
|
| 834 |
/* check_functions_in_node callback */ |
| 835 |
static bool |
| 836 |
max_parallel_hazard_checker(Oid func_id, void *context) |
| 837 |
{ |
| 838 |
return max_parallel_hazard_test(func_parallel(func_id), |
| 839 |
(max_parallel_hazard_context *) context); |
| 840 |
} |
| 841 |
|
| 842 |
static bool |
| 843 |
max_parallel_hazard_walker(Node *node, max_parallel_hazard_context *context) |
| 844 |
{ |
| 845 |
if (node == NULL) |
| 846 |
return false; |
| 847 |
|
| 848 |
/* Check for hazardous functions in node itself */ |
| 849 |
if (check_functions_in_node(node, max_parallel_hazard_checker, |
| 850 |
context)) |
| 851 |
return true; |
| 852 |
|
| 853 |
/* |
| 854 |
* It should be OK to treat MinMaxExpr as parallel-safe, since btree |
| 855 |
* opclass support functions are generally parallel-safe. XmlExpr is a |
| 856 |
* bit more dubious but we can probably get away with it. We err on the |
| 857 |
* side of caution by treating CoerceToDomain as parallel-restricted. |
| 858 |
* (Note: in principle that's wrong because a domain constraint could |
| 859 |
* contain a parallel-unsafe function; but useful constraints probably |
| 860 |
* never would have such, and assuming they do would cripple use of |
| 861 |
* parallel query in the presence of domain types.) SQLValueFunction |
| 862 |
* should be safe in all cases. NextValueExpr is parallel-unsafe. |
| 863 |
*/ |
| 864 |
if (IsA(node, CoerceToDomain)) |
| 865 |
{ |
| 866 |
if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context)) |
| 867 |
return true; |
| 868 |
} |
| 869 |
|
| 870 |
else if (IsA(node, NextValueExpr)) |
| 871 |
{ |
| 872 |
if (max_parallel_hazard_test(PROPARALLEL_UNSAFE, context)) |
| 873 |
return true; |
| 874 |
} |
| 875 |
|
| 876 |
/* |
| 877 |
* Treat window functions as parallel-restricted because we aren't sure |
| 878 |
* whether the input row ordering is fully deterministic, and the output |
| 879 |
* of window functions might vary across workers if not. (In some cases, |
| 880 |
* like where the window frame orders by a primary key, we could relax |
| 881 |
* this restriction. But it doesn't currently seem worth expending extra |
| 882 |
* effort to do so.) |
| 883 |
*/ |
| 884 |
else if (IsA(node, WindowFunc)) |
| 885 |
{ |
| 886 |
if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context)) |
| 887 |
return true; |
| 888 |
} |
| 889 |
|
| 890 |
/* |
| 891 |
* As a notational convenience for callers, look through RestrictInfo. |
| 892 |
*/ |
| 893 |
else if (IsA(node, RestrictInfo)) |
| 894 |
{ |
| 895 |
RestrictInfo *rinfo = (RestrictInfo *) node; |
| 896 |
|
| 897 |
return max_parallel_hazard_walker((Node *) rinfo->clause, context); |
| 898 |
} |
| 899 |
|
| 900 |
/* |
| 901 |
* Really we should not see SubLink during a max_interesting == restricted |
| 902 |
* scan, but if we do, return true. |
| 903 |
*/ |
| 904 |
else if (IsA(node, SubLink)) |
| 905 |
{ |
| 906 |
if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context)) |
| 907 |
return true; |
| 908 |
} |
| 909 |
|
| 910 |
/* |
| 911 |
* Only parallel-safe SubPlans can be sent to workers. Within the |
| 912 |
* testexpr of the SubPlan, Params representing the output columns of the |
| 913 |
* subplan can be treated as parallel-safe, so temporarily add their IDs |
| 914 |
* to the safe_param_ids list while examining the testexpr. |
| 915 |
*/ |
| 916 |
else if (IsA(node, SubPlan)) |
| 917 |
{ |
| 918 |
SubPlan *subplan = (SubPlan *) node; |
| 919 |
List *save_safe_param_ids; |
| 920 |
|
| 921 |
if (!subplan->parallel_safe && |
| 922 |
max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context)) |
| 923 |
return true; |
| 924 |
save_safe_param_ids = context->safe_param_ids; |
| 925 |
context->safe_param_ids = list_concat_copy(context->safe_param_ids, |
| 926 |
subplan->paramIds); |
| 927 |
if (max_parallel_hazard_walker(subplan->testexpr, context)) |
| 928 |
return true; /* no need to restore safe_param_ids */ |
| 929 |
list_free(context->safe_param_ids); |
| 930 |
context->safe_param_ids = save_safe_param_ids; |
| 931 |
/* we must also check args, but no special Param treatment there */ |
| 932 |
if (max_parallel_hazard_walker((Node *) subplan->args, context)) |
| 933 |
return true; |
| 934 |
/* don't want to recurse normally, so we're done */ |
| 935 |
return false; |
| 936 |
} |
| 937 |
|
| 938 |
/* |
| 939 |
* We can't pass Params to workers at the moment either, so they are also |
| 940 |
* parallel-restricted, unless they are PARAM_EXTERN Params or are |
| 941 |
* PARAM_EXEC Params listed in safe_param_ids, meaning they could be |
| 942 |
* either generated within workers or can be computed by the leader and |
| 943 |
* then their value can be passed to workers. |
| 944 |
*/ |
| 945 |
else if (IsA(node, Param)) |
| 946 |
{ |
| 947 |
Param *param = (Param *) node; |
| 948 |
|
| 949 |
if (param->paramkind == PARAM_EXTERN) |
| 950 |
return false; |
| 951 |
|
| 952 |
if (param->paramkind != PARAM_EXEC || |
| 953 |
!list_member_int(context->safe_param_ids, param->paramid)) |
| 954 |
{ |
| 955 |
if (max_parallel_hazard_test(PROPARALLEL_RESTRICTED, context)) |
| 956 |
return true; |
| 957 |
} |
| 958 |
return false; /* nothing to recurse to */ |
| 959 |
} |
| 960 |
|
| 961 |
/* |
| 962 |
* When we're first invoked on a completely unplanned tree, we must |
| 963 |
* recurse into subqueries so to as to locate parallel-unsafe constructs |
| 964 |
* anywhere in the tree. |
| 965 |
*/ |
| 966 |
else if (IsA(node, Query)) |
| 967 |
{ |
| 968 |
Query *query = (Query *) node; |
| 969 |
|
| 970 |
/* SELECT FOR UPDATE/SHARE must be treated as unsafe */ |
| 971 |
if (query->rowMarks != NULL) |
| 972 |
{ |
| 973 |
context->max_hazard = PROPARALLEL_UNSAFE; |
| 974 |
return true; |
| 975 |
} |
| 976 |
|
| 977 |
/* Recurse into subselects */ |
| 978 |
return query_tree_walker(query, |
| 979 |
max_parallel_hazard_walker, |
| 980 |
context, 0); |
| 981 |
} |
| 982 |
|
| 983 |
/* Recurse to check arguments */ |
| 984 |
return expression_tree_walker(node, |
| 985 |
max_parallel_hazard_walker, |
| 986 |
context); |
| 987 |
} |
| 988 |
|
| 989 |
|
| 990 |
/***************************************************************************** |
| 991 |
* Check clauses for nonstrict functions |
| 992 |
*****************************************************************************/ |
| 993 |
|
| 994 |
/* |
| 995 |
* contain_nonstrict_functions |
| 996 |
* Recursively search for nonstrict functions within a clause. |
| 997 |
* |
| 998 |
* Returns true if any nonstrict construct is found --- ie, anything that |
| 999 |
* could produce non-NULL output with a NULL input. |
| 1000 |
* |
| 1001 |
* The idea here is that the caller has verified that the expression contains |
| 1002 |
* one or more Var or Param nodes (as appropriate for the caller's need), and |
| 1003 |
* now wishes to prove that the expression result will be NULL if any of these |
| 1004 |
* inputs is NULL. If we return false, then the proof succeeded. |
| 1005 |
*/ |
| 1006 |
bool |
| 1007 |
contain_nonstrict_functions(Node *clause) |
| 1008 |
{ |
| 1009 |
return contain_nonstrict_functions_walker(clause, NULL); |
| 1010 |
} |
| 1011 |
|
| 1012 |
static bool |
| 1013 |
contain_nonstrict_functions_checker(Oid func_id, void *context) |
| 1014 |
{ |
| 1015 |
return !func_strict(func_id); |
| 1016 |
} |
| 1017 |
|
| 1018 |
static bool |
| 1019 |
contain_nonstrict_functions_walker(Node *node, void *context) |
| 1020 |
{ |
| 1021 |
if (node == NULL) |
| 1022 |
return false; |
| 1023 |
if (IsA(node, Aggref)) |
| 1024 |
{ |
| 1025 |
/* an aggregate could return non-null with null input */ |
| 1026 |
return true; |
| 1027 |
} |
| 1028 |
else if (IsA(node, GroupingFunc)) |
| 1029 |
{ |
| 1030 |
/* |
| 1031 |
* A GroupingFunc doesn't evaluate its arguments, and therefore must |
| 1032 |
* be treated as nonstrict. |
| 1033 |
*/ |
| 1034 |
return true; |
| 1035 |
} |
| 1036 |
else if (IsA(node, WindowFunc)) |
| 1037 |
{ |
| 1038 |
/* a window function could return non-null with null input */ |
| 1039 |
return true; |
| 1040 |
} |
| 1041 |
else if (IsA(node, SubscriptingRef)) |
| 1042 |
{ |
| 1043 |
SubscriptingRef *sbsref = (SubscriptingRef *) node; |
| 1044 |
const SubscriptRoutines *sbsroutines; |
| 1045 |
|
| 1046 |
/* Subscripting assignment is always presumed nonstrict */ |
| 1047 |
if (sbsref->refassgnexpr != NULL) |
| 1048 |
return true; |
| 1049 |
/* Otherwise we must look up the subscripting support methods */ |
| 1050 |
sbsroutines = getSubscriptingRoutines(sbsref->refcontainertype, NULL); |
| 1051 |
if (!(sbsroutines && sbsroutines->fetch_strict)) |
| 1052 |
return true; |
| 1053 |
/* else fall through to check args */ |
| 1054 |
} |
| 1055 |
else if (IsA(node, DistinctExpr)) |
| 1056 |
{ |
| 1057 |
/* IS DISTINCT FROM is inherently non-strict */ |
| 1058 |
return true; |
| 1059 |
} |
| 1060 |
else if (IsA(node, NullIfExpr)) |
| 1061 |
{ |
| 1062 |
/* NULLIF is inherently non-strict */ |
| 1063 |
return true; |
| 1064 |
} |
| 1065 |
else if (IsA(node, ScalarArrayOpExpr)) |
| 1066 |
{ |
| 1067 |
ScalarArrayOpExpr *expr = (ScalarArrayOpExpr *) node; |
| 1068 |
|
| 1069 |
if (!is_strict_saop(expr, false)) |
| 1070 |
return true; |
| 1071 |
/* else fall through to check args */ |
| 1072 |
} |
| 1073 |
else if (IsA(node, BoolExpr)) |
| 1074 |
{ |
| 1075 |
BoolExpr *expr = (BoolExpr *) node; |
| 1076 |
|
| 1077 |
switch (expr->boolop) |
| 1078 |
{ |
| 1079 |
case AND_EXPR: |
| 1080 |
case OR_EXPR: |
| 1081 |
/* AND, OR are inherently non-strict */ |
| 1082 |
return true; |
| 1083 |
default: |
| 1084 |
break; |
| 1085 |
} |
| 1086 |
} |
| 1087 |
else if (IsA(node, SubLink)) |
| 1088 |
{ |
| 1089 |
/* In some cases a sublink might be strict, but in general not */ |
| 1090 |
return true; |
| 1091 |
} |
| 1092 |
else if (IsA(node, SubPlan)) |
| 1093 |
return true; |
| 1094 |
else if (IsA(node, AlternativeSubPlan)) |
| 1095 |
return true; |
| 1096 |
else if (IsA(node, FieldStore)) |
| 1097 |
return true; |
| 1098 |
else if (IsA(node, CoerceViaIO)) |
| 1099 |
{ |
| 1100 |
/* |
| 1101 |
* CoerceViaIO is strict regardless of whether the I/O functions are, |
| 1102 |
* so we should skip check_functions_in_node() and just fall through |
| 1103 |
* to check the arguments. |
| 1104 |
*/ |
| 1105 |
} |
| 1106 |
else if (IsA(node, ArrayCoerceExpr)) |
| 1107 |
{ |
| 1108 |
/* |
| 1109 |
* ArrayCoerceExpr is strict at the array level, regardless of what |
| 1110 |
* the per-element expression is; so we should ignore elemexpr and |
| 1111 |
* recurse only into the arg. |
| 1112 |
*/ |
| 1113 |
return contain_nonstrict_functions_walker((Node *) ((ArrayCoerceExpr *) node)->arg, |
| 1114 |
context); |
| 1115 |
} |
| 1116 |
else if (IsA(node, CaseExpr)) |
| 1117 |
return true; |
| 1118 |
else if (IsA(node, ArrayExpr)) |
| 1119 |
return true; |
| 1120 |
else if (IsA(node, RowExpr)) |
| 1121 |
return true; |
| 1122 |
else if (IsA(node, RowCompareExpr)) |
| 1123 |
return true; |
| 1124 |
else if (IsA(node, CoalesceExpr)) |
| 1125 |
return true; |
| 1126 |
else if (IsA(node, MinMaxExpr)) |
| 1127 |
return true; |
| 1128 |
else if (IsA(node, XmlExpr)) |
| 1129 |
return true; |
| 1130 |
else if (IsA(node, NullTest)) |
| 1131 |
return true; |
| 1132 |
else if (IsA(node, BooleanTest)) |
| 1133 |
return true; |
| 1134 |
else if (IsA(node, JsonConstructorExpr)) |
| 1135 |
return true; |
| 1136 |
else |
| 1137 |
{ |
| 1138 |
/* Check other function-containing nodes */ |
| 1139 |
if (check_functions_in_node(node, contain_nonstrict_functions_checker, |
| 1140 |
context)) |
| 1141 |
return true; |
| 1142 |
} |
| 1143 |
|
| 1144 |
return expression_tree_walker(node, contain_nonstrict_functions_walker, |
| 1145 |
context); |
| 1146 |
} |
| 1147 |
|
| 1148 |
/***************************************************************************** |
| 1149 |
* Check clauses for Params |
| 1150 |
*****************************************************************************/ |
| 1151 |
|
| 1152 |
/* |
| 1153 |
* contain_exec_param |
| 1154 |
* Recursively search for PARAM_EXEC Params within a clause. |
| 1155 |
* |
| 1156 |
* Returns true if the clause contains any PARAM_EXEC Param with a paramid |
| 1157 |
* appearing in the given list of Param IDs. Does not descend into |
| 1158 |
* subqueries! |
| 1159 |
*/ |
| 1160 |
bool |
| 1161 |
contain_exec_param(Node *clause, List *param_ids) |
| 1162 |
{ |
| 1163 |
return contain_exec_param_walker(clause, param_ids); |
| 1164 |
} |
| 1165 |
|
| 1166 |
static bool |
| 1167 |
contain_exec_param_walker(Node *node, List *param_ids) |
| 1168 |
{ |
| 1169 |
if (node == NULL) |
| 1170 |
return false; |
| 1171 |
if (IsA(node, Param)) |
| 1172 |
{ |
| 1173 |
Param *p = (Param *) node; |
| 1174 |
|
| 1175 |
if (p->paramkind == PARAM_EXEC && |
| 1176 |
list_member_int(param_ids, p->paramid)) |
| 1177 |
return true; |
| 1178 |
} |
| 1179 |
return expression_tree_walker(node, contain_exec_param_walker, param_ids); |
| 1180 |
} |
| 1181 |
|
| 1182 |
/***************************************************************************** |
| 1183 |
* Check clauses for context-dependent nodes |
| 1184 |
*****************************************************************************/ |
| 1185 |
|
| 1186 |
/* |
| 1187 |
* contain_context_dependent_node |
| 1188 |
* Recursively search for context-dependent nodes within a clause. |
| 1189 |
* |
| 1190 |
* CaseTestExpr nodes must appear directly within the corresponding CaseExpr, |
| 1191 |
* not nested within another one, or they'll see the wrong test value. If one |
| 1192 |
* appears "bare" in the arguments of a SQL function, then we can't inline the |
| 1193 |
* SQL function for fear of creating such a situation. The same applies for |
| 1194 |
* CaseTestExpr used within the elemexpr of an ArrayCoerceExpr. |
| 1195 |
* |
| 1196 |
* CoerceToDomainValue would have the same issue if domain CHECK expressions |
| 1197 |
* could get inlined into larger expressions, but presently that's impossible. |
| 1198 |
* Still, it might be allowed in future, or other node types with similar |
| 1199 |
* issues might get invented. So give this function a generic name, and set |
| 1200 |
* up the recursion state to allow multiple flag bits. |
| 1201 |
*/ |
| 1202 |
static bool |
| 1203 |
contain_context_dependent_node(Node *clause) |
| 1204 |
{ |
| 1205 |
int flags = 0; |
| 1206 |
|
| 1207 |
return contain_context_dependent_node_walker(clause, &flags); |
| 1208 |
} |
| 1209 |
|
| 1210 |
#define CCDN_CASETESTEXPR_OK 0x0001 /* CaseTestExpr okay here? */ |
| 1211 |
|
| 1212 |
static bool |
| 1213 |
contain_context_dependent_node_walker(Node *node, int *flags) |
| 1214 |
{ |
| 1215 |
if (node == NULL) |
| 1216 |
return false; |
| 1217 |
if (IsA(node, CaseTestExpr)) |
| 1218 |
return !(*flags & CCDN_CASETESTEXPR_OK); |
| 1219 |
else if (IsA(node, CaseExpr)) |
| 1220 |
{ |
| 1221 |
CaseExpr *caseexpr = (CaseExpr *) node; |
| 1222 |
|
| 1223 |
/* |
| 1224 |
* If this CASE doesn't have a test expression, then it doesn't create |
| 1225 |
* a context in which CaseTestExprs should appear, so just fall |
| 1226 |
* through and treat it as a generic expression node. |
| 1227 |
*/ |
| 1228 |
if (caseexpr->arg) |
| 1229 |
{ |
| 1230 |
int save_flags = *flags; |
| 1231 |
bool res; |
| 1232 |
|
| 1233 |
/* |
| 1234 |
* Note: in principle, we could distinguish the various sub-parts |
| 1235 |
* of a CASE construct and set the flag bit only for some of them, |
| 1236 |
* since we are only expecting CaseTestExprs to appear in the |
| 1237 |
* "expr" subtree of the CaseWhen nodes. But it doesn't really |
| 1238 |
* seem worth any extra code. If there are any bare CaseTestExprs |
| 1239 |
* elsewhere in the CASE, something's wrong already. |
| 1240 |
*/ |
| 1241 |
*flags |= CCDN_CASETESTEXPR_OK; |
| 1242 |
res = expression_tree_walker(node, |
| 1243 |
contain_context_dependent_node_walker, |
| 1244 |
flags); |
| 1245 |
*flags = save_flags; |
| 1246 |
return res; |
| 1247 |
} |
| 1248 |
} |
| 1249 |
else if (IsA(node, ArrayCoerceExpr)) |
| 1250 |
{ |
| 1251 |
ArrayCoerceExpr *ac = (ArrayCoerceExpr *) node; |
| 1252 |
int save_flags; |
| 1253 |
bool res; |
| 1254 |
|
| 1255 |
/* Check the array expression */ |
| 1256 |
if (contain_context_dependent_node_walker((Node *) ac->arg, flags)) |
| 1257 |
return true; |
| 1258 |
|
| 1259 |
/* Check the elemexpr, which is allowed to contain CaseTestExpr */ |
| 1260 |
save_flags = *flags; |
| 1261 |
*flags |= CCDN_CASETESTEXPR_OK; |
| 1262 |
res = contain_context_dependent_node_walker((Node *) ac->elemexpr, |
| 1263 |
flags); |
| 1264 |
*flags = save_flags; |
| 1265 |
return res; |
| 1266 |
} |
| 1267 |
return expression_tree_walker(node, contain_context_dependent_node_walker, |
| 1268 |
flags); |
| 1269 |
} |
| 1270 |
|
| 1271 |
/***************************************************************************** |
| 1272 |
* Check clauses for Vars passed to non-leakproof functions |
| 1273 |
*****************************************************************************/ |
| 1274 |
|
| 1275 |
/* |
| 1276 |
* contain_leaked_vars |
| 1277 |
* Recursively scan a clause to discover whether it contains any Var |
| 1278 |
* nodes (of the current query level) that are passed as arguments to |
| 1279 |
* leaky functions. |
| 1280 |
* |
| 1281 |
* Returns true if the clause contains any non-leakproof functions that are |
| 1282 |
* passed Var nodes of the current query level, and which might therefore leak |
| 1283 |
* data. Such clauses must be applied after any lower-level security barrier |
| 1284 |
* clauses. |
| 1285 |
*/ |
| 1286 |
bool |
| 1287 |
contain_leaked_vars(Node *clause) |
| 1288 |
{ |
| 1289 |
return contain_leaked_vars_walker(clause, NULL); |
| 1290 |
} |
| 1291 |
|
| 1292 |
static bool |
| 1293 |
contain_leaked_vars_checker(Oid func_id, void *context) |
| 1294 |
{ |
| 1295 |
return !get_func_leakproof(func_id); |
| 1296 |
} |
| 1297 |
|
| 1298 |
static bool |
| 1299 |
contain_leaked_vars_walker(Node *node, void *context) |
| 1300 |
{ |
| 1301 |
if (node == NULL) |
| 1302 |
return false; |
| 1303 |
|
| 1304 |
switch (nodeTag(node)) |
| 1305 |
{ |
| 1306 |
case T_Var: |
| 1307 |
case T_Const: |
| 1308 |
case T_Param: |
| 1309 |
case T_ArrayExpr: |
| 1310 |
case T_FieldSelect: |
| 1311 |
case T_FieldStore: |
| 1312 |
case T_NamedArgExpr: |
| 1313 |
case T_BoolExpr: |
| 1314 |
case T_RelabelType: |
| 1315 |
case T_CollateExpr: |
| 1316 |
case T_CaseExpr: |
| 1317 |
case T_CaseTestExpr: |
| 1318 |
case T_RowExpr: |
| 1319 |
case T_SQLValueFunction: |
| 1320 |
case T_NullTest: |
| 1321 |
case T_BooleanTest: |
| 1322 |
case T_NextValueExpr: |
| 1323 |
case T_ReturningExpr: |
| 1324 |
case T_List: |
| 1325 |
|
| 1326 |
/* |
| 1327 |
* We know these node types don't contain function calls; but |
| 1328 |
* something further down in the node tree might. |
| 1329 |
*/ |
| 1330 |
break; |
| 1331 |
|
| 1332 |
case T_FuncExpr: |
| 1333 |
case T_OpExpr: |
| 1334 |
case T_DistinctExpr: |
| 1335 |
case T_NullIfExpr: |
| 1336 |
case T_ScalarArrayOpExpr: |
| 1337 |
case T_CoerceViaIO: |
| 1338 |
case T_ArrayCoerceExpr: |
| 1339 |
|
| 1340 |
/* |
| 1341 |
* If node contains a leaky function call, and there's any Var |
| 1342 |
* underneath it, reject. |
| 1343 |
*/ |
| 1344 |
if (check_functions_in_node(node, contain_leaked_vars_checker, |
| 1345 |
context) && |
| 1346 |
contain_var_clause(node)) |
| 1347 |
return true; |
| 1348 |
break; |
| 1349 |
|
| 1350 |
case T_SubscriptingRef: |
| 1351 |
{ |
| 1352 |
SubscriptingRef *sbsref = (SubscriptingRef *) node; |
| 1353 |
const SubscriptRoutines *sbsroutines; |
| 1354 |
|
| 1355 |
/* Consult the subscripting support method info */ |
| 1356 |
sbsroutines = getSubscriptingRoutines(sbsref->refcontainertype, |
| 1357 |
NULL); |
| 1358 |
if (!sbsroutines || |
| 1359 |
!(sbsref->refassgnexpr != NULL ? |
| 1360 |
sbsroutines->store_leakproof : |
| 1361 |
sbsroutines->fetch_leakproof)) |
| 1362 |
{ |
| 1363 |
/* Node is leaky, so reject if it contains Vars */ |
| 1364 |
if (contain_var_clause(node)) |
| 1365 |
return true; |
| 1366 |
} |
| 1367 |
} |
| 1368 |
break; |
| 1369 |
|
| 1370 |
case T_RowCompareExpr: |
| 1371 |
{ |
| 1372 |
/* |
| 1373 |
* It's worth special-casing this because a leaky comparison |
| 1374 |
* function only compromises one pair of row elements, which |
| 1375 |
* might not contain Vars while others do. |
| 1376 |
*/ |
| 1377 |
RowCompareExpr *rcexpr = (RowCompareExpr *) node; |
| 1378 |
ListCell *opid; |
| 1379 |
ListCell *larg; |
| 1380 |
ListCell *rarg; |
| 1381 |
|
| 1382 |
forthree(opid, rcexpr->opnos, |
| 1383 |
larg, rcexpr->largs, |
| 1384 |
rarg, rcexpr->rargs) |
| 1385 |
{ |
| 1386 |
Oid funcid = get_opcode(lfirst_oid(opid)); |
| 1387 |
|
| 1388 |
if (!get_func_leakproof(funcid) && |
| 1389 |
(contain_var_clause((Node *) lfirst(larg)) || |
| 1390 |
contain_var_clause((Node *) lfirst(rarg)))) |
| 1391 |
return true; |
| 1392 |
} |
| 1393 |
} |
| 1394 |
break; |
| 1395 |
|
| 1396 |
case T_MinMaxExpr: |
| 1397 |
{ |
| 1398 |
/* |
| 1399 |
* MinMaxExpr is leakproof if the comparison function it calls |
| 1400 |
* is leakproof. |
| 1401 |
*/ |
| 1402 |
MinMaxExpr *minmaxexpr = (MinMaxExpr *) node; |
| 1403 |
TypeCacheEntry *typentry; |
| 1404 |
bool leakproof; |
| 1405 |
|
| 1406 |
/* Look up the btree comparison function for the datatype */ |
| 1407 |
typentry = lookup_type_cache(minmaxexpr->minmaxtype, |
| 1408 |
TYPECACHE_CMP_PROC); |
| 1409 |
if (OidIsValid(typentry->cmp_proc)) |
| 1410 |
leakproof = get_func_leakproof(typentry->cmp_proc); |
| 1411 |
else |
| 1412 |
{ |
| 1413 |
/* |
| 1414 |
* The executor will throw an error, but here we just |
| 1415 |
* treat the missing function as leaky. |
| 1416 |
*/ |
| 1417 |
leakproof = false; |
| 1418 |
} |
| 1419 |
|
| 1420 |
if (!leakproof && |
| 1421 |
contain_var_clause((Node *) minmaxexpr->args)) |
| 1422 |
return true; |
| 1423 |
} |
| 1424 |
break; |
| 1425 |
|
| 1426 |
case T_CurrentOfExpr: |
| 1427 |
|
| 1428 |
/* |
| 1429 |
* WHERE CURRENT OF doesn't contain leaky function calls. |
| 1430 |
* Moreover, it is essential that this is considered non-leaky, |
| 1431 |
* since the planner must always generate a TID scan when CURRENT |
| 1432 |
* OF is present -- cf. cost_tidscan. |
| 1433 |
*/ |
| 1434 |
return false; |
| 1435 |
|
| 1436 |
default: |
| 1437 |
|
| 1438 |
/* |
| 1439 |
* If we don't recognize the node tag, assume it might be leaky. |
| 1440 |
* This prevents an unexpected security hole if someone adds a new |
| 1441 |
* node type that can call a function. |
| 1442 |
*/ |
| 1443 |
return true; |
| 1444 |
} |
| 1445 |
return expression_tree_walker(node, contain_leaked_vars_walker, |
| 1446 |
context); |
| 1447 |
} |
| 1448 |
|
| 1449 |
/* |
| 1450 |
* find_nonnullable_rels |
| 1451 |
* Determine which base rels are forced nonnullable by given clause. |
| 1452 |
* |
| 1453 |
* Returns the set of all Relids that are referenced in the clause in such |
| 1454 |
* a way that the clause cannot possibly return TRUE if any of these Relids |
| 1455 |
* is an all-NULL row. (It is OK to err on the side of conservatism; hence |
| 1456 |
* the analysis here is simplistic.) |
| 1457 |
* |
| 1458 |
* The semantics here are subtly different from contain_nonstrict_functions: |
| 1459 |
* that function is concerned with NULL results from arbitrary expressions, |
| 1460 |
* but here we assume that the input is a Boolean expression, and wish to |
| 1461 |
* see if NULL inputs will provably cause a FALSE-or-NULL result. We expect |
| 1462 |
* the expression to have been AND/OR flattened and converted to implicit-AND |
| 1463 |
* format. |
| 1464 |
* |
| 1465 |
* Note: this function is largely duplicative of find_nonnullable_vars(). |
| 1466 |
* The reason not to simplify this function into a thin wrapper around |
| 1467 |
* find_nonnullable_vars() is that the tested conditions really are different: |
| 1468 |
* a clause like "t1.v1 IS NOT NULL OR t1.v2 IS NOT NULL" does not prove |
| 1469 |
* that either v1 or v2 can't be NULL, but it does prove that the t1 row |
| 1470 |
* as a whole can't be all-NULL. Also, the behavior for PHVs is different. |
| 1471 |
* |
| 1472 |
* top_level is true while scanning top-level AND/OR structure; here, showing |
| 1473 |
* the result is either FALSE or NULL is good enough. top_level is false when |
| 1474 |
* we have descended below a NOT or a strict function: now we must be able to |
| 1475 |
* prove that the subexpression goes to NULL. |
| 1476 |
* |
| 1477 |
* We don't use expression_tree_walker here because we don't want to descend |
| 1478 |
* through very many kinds of nodes; only the ones we can be sure are strict. |
| 1479 |
*/ |
| 1480 |
Relids |
| 1481 |
find_nonnullable_rels(Node *clause) |
| 1482 |
{ |
| 1483 |
return find_nonnullable_rels_walker(clause, true); |
| 1484 |
} |
| 1485 |
|
| 1486 |
static Relids |
| 1487 |
find_nonnullable_rels_walker(Node *node, bool top_level) |
| 1488 |
{ |
| 1489 |
Relids result = NULL; |
| 1490 |
ListCell *l; |
| 1491 |
|
| 1492 |
if (node == NULL) |
| 1493 |
return NULL; |
| 1494 |
if (IsA(node, Var)) |
| 1495 |
{ |
| 1496 |
Var *var = (Var *) node; |
| 1497 |
|
| 1498 |
if (var->varlevelsup == 0) |
| 1499 |
result = bms_make_singleton(var->varno); |
| 1500 |
} |
| 1501 |
else if (IsA(node, List)) |
| 1502 |
{ |
| 1503 |
/* |
| 1504 |
* At top level, we are examining an implicit-AND list: if any of the |
| 1505 |
* arms produces FALSE-or-NULL then the result is FALSE-or-NULL. If |
| 1506 |
* not at top level, we are examining the arguments of a strict |
| 1507 |
* function: if any of them produce NULL then the result of the |
| 1508 |
* function must be NULL. So in both cases, the set of nonnullable |
| 1509 |
* rels is the union of those found in the arms, and we pass down the |
| 1510 |
* top_level flag unmodified. |
| 1511 |
*/ |
| 1512 |
foreach(l, (List *) node) |
| 1513 |
{ |
| 1514 |
result = bms_join(result, |
| 1515 |
find_nonnullable_rels_walker(lfirst(l), |
| 1516 |
top_level)); |
| 1517 |
} |
| 1518 |
} |
| 1519 |
else if (IsA(node, FuncExpr)) |
| 1520 |
{ |
| 1521 |
FuncExpr *expr = (FuncExpr *) node; |
| 1522 |
|
| 1523 |
if (func_strict(expr->funcid)) |
| 1524 |
result = find_nonnullable_rels_walker((Node *) expr->args, false); |
| 1525 |
} |
| 1526 |
else if (IsA(node, OpExpr)) |
| 1527 |
{ |
| 1528 |
OpExpr *expr = (OpExpr *) node; |
| 1529 |
|
| 1530 |
set_opfuncid(expr); |
| 1531 |
if (func_strict(expr->opfuncid)) |
| 1532 |
result = find_nonnullable_rels_walker((Node *) expr->args, false); |
| 1533 |
} |
| 1534 |
else if (IsA(node, ScalarArrayOpExpr)) |
| 1535 |
{ |
| 1536 |
ScalarArrayOpExpr *expr = (ScalarArrayOpExpr *) node; |
| 1537 |
|
| 1538 |
if (is_strict_saop(expr, top_level)) |
| 1539 |
result = find_nonnullable_rels_walker((Node *) expr->args, false); |
| 1540 |
} |
| 1541 |
else if (IsA(node, BoolExpr)) |
| 1542 |
{ |
| 1543 |
BoolExpr *expr = (BoolExpr *) node; |
| 1544 |
|
| 1545 |
switch (expr->boolop) |
| 1546 |
{ |
| 1547 |
case AND_EXPR: |
| 1548 |
/* At top level we can just recurse (to the List case) */ |
| 1549 |
if (top_level) |
| 1550 |
{ |
| 1551 |
result = find_nonnullable_rels_walker((Node *) expr->args, |
| 1552 |
top_level); |
| 1553 |
break; |
| 1554 |
} |
| 1555 |
|
| 1556 |
/* |
| 1557 |
* Below top level, even if one arm produces NULL, the result |
| 1558 |
* could be FALSE (hence not NULL). However, if *all* the |
| 1559 |
* arms produce NULL then the result is NULL, so we can take |
| 1560 |
* the intersection of the sets of nonnullable rels, just as |
| 1561 |
* for OR. Fall through to share code. |
| 1562 |
*/ |
| 1563 |
/* FALL THRU */ |
| 1564 |
case OR_EXPR: |
| 1565 |
|
| 1566 |
/* |
| 1567 |
* OR is strict if all of its arms are, so we can take the |
| 1568 |
* intersection of the sets of nonnullable rels for each arm. |
| 1569 |
* This works for both values of top_level. |
| 1570 |
*/ |
| 1571 |
foreach(l, expr->args) |
| 1572 |
{ |
| 1573 |
Relids subresult; |
| 1574 |
|
| 1575 |
subresult = find_nonnullable_rels_walker(lfirst(l), |
| 1576 |
top_level); |
| 1577 |
if (result == NULL) /* first subresult? */ |
| 1578 |
result = subresult; |
| 1579 |
else |
| 1580 |
result = bms_int_members(result, subresult); |
| 1581 |
|
| 1582 |
/* |
| 1583 |
* If the intersection is empty, we can stop looking. This |
| 1584 |
* also justifies the test for first-subresult above. |
| 1585 |
*/ |
| 1586 |
if (bms_is_empty(result)) |
| 1587 |
break; |
| 1588 |
} |
| 1589 |
break; |
| 1590 |
case NOT_EXPR: |
| 1591 |
/* NOT will return null if its arg is null */ |
| 1592 |
result = find_nonnullable_rels_walker((Node *) expr->args, |
| 1593 |
false); |
| 1594 |
break; |
| 1595 |
default: |
| 1596 |
elog(ERROR, "unrecognized boolop: %d", (int) expr->boolop); |
| 1597 |
break; |
| 1598 |
} |
| 1599 |
} |
| 1600 |
else if (IsA(node, RelabelType)) |
| 1601 |
{ |
| 1602 |
RelabelType *expr = (RelabelType *) node; |
| 1603 |
|
| 1604 |
result = find_nonnullable_rels_walker((Node *) expr->arg, top_level); |
| 1605 |
} |
| 1606 |
else if (IsA(node, CoerceViaIO)) |
| 1607 |
{ |
| 1608 |
/* not clear this is useful, but it can't hurt */ |
| 1609 |
CoerceViaIO *expr = (CoerceViaIO *) node; |
| 1610 |
|
| 1611 |
result = find_nonnullable_rels_walker((Node *) expr->arg, top_level); |
| 1612 |
} |
| 1613 |
else if (IsA(node, ArrayCoerceExpr)) |
| 1614 |
{ |
| 1615 |
/* ArrayCoerceExpr is strict at the array level; ignore elemexpr */ |
| 1616 |
ArrayCoerceExpr *expr = (ArrayCoerceExpr *) node; |
| 1617 |
|
| 1618 |
result = find_nonnullable_rels_walker((Node *) expr->arg, top_level); |
| 1619 |
} |
| 1620 |
else if (IsA(node, ConvertRowtypeExpr)) |
| 1621 |
{ |
| 1622 |
/* not clear this is useful, but it can't hurt */ |
| 1623 |
ConvertRowtypeExpr *expr = (ConvertRowtypeExpr *) node; |
| 1624 |
|
| 1625 |
result = find_nonnullable_rels_walker((Node *) expr->arg, top_level); |
| 1626 |
} |
| 1627 |
else if (IsA(node, CollateExpr)) |
| 1628 |
{ |
| 1629 |
CollateExpr *expr = (CollateExpr *) node; |
| 1630 |
|
| 1631 |
result = find_nonnullable_rels_walker((Node *) expr->arg, top_level); |
| 1632 |
} |
| 1633 |
else if (IsA(node, NullTest)) |
| 1634 |
{ |
| 1635 |
/* IS NOT NULL can be considered strict, but only at top level */ |
| 1636 |
NullTest *expr = (NullTest *) node; |
| 1637 |
|
| 1638 |
if (top_level && expr->nulltesttype == IS_NOT_NULL && !expr->argisrow) |
| 1639 |
result = find_nonnullable_rels_walker((Node *) expr->arg, false); |
| 1640 |
} |
| 1641 |
else if (IsA(node, BooleanTest)) |
| 1642 |
{ |
| 1643 |
/* Boolean tests that reject NULL are strict at top level */ |
| 1644 |
BooleanTest *expr = (BooleanTest *) node; |
| 1645 |
|
| 1646 |
if (top_level && |
| 1647 |
(expr->booltesttype == IS_TRUE || |
| 1648 |
expr->booltesttype == IS_FALSE || |
| 1649 |
expr->booltesttype == IS_NOT_UNKNOWN)) |
| 1650 |
result = find_nonnullable_rels_walker((Node *) expr->arg, false); |
| 1651 |
} |
| 1652 |
else if (IsA(node, SubPlan)) |
| 1653 |
{ |
| 1654 |
SubPlan *splan = (SubPlan *) node; |
| 1655 |
|
| 1656 |
/* |
| 1657 |
* For some types of SubPlan, we can infer strictness from Vars in the |
| 1658 |
* testexpr (the LHS of the original SubLink). |
| 1659 |
* |
| 1660 |
* For ANY_SUBLINK, if the subquery produces zero rows, the result is |
| 1661 |
* always FALSE. If the subquery produces more than one row, the |
| 1662 |
* per-row results of the testexpr are combined using OR semantics. |
| 1663 |
* Hence ANY_SUBLINK can be strict only at top level, but there it's |
| 1664 |
* as strict as the testexpr is. |
| 1665 |
* |
| 1666 |
* For ROWCOMPARE_SUBLINK, if the subquery produces zero rows, the |
| 1667 |
* result is always NULL. Otherwise, the result is as strict as the |
| 1668 |
* testexpr is. So we can check regardless of top_level. |
| 1669 |
* |
| 1670 |
* We can't prove anything for other sublink types (in particular, |
| 1671 |
* note that ALL_SUBLINK will return TRUE if the subquery is empty). |
| 1672 |
*/ |
| 1673 |
if ((top_level && splan->subLinkType == ANY_SUBLINK) || |
| 1674 |
splan->subLinkType == ROWCOMPARE_SUBLINK) |
| 1675 |
result = find_nonnullable_rels_walker(splan->testexpr, top_level); |
| 1676 |
} |
| 1677 |
else if (IsA(node, PlaceHolderVar)) |
| 1678 |
{ |
| 1679 |
PlaceHolderVar *phv = (PlaceHolderVar *) node; |
| 1680 |
|
| 1681 |
/* |
| 1682 |
* If the contained expression forces any rels non-nullable, so does |
| 1683 |
* the PHV. |
| 1684 |
*/ |
| 1685 |
result = find_nonnullable_rels_walker((Node *) phv->phexpr, top_level); |
| 1686 |
|
| 1687 |
/* |
| 1688 |
* If the PHV's syntactic scope is exactly one rel, it will be forced |
| 1689 |
* to be evaluated at that rel, and so it will behave like a Var of |
| 1690 |
* that rel: if the rel's entire output goes to null, so will the PHV. |
| 1691 |
* (If the syntactic scope is a join, we know that the PHV will go to |
| 1692 |
* null if the whole join does; but that is AND semantics while we |
| 1693 |
* need OR semantics for find_nonnullable_rels' result, so we can't do |
| 1694 |
* anything with the knowledge.) |
| 1695 |
*/ |
| 1696 |
if (phv->phlevelsup == 0 && |
| 1697 |
bms_membership(phv->phrels) == BMS_SINGLETON) |
| 1698 |
result = bms_add_members(result, phv->phrels); |
| 1699 |
} |
| 1700 |
return result; |
| 1701 |
} |
| 1702 |
|
| 1703 |
/* |
| 1704 |
* find_nonnullable_vars |
| 1705 |
* Determine which Vars are forced nonnullable by given clause. |
| 1706 |
* |
| 1707 |
* Returns the set of all level-zero Vars that are referenced in the clause in |
| 1708 |
* such a way that the clause cannot possibly return TRUE if any of these Vars |
| 1709 |
* is NULL. (It is OK to err on the side of conservatism; hence the analysis |
| 1710 |
* here is simplistic.) |
| 1711 |
* |
| 1712 |
* The semantics here are subtly different from contain_nonstrict_functions: |
| 1713 |
* that function is concerned with NULL results from arbitrary expressions, |
| 1714 |
* but here we assume that the input is a Boolean expression, and wish to |
| 1715 |
* see if NULL inputs will provably cause a FALSE-or-NULL result. We expect |
| 1716 |
* the expression to have been AND/OR flattened and converted to implicit-AND |
| 1717 |
* format. |
| 1718 |
* |
| 1719 |
* Attnos of the identified Vars are returned in a multibitmapset (a List of |
| 1720 |
* Bitmapsets). List indexes correspond to relids (varnos), while the per-rel |
| 1721 |
* Bitmapsets hold varattnos offset by FirstLowInvalidHeapAttributeNumber. |
| 1722 |
* |
| 1723 |
* top_level is true while scanning top-level AND/OR structure; here, showing |
| 1724 |
* the result is either FALSE or NULL is good enough. top_level is false when |
| 1725 |
* we have descended below a NOT or a strict function: now we must be able to |
| 1726 |
* prove that the subexpression goes to NULL. |
| 1727 |
* |
| 1728 |
* We don't use expression_tree_walker here because we don't want to descend |
| 1729 |
* through very many kinds of nodes; only the ones we can be sure are strict. |
| 1730 |
*/ |
| 1731 |
List * |
| 1732 |
find_nonnullable_vars(Node *clause) |
| 1733 |
{ |
| 1734 |
return find_nonnullable_vars_walker(clause, true); |
| 1735 |
} |
| 1736 |
|
| 1737 |
static List * |
| 1738 |
find_nonnullable_vars_walker(Node *node, bool top_level) |
| 1739 |
{ |
| 1740 |
List *result = NIL; |
| 1741 |
ListCell *l; |
| 1742 |
|
| 1743 |
if (node == NULL) |
| 1744 |
return NIL; |
| 1745 |
if (IsA(node, Var)) |
| 1746 |
{ |
| 1747 |
Var *var = (Var *) node; |
| 1748 |
|
| 1749 |
if (var->varlevelsup == 0) |
| 1750 |
result = mbms_add_member(result, |
| 1751 |
var->varno, |
| 1752 |
var->varattno - FirstLowInvalidHeapAttributeNumber); |
| 1753 |
} |
| 1754 |
else if (IsA(node, List)) |
| 1755 |
{ |
| 1756 |
/* |
| 1757 |
* At top level, we are examining an implicit-AND list: if any of the |
| 1758 |
* arms produces FALSE-or-NULL then the result is FALSE-or-NULL. If |
| 1759 |
* not at top level, we are examining the arguments of a strict |
| 1760 |
* function: if any of them produce NULL then the result of the |
| 1761 |
* function must be NULL. So in both cases, the set of nonnullable |
| 1762 |
* vars is the union of those found in the arms, and we pass down the |
| 1763 |
* top_level flag unmodified. |
| 1764 |
*/ |
| 1765 |
foreach(l, (List *) node) |
| 1766 |
{ |
| 1767 |
result = mbms_add_members(result, |
| 1768 |
find_nonnullable_vars_walker(lfirst(l), |
| 1769 |
top_level)); |
| 1770 |
} |
| 1771 |
} |
| 1772 |
else if (IsA(node, FuncExpr)) |
| 1773 |
{ |
| 1774 |
FuncExpr *expr = (FuncExpr *) node; |
| 1775 |
|
| 1776 |
if (func_strict(expr->funcid)) |
| 1777 |
result = find_nonnullable_vars_walker((Node *) expr->args, false); |
| 1778 |
} |
| 1779 |
else if (IsA(node, OpExpr)) |
| 1780 |
{ |
| 1781 |
OpExpr *expr = (OpExpr *) node; |
| 1782 |
|
| 1783 |
set_opfuncid(expr); |
| 1784 |
if (func_strict(expr->opfuncid)) |
| 1785 |
result = find_nonnullable_vars_walker((Node *) expr->args, false); |
| 1786 |
} |
| 1787 |
else if (IsA(node, ScalarArrayOpExpr)) |
| 1788 |
{ |
| 1789 |
ScalarArrayOpExpr *expr = (ScalarArrayOpExpr *) node; |
| 1790 |
|
| 1791 |
if (is_strict_saop(expr, top_level)) |
| 1792 |
result = find_nonnullable_vars_walker((Node *) expr->args, false); |
| 1793 |
} |
| 1794 |
else if (IsA(node, BoolExpr)) |
| 1795 |
{ |
| 1796 |
BoolExpr *expr = (BoolExpr *) node; |
| 1797 |
|
| 1798 |
switch (expr->boolop) |
| 1799 |
{ |
| 1800 |
case AND_EXPR: |
| 1801 |
|
| 1802 |
/* |
| 1803 |
* At top level we can just recurse (to the List case), since |
| 1804 |
* the result should be the union of what we can prove in each |
| 1805 |
* arm. |
| 1806 |
*/ |
| 1807 |
if (top_level) |
| 1808 |
{ |
| 1809 |
result = find_nonnullable_vars_walker((Node *) expr->args, |
| 1810 |
top_level); |
| 1811 |
break; |
| 1812 |
} |
| 1813 |
|
| 1814 |
/* |
| 1815 |
* Below top level, even if one arm produces NULL, the result |
| 1816 |
* could be FALSE (hence not NULL). However, if *all* the |
| 1817 |
* arms produce NULL then the result is NULL, so we can take |
| 1818 |
* the intersection of the sets of nonnullable vars, just as |
| 1819 |
* for OR. Fall through to share code. |
| 1820 |
*/ |
| 1821 |
/* FALL THRU */ |
| 1822 |
case OR_EXPR: |
| 1823 |
|
| 1824 |
/* |
| 1825 |
* OR is strict if all of its arms are, so we can take the |
| 1826 |
* intersection of the sets of nonnullable vars for each arm. |
| 1827 |
* This works for both values of top_level. |
| 1828 |
*/ |
| 1829 |
foreach(l, expr->args) |
| 1830 |
{ |
| 1831 |
List *subresult; |
| 1832 |
|
| 1833 |
subresult = find_nonnullable_vars_walker(lfirst(l), |
| 1834 |
top_level); |
| 1835 |
if (result == NIL) /* first subresult? */ |
| 1836 |
result = subresult; |
| 1837 |
else |
| 1838 |
result = mbms_int_members(result, subresult); |
| 1839 |
|
| 1840 |
/* |
| 1841 |
* If the intersection is empty, we can stop looking. This |
| 1842 |
* also justifies the test for first-subresult above. |
| 1843 |
*/ |
| 1844 |
if (result == NIL) |
| 1845 |
break; |
| 1846 |
} |
| 1847 |
break; |
| 1848 |
case NOT_EXPR: |
| 1849 |
/* NOT will return null if its arg is null */ |
| 1850 |
result = find_nonnullable_vars_walker((Node *) expr->args, |
| 1851 |
false); |
| 1852 |
break; |
| 1853 |
default: |
| 1854 |
elog(ERROR, "unrecognized boolop: %d", (int) expr->boolop); |
| 1855 |
break; |
| 1856 |
} |
| 1857 |
} |
| 1858 |
else if (IsA(node, RelabelType)) |
| 1859 |
{ |
| 1860 |
RelabelType *expr = (RelabelType *) node; |
| 1861 |
|
| 1862 |
result = find_nonnullable_vars_walker((Node *) expr->arg, top_level); |
| 1863 |
} |
| 1864 |
else if (IsA(node, CoerceViaIO)) |
| 1865 |
{ |
| 1866 |
/* not clear this is useful, but it can't hurt */ |
| 1867 |
CoerceViaIO *expr = (CoerceViaIO *) node; |
| 1868 |
|
| 1869 |
result = find_nonnullable_vars_walker((Node *) expr->arg, false); |
| 1870 |
} |
| 1871 |
else if (IsA(node, ArrayCoerceExpr)) |
| 1872 |
{ |
| 1873 |
/* ArrayCoerceExpr is strict at the array level; ignore elemexpr */ |
| 1874 |
ArrayCoerceExpr *expr = (ArrayCoerceExpr *) node; |
| 1875 |
|
| 1876 |
result = find_nonnullable_vars_walker((Node *) expr->arg, top_level); |
| 1877 |
} |
| 1878 |
else if (IsA(node, ConvertRowtypeExpr)) |
| 1879 |
{ |
| 1880 |
/* not clear this is useful, but it can't hurt */ |
| 1881 |
ConvertRowtypeExpr *expr = (ConvertRowtypeExpr *) node; |
| 1882 |
|
| 1883 |
result = find_nonnullable_vars_walker((Node *) expr->arg, top_level); |
| 1884 |
} |
| 1885 |
else if (IsA(node, CollateExpr)) |
| 1886 |
{ |
| 1887 |
CollateExpr *expr = (CollateExpr *) node; |
| 1888 |
|
| 1889 |
result = find_nonnullable_vars_walker((Node *) expr->arg, top_level); |
| 1890 |
} |
| 1891 |
else if (IsA(node, NullTest)) |
| 1892 |
{ |
| 1893 |
/* IS NOT NULL can be considered strict, but only at top level */ |
| 1894 |
NullTest *expr = (NullTest *) node; |
| 1895 |
|
| 1896 |
if (top_level && expr->nulltesttype == IS_NOT_NULL && !expr->argisrow) |
| 1897 |
result = find_nonnullable_vars_walker((Node *) expr->arg, false); |
| 1898 |
} |
| 1899 |
else if (IsA(node, BooleanTest)) |
| 1900 |
{ |
| 1901 |
/* Boolean tests that reject NULL are strict at top level */ |
| 1902 |
BooleanTest *expr = (BooleanTest *) node; |
| 1903 |
|
| 1904 |
if (top_level && |
| 1905 |
(expr->booltesttype == IS_TRUE || |
| 1906 |
expr->booltesttype == IS_FALSE || |
| 1907 |
expr->booltesttype == IS_NOT_UNKNOWN)) |
| 1908 |
result = find_nonnullable_vars_walker((Node *) expr->arg, false); |
| 1909 |
} |
| 1910 |
else if (IsA(node, SubPlan)) |
| 1911 |
{ |
| 1912 |
SubPlan *splan = (SubPlan *) node; |
| 1913 |
|
| 1914 |
/* See analysis in find_nonnullable_rels_walker */ |
| 1915 |
if ((top_level && splan->subLinkType == ANY_SUBLINK) || |
| 1916 |
splan->subLinkType == ROWCOMPARE_SUBLINK) |
| 1917 |
result = find_nonnullable_vars_walker(splan->testexpr, top_level); |
| 1918 |
} |
| 1919 |
else if (IsA(node, PlaceHolderVar)) |
| 1920 |
{ |
| 1921 |
PlaceHolderVar *phv = (PlaceHolderVar *) node; |
| 1922 |
|
| 1923 |
result = find_nonnullable_vars_walker((Node *) phv->phexpr, top_level); |
| 1924 |
} |
| 1925 |
return result; |
| 1926 |
} |
| 1927 |
|
| 1928 |
/* |
| 1929 |
* find_forced_null_vars |
| 1930 |
* Determine which Vars must be NULL for the given clause to return TRUE. |
| 1931 |
* |
| 1932 |
* This is the complement of find_nonnullable_vars: find the level-zero Vars |
| 1933 |
* that must be NULL for the clause to return TRUE. (It is OK to err on the |
| 1934 |
* side of conservatism; hence the analysis here is simplistic. In fact, |
| 1935 |
* we only detect simple "var IS NULL" tests at the top level.) |
| 1936 |
* |
| 1937 |
* As with find_nonnullable_vars, we return the varattnos of the identified |
| 1938 |
* Vars in a multibitmapset. |
| 1939 |
*/ |
| 1940 |
List * |
| 1941 |
find_forced_null_vars(Node *node) |
| 1942 |
{ |
| 1943 |
List *result = NIL; |
| 1944 |
Var *var; |
| 1945 |
ListCell *l; |
| 1946 |
|
| 1947 |
if (node == NULL) |
| 1948 |
return NIL; |
| 1949 |
/* Check single-clause cases using subroutine */ |
| 1950 |
var = find_forced_null_var(node); |
| 1951 |
if (var) |
| 1952 |
{ |
| 1953 |
result = mbms_add_member(result, |
| 1954 |
var->varno, |
| 1955 |
var->varattno - FirstLowInvalidHeapAttributeNumber); |
| 1956 |
} |
| 1957 |
/* Otherwise, handle AND-conditions */ |
| 1958 |
else if (IsA(node, List)) |
| 1959 |
{ |
| 1960 |
/* |
| 1961 |
* At top level, we are examining an implicit-AND list: if any of the |
| 1962 |
* arms produces FALSE-or-NULL then the result is FALSE-or-NULL. |
| 1963 |
*/ |
| 1964 |
foreach(l, (List *) node) |
| 1965 |
{ |
| 1966 |
result = mbms_add_members(result, |
| 1967 |
find_forced_null_vars((Node *) lfirst(l))); |
| 1968 |
} |
| 1969 |
} |
| 1970 |
else if (IsA(node, BoolExpr)) |
| 1971 |
{ |
| 1972 |
BoolExpr *expr = (BoolExpr *) node; |
| 1973 |
|
| 1974 |
/* |
| 1975 |
* We don't bother considering the OR case, because it's fairly |
| 1976 |
* unlikely anyone would write "v1 IS NULL OR v1 IS NULL". Likewise, |
| 1977 |
* the NOT case isn't worth expending code on. |
| 1978 |
*/ |
| 1979 |
if (expr->boolop == AND_EXPR) |
| 1980 |
{ |
| 1981 |
/* At top level we can just recurse (to the List case) */ |
| 1982 |
result = find_forced_null_vars((Node *) expr->args); |
| 1983 |
} |
| 1984 |
} |
| 1985 |
return result; |
| 1986 |
} |
| 1987 |
|
| 1988 |
/* |
| 1989 |
* find_forced_null_var |
| 1990 |
* Return the Var forced null by the given clause, or NULL if it's |
| 1991 |
* not an IS NULL-type clause. For success, the clause must enforce |
| 1992 |
* *only* nullness of the particular Var, not any other conditions. |
| 1993 |
* |
| 1994 |
* This is just the single-clause case of find_forced_null_vars(), without |
| 1995 |
* any allowance for AND conditions. It's used by initsplan.c on individual |
| 1996 |
* qual clauses. The reason for not just applying find_forced_null_vars() |
| 1997 |
* is that if an AND of an IS NULL clause with something else were to somehow |
| 1998 |
* survive AND/OR flattening, initsplan.c might get fooled into discarding |
| 1999 |
* the whole clause when only the IS NULL part of it had been proved redundant. |
| 2000 |
*/ |
| 2001 |
Var * |
| 2002 |
find_forced_null_var(Node *node) |
| 2003 |
{ |
| 2004 |
if (node == NULL) |
| 2005 |
return NULL; |
| 2006 |
if (IsA(node, NullTest)) |
| 2007 |
{ |
| 2008 |
/* check for var IS NULL */ |
| 2009 |
NullTest *expr = (NullTest *) node; |
| 2010 |
|
| 2011 |
if (expr->nulltesttype == IS_NULL && !expr->argisrow) |
| 2012 |
{ |
| 2013 |
Var *var = (Var *) expr->arg; |
| 2014 |
|
| 2015 |
if (var && IsA(var, Var) && |
| 2016 |
var->varlevelsup == 0) |
| 2017 |
return var; |
| 2018 |
} |
| 2019 |
} |
| 2020 |
else if (IsA(node, BooleanTest)) |
| 2021 |
{ |
| 2022 |
/* var IS UNKNOWN is equivalent to var IS NULL */ |
| 2023 |
BooleanTest *expr = (BooleanTest *) node; |
| 2024 |
|
| 2025 |
if (expr->booltesttype == IS_UNKNOWN) |
| 2026 |
{ |
| 2027 |
Var *var = (Var *) expr->arg; |
| 2028 |
|
| 2029 |
if (var && IsA(var, Var) && |
| 2030 |
var->varlevelsup == 0) |
| 2031 |
return var; |
| 2032 |
} |
| 2033 |
} |
| 2034 |
return NULL; |
| 2035 |
} |
| 2036 |
|
| 2037 |
/* |
| 2038 |
* Can we treat a ScalarArrayOpExpr as strict? |
| 2039 |
* |
| 2040 |
* If "falseOK" is true, then a "false" result can be considered strict, |
| 2041 |
* else we need to guarantee an actual NULL result for NULL input. |
| 2042 |
* |
| 2043 |
* "foo op ALL array" is strict if the op is strict *and* we can prove |
| 2044 |
* that the array input isn't an empty array. We can check that |
| 2045 |
* for the cases of an array constant and an ARRAY[] construct. |
| 2046 |
* |
| 2047 |
* "foo op ANY array" is strict in the falseOK sense if the op is strict. |
| 2048 |
* If not falseOK, the test is the same as for "foo op ALL array". |
| 2049 |
*/ |
| 2050 |
static bool |
| 2051 |
is_strict_saop(ScalarArrayOpExpr *expr, bool falseOK) |
| 2052 |
{ |
| 2053 |
Node *rightop; |
| 2054 |
|
| 2055 |
/* The contained operator must be strict. */ |
| 2056 |
set_sa_opfuncid(expr); |
| 2057 |
if (!func_strict(expr->opfuncid)) |
| 2058 |
return false; |
| 2059 |
/* If ANY and falseOK, that's all we need to check. */ |
| 2060 |
if (expr->useOr && falseOK) |
| 2061 |
return true; |
| 2062 |
/* Else, we have to see if the array is provably non-empty. */ |
| 2063 |
Assert(list_length(expr->args) == 2); |
| 2064 |
rightop = (Node *) lsecond(expr->args); |
| 2065 |
if (rightop && IsA(rightop, Const)) |
| 2066 |
{ |
| 2067 |
Datum arraydatum = ((Const *) rightop)->constvalue; |
| 2068 |
bool arrayisnull = ((Const *) rightop)->constisnull; |
| 2069 |
ArrayType *arrayval; |
| 2070 |
int nitems; |
| 2071 |
|
| 2072 |
if (arrayisnull) |
| 2073 |
return false; |
| 2074 |
arrayval = DatumGetArrayTypeP(arraydatum); |
| 2075 |
nitems = ArrayGetNItems(ARR_NDIM(arrayval), ARR_DIMS(arrayval)); |
| 2076 |
if (nitems > 0) |
| 2077 |
return true; |
| 2078 |
} |
| 2079 |
else if (rightop && IsA(rightop, ArrayExpr)) |
| 2080 |
{ |
| 2081 |
ArrayExpr *arrayexpr = (ArrayExpr *) rightop; |
| 2082 |
|
| 2083 |
if (arrayexpr->elements != NIL && !arrayexpr->multidims) |
| 2084 |
return true; |
| 2085 |
} |
| 2086 |
return false; |
| 2087 |
} |
| 2088 |
|
| 2089 |
|
| 2090 |
/***************************************************************************** |
| 2091 |
* Check for "pseudo-constant" clauses |
| 2092 |
*****************************************************************************/ |
| 2093 |
|
| 2094 |
/* |
| 2095 |
* is_pseudo_constant_clause |
| 2096 |
* Detect whether an expression is "pseudo constant", ie, it contains no |
| 2097 |
* variables of the current query level and no uses of volatile functions. |
| 2098 |
* Such an expr is not necessarily a true constant: it can still contain |
| 2099 |
* Params and outer-level Vars, not to mention functions whose results |
| 2100 |
* may vary from one statement to the next. However, the expr's value |
| 2101 |
* will be constant over any one scan of the current query, so it can be |
| 2102 |
* used as, eg, an indexscan key. (Actually, the condition for indexscan |
| 2103 |
* keys is weaker than this; see is_pseudo_constant_for_index().) |
| 2104 |
* |
| 2105 |
* CAUTION: this function omits to test for one very important class of |
| 2106 |
* not-constant expressions, namely aggregates (Aggrefs). In current usage |
| 2107 |
* this is only applied to WHERE clauses and so a check for Aggrefs would be |
| 2108 |
* a waste of cycles; but be sure to also check contain_agg_clause() if you |
| 2109 |
* want to know about pseudo-constness in other contexts. The same goes |
| 2110 |
* for window functions (WindowFuncs). |
| 2111 |
*/ |
| 2112 |
bool |
| 2113 |
is_pseudo_constant_clause(Node *clause) |
| 2114 |
{ |
| 2115 |
/* |
| 2116 |
* We could implement this check in one recursive scan. But since the |
| 2117 |
* check for volatile functions is both moderately expensive and unlikely |
| 2118 |
* to fail, it seems better to look for Vars first and only check for |
| 2119 |
* volatile functions if we find no Vars. |
| 2120 |
*/ |
| 2121 |
if (!contain_var_clause(clause) && |
| 2122 |
!contain_volatile_functions(clause)) |
| 2123 |
return true; |
| 2124 |
return false; |
| 2125 |
} |
| 2126 |
|
| 2127 |
/* |
| 2128 |
* is_pseudo_constant_clause_relids |
| 2129 |
* Same as above, except caller already has available the var membership |
| 2130 |
* of the expression; this lets us avoid the contain_var_clause() scan. |
| 2131 |
*/ |
| 2132 |
bool |
| 2133 |
is_pseudo_constant_clause_relids(Node *clause, Relids relids) |
| 2134 |
{ |
| 2135 |
if (bms_is_empty(relids) && |
| 2136 |
!contain_volatile_functions(clause)) |
| 2137 |
return true; |
| 2138 |
return false; |
| 2139 |
} |
| 2140 |
|
| 2141 |
|
| 2142 |
/***************************************************************************** |
| 2143 |
* * |
| 2144 |
* General clause-manipulating routines * |
| 2145 |
* * |
| 2146 |
*****************************************************************************/ |
| 2147 |
|
| 2148 |
/* |
| 2149 |
* NumRelids |
| 2150 |
* (formerly clause_relids) |
| 2151 |
* |
| 2152 |
* Returns the number of different base relations referenced in 'clause'. |
| 2153 |
*/ |
| 2154 |
int |
| 2155 |
NumRelids(PlannerInfo *root, Node *clause) |
| 2156 |
{ |
| 2157 |
int result; |
| 2158 |
Relids varnos = pull_varnos(root, clause); |
| 2159 |
|
| 2160 |
varnos = bms_del_members(varnos, root->outer_join_rels); |
| 2161 |
result = bms_num_members(varnos); |
| 2162 |
bms_free(varnos); |
| 2163 |
return result; |
| 2164 |
} |
| 2165 |
|
| 2166 |
/* |
| 2167 |
* CommuteOpExpr: commute a binary operator clause |
| 2168 |
* |
| 2169 |
* XXX the clause is destructively modified! |
| 2170 |
*/ |
| 2171 |
void |
| 2172 |
CommuteOpExpr(OpExpr *clause) |
| 2173 |
{ |
| 2174 |
Oid opoid; |
| 2175 |
Node *temp; |
| 2176 |
|
| 2177 |
/* Sanity checks: caller is at fault if these fail */ |
| 2178 |
if (!is_opclause(clause) || |
| 2179 |
list_length(clause->args) != 2) |
| 2180 |
elog(ERROR, "cannot commute non-binary-operator clause"); |
| 2181 |
|
| 2182 |
opoid = get_commutator(clause->opno); |
| 2183 |
|
| 2184 |
if (!OidIsValid(opoid)) |
| 2185 |
elog(ERROR, "could not find commutator for operator %u", |
| 2186 |
clause->opno); |
| 2187 |
|
| 2188 |
/* |
| 2189 |
* modify the clause in-place! |
| 2190 |
*/ |
| 2191 |
clause->opno = opoid; |
| 2192 |
clause->opfuncid = InvalidOid; |
| 2193 |
/* opresulttype, opretset, opcollid, inputcollid need not change */ |
| 2194 |
|
| 2195 |
temp = linitial(clause->args); |
| 2196 |
linitial(clause->args) = lsecond(clause->args); |
| 2197 |
lsecond(clause->args) = temp; |
| 2198 |
} |
| 2199 |
|
| 2200 |
/* |
| 2201 |
* Helper for eval_const_expressions: check that datatype of an attribute |
| 2202 |
* is still what it was when the expression was parsed. This is needed to |
| 2203 |
* guard against improper simplification after ALTER COLUMN TYPE. (XXX we |
| 2204 |
* may well need to make similar checks elsewhere?) |
| 2205 |
* |
| 2206 |
* rowtypeid may come from a whole-row Var, and therefore it can be a domain |
| 2207 |
* over composite, but for this purpose we only care about checking the type |
| 2208 |
* of a contained field. |
| 2209 |
*/ |
| 2210 |
static bool |
| 2211 |
rowtype_field_matches(Oid rowtypeid, int fieldnum, |
| 2212 |
Oid expectedtype, int32 expectedtypmod, |
| 2213 |
Oid expectedcollation) |
| 2214 |
{ |
| 2215 |
TupleDesc tupdesc; |
| 2216 |
Form_pg_attribute attr; |
| 2217 |
|
| 2218 |
/* No issue for RECORD, since there is no way to ALTER such a type */ |
| 2219 |
if (rowtypeid == RECORDOID) |
| 2220 |
return true; |
| 2221 |
tupdesc = lookup_rowtype_tupdesc_domain(rowtypeid, -1, false); |
| 2222 |
if (fieldnum <= 0 || fieldnum > tupdesc->natts) |
| 2223 |
{ |
| 2224 |
ReleaseTupleDesc(tupdesc); |
| 2225 |
return false; |
| 2226 |
} |
| 2227 |
attr = TupleDescAttr(tupdesc, fieldnum - 1); |
| 2228 |
if (attr->attisdropped || |
| 2229 |
attr->atttypid != expectedtype || |
| 2230 |
attr->atttypmod != expectedtypmod || |
| 2231 |
attr->attcollation != expectedcollation) |
| 2232 |
{ |
| 2233 |
ReleaseTupleDesc(tupdesc); |
| 2234 |
return false; |
| 2235 |
} |
| 2236 |
ReleaseTupleDesc(tupdesc); |
| 2237 |
return true; |
| 2238 |
} |
| 2239 |
|
| 2240 |
|
| 2241 |
/*-------------------- |
| 2242 |
* eval_const_expressions |
| 2243 |
* |
| 2244 |
* Reduce any recognizably constant subexpressions of the given |
| 2245 |
* expression tree, for example "2 + 2" => "4". More interestingly, |
| 2246 |
* we can reduce certain boolean expressions even when they contain |
| 2247 |
* non-constant subexpressions: "x OR true" => "true" no matter what |
| 2248 |
* the subexpression x is. (XXX We assume that no such subexpression |
| 2249 |
* will have important side-effects, which is not necessarily a good |
| 2250 |
* assumption in the presence of user-defined functions; do we need a |
| 2251 |
* pg_proc flag that prevents discarding the execution of a function?) |
| 2252 |
* |
| 2253 |
* We do understand that certain functions may deliver non-constant |
| 2254 |
* results even with constant inputs, "nextval()" being the classic |
| 2255 |
* example. Functions that are not marked "immutable" in pg_proc |
| 2256 |
* will not be pre-evaluated here, although we will reduce their |
| 2257 |
* arguments as far as possible. |
| 2258 |
* |
| 2259 |
* Whenever a function is eliminated from the expression by means of |
| 2260 |
* constant-expression evaluation or inlining, we add the function to |
| 2261 |
* root->glob->invalItems. This ensures the plan is known to depend on |
| 2262 |
* such functions, even though they aren't referenced anymore. |
| 2263 |
* |
| 2264 |
* We assume that the tree has already been type-checked and contains |
| 2265 |
* only operators and functions that are reasonable to try to execute. |
| 2266 |
* |
| 2267 |
* NOTE: "root" can be passed as NULL if the caller never wants to do any |
| 2268 |
* Param substitutions nor receive info about inlined functions. |
| 2269 |
* |
| 2270 |
* NOTE: the planner assumes that this will always flatten nested AND and |
| 2271 |
* OR clauses into N-argument form. See comments in prepqual.c. |
| 2272 |
* |
| 2273 |
* NOTE: another critical effect is that any function calls that require |
| 2274 |
* default arguments will be expanded, and named-argument calls will be |
| 2275 |
* converted to positional notation. The executor won't handle either. |
| 2276 |
*-------------------- |
| 2277 |
*/ |
| 2278 |
Node * |
| 2279 |
eval_const_expressions(PlannerInfo *root, Node *node) |
| 2280 |
{ |
| 2281 |
eval_const_expressions_context context; |
| 2282 |
|
| 2283 |
if (root) |
| 2284 |
context.boundParams = root->glob->boundParams; /* bound Params */ |
| 2285 |
else |
| 2286 |
context.boundParams = NULL; |
| 2287 |
context.root = root; /* for inlined-function dependencies */ |
| 2288 |
context.active_fns = NIL; /* nothing being recursively simplified */ |
| 2289 |
context.case_val = NULL; /* no CASE being examined */ |
| 2290 |
context.estimate = false; /* safe transformations only */ |
| 2291 |
return eval_const_expressions_mutator(node, &context); |
| 2292 |
} |
| 2293 |
|
| 2294 |
#define MIN_ARRAY_SIZE_FOR_HASHED_SAOP 9 |
| 2295 |
/*-------------------- |
| 2296 |
* convert_saop_to_hashed_saop |
| 2297 |
* |
| 2298 |
* Recursively search 'node' for ScalarArrayOpExprs and fill in the hash |
| 2299 |
* function for any ScalarArrayOpExpr that looks like it would be useful to |
| 2300 |
* evaluate using a hash table rather than a linear search. |
| 2301 |
* |
| 2302 |
* We'll use a hash table if all of the following conditions are met: |
| 2303 |
* 1. The 2nd argument of the array contain only Consts. |
| 2304 |
* 2. useOr is true or there is a valid negator operator for the |
| 2305 |
* ScalarArrayOpExpr's opno. |
| 2306 |
* 3. There's valid hash function for both left and righthand operands and |
| 2307 |
* these hash functions are the same. |
| 2308 |
* 4. If the array contains enough elements for us to consider it to be |
| 2309 |
* worthwhile using a hash table rather than a linear search. |
| 2310 |
*/ |
| 2311 |
void |
| 2312 |
convert_saop_to_hashed_saop(Node *node) |
| 2313 |
{ |
| 2314 |
(void) convert_saop_to_hashed_saop_walker(node, NULL); |
| 2315 |
} |
| 2316 |
|
| 2317 |
static bool |
| 2318 |
convert_saop_to_hashed_saop_walker(Node *node, void *context) |
| 2319 |
{ |
| 2320 |
if (node == NULL) |
| 2321 |
return false; |
| 2322 |
|
| 2323 |
if (IsA(node, ScalarArrayOpExpr)) |
| 2324 |
{ |
| 2325 |
ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) node; |
| 2326 |
Node *leftarg = (Node *) linitial(saop->args); |
| 2327 |
Node *arrayarg = (Node *) lsecond(saop->args); |
| 2328 |
Oid lefthashfunc; |
| 2329 |
Oid righthashfunc; |
| 2330 |
|
| 2331 |
if (arrayarg && IsA(arrayarg, Const) && |
| 2332 |
!((Const *) arrayarg)->constisnull) |
| 2333 |
{ |
| 2334 |
if (saop->useOr) |
| 2335 |
{ |
| 2336 |
if (get_op_hash_functions_ext(saop->opno, exprType(leftarg), |
| 2337 |
&lefthashfunc, &righthashfunc) && |
| 2338 |
lefthashfunc == righthashfunc) |
| 2339 |
{ |
| 2340 |
Datum arrdatum = ((Const *) arrayarg)->constvalue; |
| 2341 |
ArrayType *arr = (ArrayType *) DatumGetPointer(arrdatum); |
| 2342 |
int nitems; |
| 2343 |
|
| 2344 |
/* |
| 2345 |
* Only fill in the hash functions if the array looks |
| 2346 |
* large enough for it to be worth hashing instead of |
| 2347 |
* doing a linear search. |
| 2348 |
*/ |
| 2349 |
nitems = ArrayGetNItems(ARR_NDIM(arr), ARR_DIMS(arr)); |
| 2350 |
|
| 2351 |
if (nitems >= MIN_ARRAY_SIZE_FOR_HASHED_SAOP) |
| 2352 |
{ |
| 2353 |
/* Looks good. Fill in the hash functions */ |
| 2354 |
saop->hashfuncid = lefthashfunc; |
| 2355 |
} |
| 2356 |
return false; |
| 2357 |
} |
| 2358 |
} |
| 2359 |
else /* !saop->useOr */ |
| 2360 |
{ |
| 2361 |
Oid negator = get_negator(saop->opno); |
| 2362 |
|
| 2363 |
/* |
| 2364 |
* Check if this is a NOT IN using an operator whose negator |
| 2365 |
* is hashable. If so we can still build a hash table and |
| 2366 |
* just ensure the lookup items are not in the hash table. |
| 2367 |
*/ |
| 2368 |
if (OidIsValid(negator) && |
| 2369 |
get_op_hash_functions_ext(negator, exprType(leftarg), |
| 2370 |
&lefthashfunc, &righthashfunc) && |
| 2371 |
lefthashfunc == righthashfunc) |
| 2372 |
{ |
| 2373 |
Datum arrdatum = ((Const *) arrayarg)->constvalue; |
| 2374 |
ArrayType *arr = (ArrayType *) DatumGetPointer(arrdatum); |
| 2375 |
int nitems; |
| 2376 |
|
| 2377 |
/* |
| 2378 |
* Only fill in the hash functions if the array looks |
| 2379 |
* large enough for it to be worth hashing instead of |
| 2380 |
* doing a linear search. |
| 2381 |
*/ |
| 2382 |
nitems = ArrayGetNItems(ARR_NDIM(arr), ARR_DIMS(arr)); |
| 2383 |
|
| 2384 |
if (nitems >= MIN_ARRAY_SIZE_FOR_HASHED_SAOP) |
| 2385 |
{ |
| 2386 |
/* Looks good. Fill in the hash functions */ |
| 2387 |
saop->hashfuncid = lefthashfunc; |
| 2388 |
|
| 2389 |
/* |
| 2390 |
* Also set the negfuncid. The executor will need |
| 2391 |
* that to perform hashtable lookups. |
| 2392 |
*/ |
| 2393 |
saop->negfuncid = get_opcode(negator); |
| 2394 |
} |
| 2395 |
return false; |
| 2396 |
} |
| 2397 |
} |
| 2398 |
} |
| 2399 |
} |
| 2400 |
|
| 2401 |
return expression_tree_walker(node, convert_saop_to_hashed_saop_walker, NULL); |
| 2402 |
} |
| 2403 |
|
| 2404 |
|
| 2405 |
/*-------------------- |
| 2406 |
* estimate_expression_value |
| 2407 |
* |
| 2408 |
* This function attempts to estimate the value of an expression for |
| 2409 |
* planning purposes. It is in essence a more aggressive version of |
| 2410 |
* eval_const_expressions(): we will perform constant reductions that are |
| 2411 |
* not necessarily 100% safe, but are reasonable for estimation purposes. |
| 2412 |
* |
| 2413 |
* Currently the extra steps that are taken in this mode are: |
| 2414 |
* 1. Substitute values for Params, where a bound Param value has been made |
| 2415 |
* available by the caller of planner(), even if the Param isn't marked |
| 2416 |
* constant. This effectively means that we plan using the first supplied |
| 2417 |
* value of the Param. |
| 2418 |
* 2. Fold stable, as well as immutable, functions to constants. |
| 2419 |
* 3. Reduce PlaceHolderVar nodes to their contained expressions. |
| 2420 |
*-------------------- |
| 2421 |
*/ |
| 2422 |
Node * |
| 2423 |
estimate_expression_value(PlannerInfo *root, Node *node) |
| 2424 |
{ |
| 2425 |
eval_const_expressions_context context; |
| 2426 |
|
| 2427 |
context.boundParams = root->glob->boundParams; /* bound Params */ |
| 2428 |
/* we do not need to mark the plan as depending on inlined functions */ |
| 2429 |
context.root = NULL; |
| 2430 |
context.active_fns = NIL; /* nothing being recursively simplified */ |
| 2431 |
context.case_val = NULL; /* no CASE being examined */ |
| 2432 |
context.estimate = true; /* unsafe transformations OK */ |
| 2433 |
return eval_const_expressions_mutator(node, &context); |
| 2434 |
} |
| 2435 |
|
| 2436 |
/* |
| 2437 |
* The generic case in eval_const_expressions_mutator is to recurse using |
| 2438 |
* expression_tree_mutator, which will copy the given node unchanged but |
| 2439 |
* const-simplify its arguments (if any) as far as possible. If the node |
| 2440 |
* itself does immutable processing, and each of its arguments were reduced |
| 2441 |
* to a Const, we can then reduce it to a Const using evaluate_expr. (Some |
| 2442 |
* node types need more complicated logic; for example, a CASE expression |
| 2443 |
* might be reducible to a constant even if not all its subtrees are.) |
| 2444 |
*/ |
| 2445 |
#define ece_generic_processing(node) \ |
| 2446 |
expression_tree_mutator((Node *) (node), eval_const_expressions_mutator, \ |
| 2447 |
context) |
| 2448 |
|
| 2449 |
/* |
| 2450 |
* Check whether all arguments of the given node were reduced to Consts. |
| 2451 |
* By going directly to expression_tree_walker, contain_non_const_walker |
| 2452 |
* is not applied to the node itself, only to its children. |
| 2453 |
*/ |
| 2454 |
#define ece_all_arguments_const(node) \ |
| 2455 |
(!expression_tree_walker((Node *) (node), contain_non_const_walker, NULL)) |
| 2456 |
|
| 2457 |
/* Generic macro for applying evaluate_expr */ |
| 2458 |
#define ece_evaluate_expr(node) \ |
| 2459 |
((Node *) evaluate_expr((Expr *) (node), \ |
| 2460 |
exprType((Node *) (node)), \ |
| 2461 |
exprTypmod((Node *) (node)), \ |
| 2462 |
exprCollation((Node *) (node)))) |
| 2463 |
|
| 2464 |
/* |
| 2465 |
* Recursive guts of eval_const_expressions/estimate_expression_value |
| 2466 |
*/ |
| 2467 |
static Node * |
| 2468 |
eval_const_expressions_mutator(Node *node, |
| 2469 |
eval_const_expressions_context *context) |
| 2470 |
{ |
| 2471 |
|
| 2472 |
/* since this function recurses, it could be driven to stack overflow */ |
| 2473 |
check_stack_depth(); |
| 2474 |
|
| 2475 |
if (node == NULL) |
| 2476 |
return NULL; |
| 2477 |
switch (nodeTag(node)) |
| 2478 |
{ |
| 2479 |
case T_Param: |
| 2480 |
{ |
| 2481 |
Param *param = (Param *) node; |
| 2482 |
ParamListInfo paramLI = context->boundParams; |
| 2483 |
|
| 2484 |
/* Look to see if we've been given a value for this Param */ |
| 2485 |
if (param->paramkind == PARAM_EXTERN && |
| 2486 |
paramLI != NULL && |
| 2487 |
param->paramid > 0 && |
| 2488 |
param->paramid <= paramLI->numParams) |
| 2489 |
{ |
| 2490 |
ParamExternData *prm; |
| 2491 |
ParamExternData prmdata; |
| 2492 |
|
| 2493 |
/* |
| 2494 |
* Give hook a chance in case parameter is dynamic. Tell |
| 2495 |
* it that this fetch is speculative, so it should avoid |
| 2496 |
* erroring out if parameter is unavailable. |
| 2497 |
*/ |
| 2498 |
if (paramLI->paramFetch != NULL) |
| 2499 |
prm = paramLI->paramFetch(paramLI, param->paramid, |
| 2500 |
true, &prmdata); |
| 2501 |
else |
| 2502 |
prm = ¶mLI->params[param->paramid - 1]; |
| 2503 |
|
| 2504 |
/* |
| 2505 |
* We don't just check OidIsValid, but insist that the |
| 2506 |
* fetched type match the Param, just in case the hook did |
| 2507 |
* something unexpected. No need to throw an error here |
| 2508 |
* though; leave that for runtime. |
| 2509 |
*/ |
| 2510 |
if (OidIsValid(prm->ptype) && |
| 2511 |
prm->ptype == param->paramtype) |
| 2512 |
{ |
| 2513 |
/* OK to substitute parameter value? */ |
| 2514 |
if (context->estimate || |
| 2515 |
(prm->pflags & PARAM_FLAG_CONST)) |
| 2516 |
{ |
| 2517 |
/* |
| 2518 |
* Return a Const representing the param value. |
| 2519 |
* Must copy pass-by-ref datatypes, since the |
| 2520 |
* Param might be in a memory context |
| 2521 |
* shorter-lived than our output plan should be. |
| 2522 |
*/ |
| 2523 |
int16 typLen; |
| 2524 |
bool typByVal; |
| 2525 |
Datum pval; |
| 2526 |
Const *con; |
| 2527 |
|
| 2528 |
get_typlenbyval(param->paramtype, |
| 2529 |
&typLen, &typByVal); |
| 2530 |
if (prm->isnull || typByVal) |
| 2531 |
pval = prm->value; |
| 2532 |
else |
| 2533 |
pval = datumCopy(prm->value, typByVal, typLen); |
| 2534 |
con = makeConst(param->paramtype, |
| 2535 |
param->paramtypmod, |
| 2536 |
param->paramcollid, |
| 2537 |
(int) typLen, |
| 2538 |
pval, |
| 2539 |
prm->isnull, |
| 2540 |
typByVal); |
| 2541 |
con->location = param->location; |
| 2542 |
return (Node *) con; |
| 2543 |
} |
| 2544 |
} |
| 2545 |
} |
| 2546 |
|
| 2547 |
/* |
| 2548 |
* Not replaceable, so just copy the Param (no need to |
| 2549 |
* recurse) |
| 2550 |
*/ |
| 2551 |
return (Node *) copyObject(param); |
| 2552 |
} |
| 2553 |
case T_WindowFunc: |
| 2554 |
{ |
| 2555 |
WindowFunc *expr = (WindowFunc *) node; |
| 2556 |
Oid funcid = expr->winfnoid; |
| 2557 |
List *args; |
| 2558 |
Expr *aggfilter; |
| 2559 |
HeapTuple func_tuple; |
| 2560 |
WindowFunc *newexpr; |
| 2561 |
|
| 2562 |
/* |
| 2563 |
* We can't really simplify a WindowFunc node, but we mustn't |
| 2564 |
* just fall through to the default processing, because we |
| 2565 |
* have to apply expand_function_arguments to its argument |
| 2566 |
* list. That takes care of inserting default arguments and |
| 2567 |
* expanding named-argument notation. |
| 2568 |
*/ |
| 2569 |
func_tuple = SearchSysCache1(PROCOID, ObjectIdGetDatum(funcid)); |
| 2570 |
if (!HeapTupleIsValid(func_tuple)) |
| 2571 |
elog(ERROR, "cache lookup failed for function %u", funcid); |
| 2572 |
|
| 2573 |
args = expand_function_arguments(expr->args, |
| 2574 |
false, expr->wintype, |
| 2575 |
func_tuple); |
| 2576 |
|
| 2577 |
ReleaseSysCache(func_tuple); |
| 2578 |
|
| 2579 |
/* Now, recursively simplify the args (which are a List) */ |
| 2580 |
args = (List *) |
| 2581 |
expression_tree_mutator((Node *) args, |
| 2582 |
eval_const_expressions_mutator, |
| 2583 |
context); |
| 2584 |
/* ... and the filter expression, which isn't */ |
| 2585 |
aggfilter = (Expr *) |
| 2586 |
eval_const_expressions_mutator((Node *) expr->aggfilter, |
| 2587 |
context); |
| 2588 |
|
| 2589 |
/* And build the replacement WindowFunc node */ |
| 2590 |
newexpr = makeNode(WindowFunc); |
| 2591 |
newexpr->winfnoid = expr->winfnoid; |
| 2592 |
newexpr->wintype = expr->wintype; |
| 2593 |
newexpr->wincollid = expr->wincollid; |
| 2594 |
newexpr->inputcollid = expr->inputcollid; |
| 2595 |
newexpr->args = args; |
| 2596 |
newexpr->aggfilter = aggfilter; |
| 2597 |
newexpr->runCondition = expr->runCondition; |
| 2598 |
newexpr->winref = expr->winref; |
| 2599 |
newexpr->winstar = expr->winstar; |
| 2600 |
newexpr->winagg = expr->winagg; |
| 2601 |
newexpr->location = expr->location; |
| 2602 |
|
| 2603 |
return (Node *) newexpr; |
| 2604 |
} |
| 2605 |
case T_FuncExpr: |
| 2606 |
{ |
| 2607 |
FuncExpr *expr = (FuncExpr *) node; |
| 2608 |
List *args = expr->args; |
| 2609 |
Expr *simple; |
| 2610 |
FuncExpr *newexpr; |
| 2611 |
|
| 2612 |
/* |
| 2613 |
* Code for op/func reduction is pretty bulky, so split it out |
| 2614 |
* as a separate function. Note: exprTypmod normally returns |
| 2615 |
* -1 for a FuncExpr, but not when the node is recognizably a |
| 2616 |
* length coercion; we want to preserve the typmod in the |
| 2617 |
* eventual Const if so. |
| 2618 |
*/ |
| 2619 |
simple = simplify_function(expr->funcid, |
| 2620 |
expr->funcresulttype, |
| 2621 |
exprTypmod(node), |
| 2622 |
expr->funccollid, |
| 2623 |
expr->inputcollid, |
| 2624 |
&args, |
| 2625 |
expr->funcvariadic, |
| 2626 |
true, |
| 2627 |
true, |
| 2628 |
context); |
| 2629 |
if (simple) /* successfully simplified it */ |
| 2630 |
return (Node *) simple; |
| 2631 |
|
| 2632 |
/* |
| 2633 |
* The expression cannot be simplified any further, so build |
| 2634 |
* and return a replacement FuncExpr node using the |
| 2635 |
* possibly-simplified arguments. Note that we have also |
| 2636 |
* converted the argument list to positional notation. |
| 2637 |
*/ |
| 2638 |
newexpr = makeNode(FuncExpr); |
| 2639 |
newexpr->funcid = expr->funcid; |
| 2640 |
newexpr->funcresulttype = expr->funcresulttype; |
| 2641 |
newexpr->funcretset = expr->funcretset; |
| 2642 |
newexpr->funcvariadic = expr->funcvariadic; |
| 2643 |
newexpr->funcformat = expr->funcformat; |
| 2644 |
newexpr->funccollid = expr->funccollid; |
| 2645 |
newexpr->inputcollid = expr->inputcollid; |
| 2646 |
newexpr->args = args; |
| 2647 |
newexpr->location = expr->location; |
| 2648 |
return (Node *) newexpr; |
| 2649 |
} |
| 2650 |
case T_OpExpr: |
| 2651 |
{ |
| 2652 |
OpExpr *expr = (OpExpr *) node; |
| 2653 |
List *args = expr->args; |
| 2654 |
Expr *simple; |
| 2655 |
OpExpr *newexpr; |
| 2656 |
|
| 2657 |
/* |
| 2658 |
* Need to get OID of underlying function. Okay to scribble |
| 2659 |
* on input to this extent. |
| 2660 |
*/ |
| 2661 |
set_opfuncid(expr); |
| 2662 |
|
| 2663 |
/* |
| 2664 |
* Code for op/func reduction is pretty bulky, so split it out |
| 2665 |
* as a separate function. |
| 2666 |
*/ |
| 2667 |
simple = simplify_function(expr->opfuncid, |
| 2668 |
expr->opresulttype, -1, |
| 2669 |
expr->opcollid, |
| 2670 |
expr->inputcollid, |
| 2671 |
&args, |
| 2672 |
false, |
| 2673 |
true, |
| 2674 |
true, |
| 2675 |
context); |
| 2676 |
if (simple) /* successfully simplified it */ |
| 2677 |
return (Node *) simple; |
| 2678 |
|
| 2679 |
/* |
| 2680 |
* If the operator is boolean equality or inequality, we know |
| 2681 |
* how to simplify cases involving one constant and one |
| 2682 |
* non-constant argument. |
| 2683 |
*/ |
| 2684 |
if (expr->opno == BooleanEqualOperator || |
| 2685 |
expr->opno == BooleanNotEqualOperator) |
| 2686 |
{ |
| 2687 |
simple = (Expr *) simplify_boolean_equality(expr->opno, |
| 2688 |
args); |
| 2689 |
if (simple) /* successfully simplified it */ |
| 2690 |
return (Node *) simple; |
| 2691 |
} |
| 2692 |
|
| 2693 |
/* |
| 2694 |
* The expression cannot be simplified any further, so build |
| 2695 |
* and return a replacement OpExpr node using the |
| 2696 |
* possibly-simplified arguments. |
| 2697 |
*/ |
| 2698 |
newexpr = makeNode(OpExpr); |
| 2699 |
newexpr->opno = expr->opno; |
| 2700 |
newexpr->opfuncid = expr->opfuncid; |
| 2701 |
newexpr->opresulttype = expr->opresulttype; |
| 2702 |
newexpr->opretset = expr->opretset; |
| 2703 |
newexpr->opcollid = expr->opcollid; |
| 2704 |
newexpr->inputcollid = expr->inputcollid; |
| 2705 |
newexpr->args = args; |
| 2706 |
newexpr->location = expr->location; |
| 2707 |
return (Node *) newexpr; |
| 2708 |
} |
| 2709 |
case T_DistinctExpr: |
| 2710 |
{ |
| 2711 |
DistinctExpr *expr = (DistinctExpr *) node; |
| 2712 |
List *args; |
| 2713 |
ListCell *arg; |
| 2714 |
bool has_null_input = false; |
| 2715 |
bool all_null_input = true; |
| 2716 |
bool has_nonconst_input = false; |
| 2717 |
Expr *simple; |
| 2718 |
DistinctExpr *newexpr; |
| 2719 |
|
| 2720 |
/* |
| 2721 |
* Reduce constants in the DistinctExpr's arguments. We know |
| 2722 |
* args is either NIL or a List node, so we can call |
| 2723 |
* expression_tree_mutator directly rather than recursing to |
| 2724 |
* self. |
| 2725 |
*/ |
| 2726 |
args = (List *) expression_tree_mutator((Node *) expr->args, |
| 2727 |
eval_const_expressions_mutator, |
| 2728 |
context); |
| 2729 |
|
| 2730 |
/* |
| 2731 |
* We must do our own check for NULLs because DistinctExpr has |
| 2732 |
* different results for NULL input than the underlying |
| 2733 |
* operator does. |
| 2734 |
*/ |
| 2735 |
foreach(arg, args) |
| 2736 |
{ |
| 2737 |
if (IsA(lfirst(arg), Const)) |
| 2738 |
{ |
| 2739 |
has_null_input |= ((Const *) lfirst(arg))->constisnull; |
| 2740 |
all_null_input &= ((Const *) lfirst(arg))->constisnull; |
| 2741 |
} |
| 2742 |
else |
| 2743 |
has_nonconst_input = true; |
| 2744 |
} |
| 2745 |
|
| 2746 |
/* all constants? then can optimize this out */ |
| 2747 |
if (!has_nonconst_input) |
| 2748 |
{ |
| 2749 |
/* all nulls? then not distinct */ |
| 2750 |
if (all_null_input) |
| 2751 |
return makeBoolConst(false, false); |
| 2752 |
|
| 2753 |
/* one null? then distinct */ |
| 2754 |
if (has_null_input) |
| 2755 |
return makeBoolConst(true, false); |
| 2756 |
|
| 2757 |
/* otherwise try to evaluate the '=' operator */ |
| 2758 |
/* (NOT okay to try to inline it, though!) */ |
| 2759 |
|
| 2760 |
/* |
| 2761 |
* Need to get OID of underlying function. Okay to |
| 2762 |
* scribble on input to this extent. |
| 2763 |
*/ |
| 2764 |
set_opfuncid((OpExpr *) expr); /* rely on struct |
| 2765 |
* equivalence */ |
| 2766 |
|
| 2767 |
/* |
| 2768 |
* Code for op/func reduction is pretty bulky, so split it |
| 2769 |
* out as a separate function. |
| 2770 |
*/ |
| 2771 |
simple = simplify_function(expr->opfuncid, |
| 2772 |
expr->opresulttype, -1, |
| 2773 |
expr->opcollid, |
| 2774 |
expr->inputcollid, |
| 2775 |
&args, |
| 2776 |
false, |
| 2777 |
false, |
| 2778 |
false, |
| 2779 |
context); |
| 2780 |
if (simple) /* successfully simplified it */ |
| 2781 |
{ |
| 2782 |
/* |
| 2783 |
* Since the underlying operator is "=", must negate |
| 2784 |
* its result |
| 2785 |
*/ |
| 2786 |
Const *csimple = castNode(Const, simple); |
| 2787 |
|
| 2788 |
csimple->constvalue = |
| 2789 |
BoolGetDatum(!DatumGetBool(csimple->constvalue)); |
| 2790 |
return (Node *) csimple; |
| 2791 |
} |
| 2792 |
} |
| 2793 |
|
| 2794 |
/* |
| 2795 |
* The expression cannot be simplified any further, so build |
| 2796 |
* and return a replacement DistinctExpr node using the |
| 2797 |
* possibly-simplified arguments. |
| 2798 |
*/ |
| 2799 |
newexpr = makeNode(DistinctExpr); |
| 2800 |
newexpr->opno = expr->opno; |
| 2801 |
newexpr->opfuncid = expr->opfuncid; |
| 2802 |
newexpr->opresulttype = expr->opresulttype; |
| 2803 |
newexpr->opretset = expr->opretset; |
| 2804 |
newexpr->opcollid = expr->opcollid; |
| 2805 |
newexpr->inputcollid = expr->inputcollid; |
| 2806 |
newexpr->args = args; |
| 2807 |
newexpr->location = expr->location; |
| 2808 |
return (Node *) newexpr; |
| 2809 |
} |
| 2810 |
case T_NullIfExpr: |
| 2811 |
{ |
| 2812 |
NullIfExpr *expr; |
| 2813 |
ListCell *arg; |
| 2814 |
bool has_nonconst_input = false; |
| 2815 |
|
| 2816 |
/* Copy the node and const-simplify its arguments */ |
| 2817 |
expr = (NullIfExpr *) ece_generic_processing(node); |
| 2818 |
|
| 2819 |
/* If either argument is NULL they can't be equal */ |
| 2820 |
foreach(arg, expr->args) |
| 2821 |
{ |
| 2822 |
if (!IsA(lfirst(arg), Const)) |
| 2823 |
has_nonconst_input = true; |
| 2824 |
else if (((Const *) lfirst(arg))->constisnull) |
| 2825 |
return (Node *) linitial(expr->args); |
| 2826 |
} |
| 2827 |
|
| 2828 |
/* |
| 2829 |
* Need to get OID of underlying function before checking if |
| 2830 |
* the function is OK to evaluate. |
| 2831 |
*/ |
| 2832 |
set_opfuncid((OpExpr *) expr); |
| 2833 |
|
| 2834 |
if (!has_nonconst_input && |
| 2835 |
ece_function_is_safe(expr->opfuncid, context)) |
| 2836 |
return ece_evaluate_expr(expr); |
| 2837 |
|
| 2838 |
return (Node *) expr; |
| 2839 |
} |
| 2840 |
case T_ScalarArrayOpExpr: |
| 2841 |
{ |
| 2842 |
ScalarArrayOpExpr *saop; |
| 2843 |
|
| 2844 |
/* Copy the node and const-simplify its arguments */ |
| 2845 |
saop = (ScalarArrayOpExpr *) ece_generic_processing(node); |
| 2846 |
|
| 2847 |
/* Make sure we know underlying function */ |
| 2848 |
set_sa_opfuncid(saop); |
| 2849 |
|
| 2850 |
/* |
| 2851 |
* If all arguments are Consts, and it's a safe function, we |
| 2852 |
* can fold to a constant |
| 2853 |
*/ |
| 2854 |
if (ece_all_arguments_const(saop) && |
| 2855 |
ece_function_is_safe(saop->opfuncid, context)) |
| 2856 |
return ece_evaluate_expr(saop); |
| 2857 |
return (Node *) saop; |
| 2858 |
} |
| 2859 |
case T_BoolExpr: |
| 2860 |
{ |
| 2861 |
BoolExpr *expr = (BoolExpr *) node; |
| 2862 |
|
| 2863 |
switch (expr->boolop) |
| 2864 |
{ |
| 2865 |
case OR_EXPR: |
| 2866 |
{ |
| 2867 |
List *newargs; |
| 2868 |
bool haveNull = false; |
| 2869 |
bool forceTrue = false; |
| 2870 |
|
| 2871 |
newargs = simplify_or_arguments(expr->args, |
| 2872 |
context, |
| 2873 |
&haveNull, |
| 2874 |
&forceTrue); |
| 2875 |
if (forceTrue) |
| 2876 |
return makeBoolConst(true, false); |
| 2877 |
if (haveNull) |
| 2878 |
newargs = lappend(newargs, |
| 2879 |
makeBoolConst(false, true)); |
| 2880 |
/* If all the inputs are FALSE, result is FALSE */ |
| 2881 |
if (newargs == NIL) |
| 2882 |
return makeBoolConst(false, false); |
| 2883 |
|
| 2884 |
/* |
| 2885 |
* If only one nonconst-or-NULL input, it's the |
| 2886 |
* result |
| 2887 |
*/ |
| 2888 |
if (list_length(newargs) == 1) |
| 2889 |
return (Node *) linitial(newargs); |
| 2890 |
/* Else we still need an OR node */ |
| 2891 |
return (Node *) make_orclause(newargs); |
| 2892 |
} |
| 2893 |
case AND_EXPR: |
| 2894 |
{ |
| 2895 |
List *newargs; |
| 2896 |
bool haveNull = false; |
| 2897 |
bool forceFalse = false; |
| 2898 |
|
| 2899 |
newargs = simplify_and_arguments(expr->args, |
| 2900 |
context, |
| 2901 |
&haveNull, |
| 2902 |
&forceFalse); |
| 2903 |
if (forceFalse) |
| 2904 |
return makeBoolConst(false, false); |
| 2905 |
if (haveNull) |
| 2906 |
newargs = lappend(newargs, |
| 2907 |
makeBoolConst(false, true)); |
| 2908 |
/* If all the inputs are TRUE, result is TRUE */ |
| 2909 |
if (newargs == NIL) |
| 2910 |
return makeBoolConst(true, false); |
| 2911 |
|
| 2912 |
/* |
| 2913 |
* If only one nonconst-or-NULL input, it's the |
| 2914 |
* result |
| 2915 |
*/ |
| 2916 |
if (list_length(newargs) == 1) |
| 2917 |
return (Node *) linitial(newargs); |
| 2918 |
/* Else we still need an AND node */ |
| 2919 |
return (Node *) make_andclause(newargs); |
| 2920 |
} |
| 2921 |
case NOT_EXPR: |
| 2922 |
{ |
| 2923 |
Node *arg; |
| 2924 |
|
| 2925 |
Assert(list_length(expr->args) == 1); |
| 2926 |
arg = eval_const_expressions_mutator(linitial(expr->args), |
| 2927 |
context); |
| 2928 |
|
| 2929 |
/* |
| 2930 |
* Use negate_clause() to see if we can simplify |
| 2931 |
* away the NOT. |
| 2932 |
*/ |
| 2933 |
return negate_clause(arg); |
| 2934 |
} |
| 2935 |
default: |
| 2936 |
elog(ERROR, "unrecognized boolop: %d", |
| 2937 |
(int) expr->boolop); |
| 2938 |
break; |
| 2939 |
} |
| 2940 |
break; |
| 2941 |
} |
| 2942 |
|
| 2943 |
case T_JsonValueExpr: |
| 2944 |
{ |
| 2945 |
JsonValueExpr *jve = (JsonValueExpr *) node; |
| 2946 |
Node *raw_expr = (Node *) jve->raw_expr; |
| 2947 |
Node *formatted_expr = (Node *) jve->formatted_expr; |
| 2948 |
|
| 2949 |
/* |
| 2950 |
* If we can fold formatted_expr to a constant, we can elide |
| 2951 |
* the JsonValueExpr altogether. Otherwise we must process |
| 2952 |
* raw_expr too. But JsonFormat is a flat node and requires |
| 2953 |
* no simplification, only copying. |
| 2954 |
*/ |
| 2955 |
formatted_expr = eval_const_expressions_mutator(formatted_expr, |
| 2956 |
context); |
| 2957 |
if (formatted_expr && IsA(formatted_expr, Const)) |
| 2958 |
return formatted_expr; |
| 2959 |
|
| 2960 |
raw_expr = eval_const_expressions_mutator(raw_expr, context); |
| 2961 |
|
| 2962 |
return (Node *) makeJsonValueExpr((Expr *) raw_expr, |
| 2963 |
(Expr *) formatted_expr, |
| 2964 |
copyObject(jve->format)); |
| 2965 |
} |
| 2966 |
|
| 2967 |
case T_SubPlan: |
| 2968 |
case T_AlternativeSubPlan: |
| 2969 |
|
| 2970 |
/* |
| 2971 |
* Return a SubPlan unchanged --- too late to do anything with it. |
| 2972 |
* |
| 2973 |
* XXX should we ereport() here instead? Probably this routine |
| 2974 |
* should never be invoked after SubPlan creation. |
| 2975 |
*/ |
| 2976 |
return node; |
| 2977 |
case T_RelabelType: |
| 2978 |
{ |
| 2979 |
RelabelType *relabel = (RelabelType *) node; |
| 2980 |
Node *arg; |
| 2981 |
|
| 2982 |
/* Simplify the input ... */ |
| 2983 |
arg = eval_const_expressions_mutator((Node *) relabel->arg, |
| 2984 |
context); |
| 2985 |
/* ... and attach a new RelabelType node, if needed */ |
| 2986 |
return applyRelabelType(arg, |
| 2987 |
relabel->resulttype, |
| 2988 |
relabel->resulttypmod, |
| 2989 |
relabel->resultcollid, |
| 2990 |
relabel->relabelformat, |
| 2991 |
relabel->location, |
| 2992 |
true); |
| 2993 |
} |
| 2994 |
case T_CoerceViaIO: |
| 2995 |
{ |
| 2996 |
CoerceViaIO *expr = (CoerceViaIO *) node; |
| 2997 |
List *args; |
| 2998 |
Oid outfunc; |
| 2999 |
bool outtypisvarlena; |
| 3000 |
Oid infunc; |
| 3001 |
Oid intypioparam; |
| 3002 |
Expr *simple; |
| 3003 |
CoerceViaIO *newexpr; |
| 3004 |
|
| 3005 |
/* Make a List so we can use simplify_function */ |
| 3006 |
args = list_make1(expr->arg); |
| 3007 |
|
| 3008 |
/* |
| 3009 |
* CoerceViaIO represents calling the source type's output |
| 3010 |
* function then the result type's input function. So, try to |
| 3011 |
* simplify it as though it were a stack of two such function |
| 3012 |
* calls. First we need to know what the functions are. |
| 3013 |
* |
| 3014 |
* Note that the coercion functions are assumed not to care |
| 3015 |
* about input collation, so we just pass InvalidOid for that. |
| 3016 |
*/ |
| 3017 |
getTypeOutputInfo(exprType((Node *) expr->arg), |
| 3018 |
&outfunc, &outtypisvarlena); |
| 3019 |
getTypeInputInfo(expr->resulttype, |
| 3020 |
&infunc, &intypioparam); |
| 3021 |
|
| 3022 |
simple = simplify_function(outfunc, |
| 3023 |
CSTRINGOID, -1, |
| 3024 |
InvalidOid, |
| 3025 |
InvalidOid, |
| 3026 |
&args, |
| 3027 |
false, |
| 3028 |
true, |
| 3029 |
true, |
| 3030 |
context); |
| 3031 |
if (simple) /* successfully simplified output fn */ |
| 3032 |
{ |
| 3033 |
/* |
| 3034 |
* Input functions may want 1 to 3 arguments. We always |
| 3035 |
* supply all three, trusting that nothing downstream will |
| 3036 |
* complain. |
| 3037 |
*/ |
| 3038 |
args = list_make3(simple, |
| 3039 |
makeConst(OIDOID, |
| 3040 |
-1, |
| 3041 |
InvalidOid, |
| 3042 |
sizeof(Oid), |
| 3043 |
ObjectIdGetDatum(intypioparam), |
| 3044 |
false, |
| 3045 |
true), |
| 3046 |
makeConst(INT4OID, |
| 3047 |
-1, |
| 3048 |
InvalidOid, |
| 3049 |
sizeof(int32), |
| 3050 |
Int32GetDatum(-1), |
| 3051 |
false, |
| 3052 |
true)); |
| 3053 |
|
| 3054 |
simple = simplify_function(infunc, |
| 3055 |
expr->resulttype, -1, |
| 3056 |
expr->resultcollid, |
| 3057 |
InvalidOid, |
| 3058 |
&args, |
| 3059 |
false, |
| 3060 |
false, |
| 3061 |
true, |
| 3062 |
context); |
| 3063 |
if (simple) /* successfully simplified input fn */ |
| 3064 |
return (Node *) simple; |
| 3065 |
} |
| 3066 |
|
| 3067 |
/* |
| 3068 |
* The expression cannot be simplified any further, so build |
| 3069 |
* and return a replacement CoerceViaIO node using the |
| 3070 |
* possibly-simplified argument. |
| 3071 |
*/ |
| 3072 |
newexpr = makeNode(CoerceViaIO); |
| 3073 |
newexpr->arg = (Expr *) linitial(args); |
| 3074 |
newexpr->resulttype = expr->resulttype; |
| 3075 |
newexpr->resultcollid = expr->resultcollid; |
| 3076 |
newexpr->coerceformat = expr->coerceformat; |
| 3077 |
newexpr->location = expr->location; |
| 3078 |
return (Node *) newexpr; |
| 3079 |
} |
| 3080 |
case T_ArrayCoerceExpr: |
| 3081 |
{ |
| 3082 |
ArrayCoerceExpr *ac = makeNode(ArrayCoerceExpr); |
| 3083 |
Node *save_case_val; |
| 3084 |
|
| 3085 |
/* |
| 3086 |
* Copy the node and const-simplify its arguments. We can't |
| 3087 |
* use ece_generic_processing() here because we need to mess |
| 3088 |
* with case_val only while processing the elemexpr. |
| 3089 |
*/ |
| 3090 |
memcpy(ac, node, sizeof(ArrayCoerceExpr)); |
| 3091 |
ac->arg = (Expr *) |
| 3092 |
eval_const_expressions_mutator((Node *) ac->arg, |
| 3093 |
context); |
| 3094 |
|
| 3095 |
/* |
| 3096 |
* Set up for the CaseTestExpr node contained in the elemexpr. |
| 3097 |
* We must prevent it from absorbing any outer CASE value. |
| 3098 |
*/ |
| 3099 |
save_case_val = context->case_val; |
| 3100 |
context->case_val = NULL; |
| 3101 |
|
| 3102 |
ac->elemexpr = (Expr *) |
| 3103 |
eval_const_expressions_mutator((Node *) ac->elemexpr, |
| 3104 |
context); |
| 3105 |
|
| 3106 |
context->case_val = save_case_val; |
| 3107 |
|
| 3108 |
/* |
| 3109 |
* If constant argument and the per-element expression is |
| 3110 |
* immutable, we can simplify the whole thing to a constant. |
| 3111 |
* Exception: although contain_mutable_functions considers |
| 3112 |
* CoerceToDomain immutable for historical reasons, let's not |
| 3113 |
* do so here; this ensures coercion to an array-over-domain |
| 3114 |
* does not apply the domain's constraints until runtime. |
| 3115 |
*/ |
| 3116 |
if (ac->arg && IsA(ac->arg, Const) && |
| 3117 |
ac->elemexpr && !IsA(ac->elemexpr, CoerceToDomain) && |
| 3118 |
!contain_mutable_functions((Node *) ac->elemexpr)) |
| 3119 |
return ece_evaluate_expr(ac); |
| 3120 |
|
| 3121 |
return (Node *) ac; |
| 3122 |
} |
| 3123 |
case T_CollateExpr: |
| 3124 |
{ |
| 3125 |
/* |
| 3126 |
* We replace CollateExpr with RelabelType, so as to improve |
| 3127 |
* uniformity of expression representation and thus simplify |
| 3128 |
* comparison of expressions. Hence this looks very nearly |
| 3129 |
* the same as the RelabelType case, and we can apply the same |
| 3130 |
* optimizations to avoid unnecessary RelabelTypes. |
| 3131 |
*/ |
| 3132 |
CollateExpr *collate = (CollateExpr *) node; |
| 3133 |
Node *arg; |
| 3134 |
|
| 3135 |
/* Simplify the input ... */ |
| 3136 |
arg = eval_const_expressions_mutator((Node *) collate->arg, |
| 3137 |
context); |
| 3138 |
/* ... and attach a new RelabelType node, if needed */ |
| 3139 |
return applyRelabelType(arg, |
| 3140 |
exprType(arg), |
| 3141 |
exprTypmod(arg), |
| 3142 |
collate->collOid, |
| 3143 |
COERCE_IMPLICIT_CAST, |
| 3144 |
collate->location, |
| 3145 |
true); |
| 3146 |
} |
| 3147 |
case T_CaseExpr: |
| 3148 |
{ |
| 3149 |
/*---------- |
| 3150 |
* CASE expressions can be simplified if there are constant |
| 3151 |
* condition clauses: |
| 3152 |
* FALSE (or NULL): drop the alternative |
| 3153 |
* TRUE: drop all remaining alternatives |
| 3154 |
* If the first non-FALSE alternative is a constant TRUE, |
| 3155 |
* we can simplify the entire CASE to that alternative's |
| 3156 |
* expression. If there are no non-FALSE alternatives, |
| 3157 |
* we simplify the entire CASE to the default result (ELSE). |
| 3158 |
* |
| 3159 |
* If we have a simple-form CASE with constant test |
| 3160 |
* expression, we substitute the constant value for contained |
| 3161 |
* CaseTestExpr placeholder nodes, so that we have the |
| 3162 |
* opportunity to reduce constant test conditions. For |
| 3163 |
* example this allows |
| 3164 |
* CASE 0 WHEN 0 THEN 1 ELSE 1/0 END |
| 3165 |
* to reduce to 1 rather than drawing a divide-by-0 error. |
| 3166 |
* Note that when the test expression is constant, we don't |
| 3167 |
* have to include it in the resulting CASE; for example |
| 3168 |
* CASE 0 WHEN x THEN y ELSE z END |
| 3169 |
* is transformed by the parser to |
| 3170 |
* CASE 0 WHEN CaseTestExpr = x THEN y ELSE z END |
| 3171 |
* which we can simplify to |
| 3172 |
* CASE WHEN 0 = x THEN y ELSE z END |
| 3173 |
* It is not necessary for the executor to evaluate the "arg" |
| 3174 |
* expression when executing the CASE, since any contained |
| 3175 |
* CaseTestExprs that might have referred to it will have been |
| 3176 |
* replaced by the constant. |
| 3177 |
*---------- |
| 3178 |
*/ |
| 3179 |
CaseExpr *caseexpr = (CaseExpr *) node; |
| 3180 |
CaseExpr *newcase; |
| 3181 |
Node *save_case_val; |
| 3182 |
Node *newarg; |
| 3183 |
List *newargs; |
| 3184 |
bool const_true_cond; |
| 3185 |
Node *defresult = NULL; |
| 3186 |
ListCell *arg; |
| 3187 |
|
| 3188 |
/* Simplify the test expression, if any */ |
| 3189 |
newarg = eval_const_expressions_mutator((Node *) caseexpr->arg, |
| 3190 |
context); |
| 3191 |
|
| 3192 |
/* Set up for contained CaseTestExpr nodes */ |
| 3193 |
save_case_val = context->case_val; |
| 3194 |
if (newarg && IsA(newarg, Const)) |
| 3195 |
{ |
| 3196 |
context->case_val = newarg; |
| 3197 |
newarg = NULL; /* not needed anymore, see above */ |
| 3198 |
} |
| 3199 |
else |
| 3200 |
context->case_val = NULL; |
| 3201 |
|
| 3202 |
/* Simplify the WHEN clauses */ |
| 3203 |
newargs = NIL; |
| 3204 |
const_true_cond = false; |
| 3205 |
foreach(arg, caseexpr->args) |
| 3206 |
{ |
| 3207 |
CaseWhen *oldcasewhen = lfirst_node(CaseWhen, arg); |
| 3208 |
Node *casecond; |
| 3209 |
Node *caseresult; |
| 3210 |
|
| 3211 |
/* Simplify this alternative's test condition */ |
| 3212 |
casecond = eval_const_expressions_mutator((Node *) oldcasewhen->expr, |
| 3213 |
context); |
| 3214 |
|
| 3215 |
/* |
| 3216 |
* If the test condition is constant FALSE (or NULL), then |
| 3217 |
* drop this WHEN clause completely, without processing |
| 3218 |
* the result. |
| 3219 |
*/ |
| 3220 |
if (casecond && IsA(casecond, Const)) |
| 3221 |
{ |
| 3222 |
Const *const_input = (Const *) casecond; |
| 3223 |
|
| 3224 |
if (const_input->constisnull || |
| 3225 |
!DatumGetBool(const_input->constvalue)) |
| 3226 |
continue; /* drop alternative with FALSE cond */ |
| 3227 |
/* Else it's constant TRUE */ |
| 3228 |
const_true_cond = true; |
| 3229 |
} |
| 3230 |
|
| 3231 |
/* Simplify this alternative's result value */ |
| 3232 |
caseresult = eval_const_expressions_mutator((Node *) oldcasewhen->result, |
| 3233 |
context); |
| 3234 |
|
| 3235 |
/* If non-constant test condition, emit a new WHEN node */ |
| 3236 |
if (!const_true_cond) |
| 3237 |
{ |
| 3238 |
CaseWhen *newcasewhen = makeNode(CaseWhen); |
| 3239 |
|
| 3240 |
newcasewhen->expr = (Expr *) casecond; |
| 3241 |
newcasewhen->result = (Expr *) caseresult; |
| 3242 |
newcasewhen->location = oldcasewhen->location; |
| 3243 |
newargs = lappend(newargs, newcasewhen); |
| 3244 |
continue; |
| 3245 |
} |
| 3246 |
|
| 3247 |
/* |
| 3248 |
* Found a TRUE condition, so none of the remaining |
| 3249 |
* alternatives can be reached. We treat the result as |
| 3250 |
* the default result. |
| 3251 |
*/ |
| 3252 |
defresult = caseresult; |
| 3253 |
break; |
| 3254 |
} |
| 3255 |
|
| 3256 |
/* Simplify the default result, unless we replaced it above */ |
| 3257 |
if (!const_true_cond) |
| 3258 |
defresult = eval_const_expressions_mutator((Node *) caseexpr->defresult, |
| 3259 |
context); |
| 3260 |
|
| 3261 |
context->case_val = save_case_val; |
| 3262 |
|
| 3263 |
/* |
| 3264 |
* If no non-FALSE alternatives, CASE reduces to the default |
| 3265 |
* result |
| 3266 |
*/ |
| 3267 |
if (newargs == NIL) |
| 3268 |
return defresult; |
| 3269 |
/* Otherwise we need a new CASE node */ |
| 3270 |
newcase = makeNode(CaseExpr); |
| 3271 |
newcase->casetype = caseexpr->casetype; |
| 3272 |
newcase->casecollid = caseexpr->casecollid; |
| 3273 |
newcase->arg = (Expr *) newarg; |
| 3274 |
newcase->args = newargs; |
| 3275 |
newcase->defresult = (Expr *) defresult; |
| 3276 |
newcase->location = caseexpr->location; |
| 3277 |
return (Node *) newcase; |
| 3278 |
} |
| 3279 |
case T_CaseTestExpr: |
| 3280 |
{ |
| 3281 |
/* |
| 3282 |
* If we know a constant test value for the current CASE |
| 3283 |
* construct, substitute it for the placeholder. Else just |
| 3284 |
* return the placeholder as-is. |
| 3285 |
*/ |
| 3286 |
if (context->case_val) |
| 3287 |
return copyObject(context->case_val); |
| 3288 |
else |
| 3289 |
return copyObject(node); |
| 3290 |
} |
| 3291 |
case T_SubscriptingRef: |
| 3292 |
case T_ArrayExpr: |
| 3293 |
case T_RowExpr: |
| 3294 |
case T_MinMaxExpr: |
| 3295 |
{ |
| 3296 |
/* |
| 3297 |
* Generic handling for node types whose own processing is |
| 3298 |
* known to be immutable, and for which we need no smarts |
| 3299 |
* beyond "simplify if all inputs are constants". |
| 3300 |
* |
| 3301 |
* Treating SubscriptingRef this way assumes that subscripting |
| 3302 |
* fetch and assignment are both immutable. This constrains |
| 3303 |
* type-specific subscripting implementations; maybe we should |
| 3304 |
* relax it someday. |
| 3305 |
* |
| 3306 |
* Treating MinMaxExpr this way amounts to assuming that the |
| 3307 |
* btree comparison function it calls is immutable; see the |
| 3308 |
* reasoning in contain_mutable_functions_walker. |
| 3309 |
*/ |
| 3310 |
|
| 3311 |
/* Copy the node and const-simplify its arguments */ |
| 3312 |
node = ece_generic_processing(node); |
| 3313 |
/* If all arguments are Consts, we can fold to a constant */ |
| 3314 |
if (ece_all_arguments_const(node)) |
| 3315 |
return ece_evaluate_expr(node); |
| 3316 |
return node; |
| 3317 |
} |
| 3318 |
case T_CoalesceExpr: |
| 3319 |
{ |
| 3320 |
CoalesceExpr *coalesceexpr = (CoalesceExpr *) node; |
| 3321 |
CoalesceExpr *newcoalesce; |
| 3322 |
List *newargs; |
| 3323 |
ListCell *arg; |
| 3324 |
|
| 3325 |
newargs = NIL; |
| 3326 |
foreach(arg, coalesceexpr->args) |
| 3327 |
{ |
| 3328 |
Node *e; |
| 3329 |
|
| 3330 |
e = eval_const_expressions_mutator((Node *) lfirst(arg), |
| 3331 |
context); |
| 3332 |
|
| 3333 |
/* |
| 3334 |
* We can remove null constants from the list. For a |
| 3335 |
* non-null constant, if it has not been preceded by any |
| 3336 |
* other non-null-constant expressions then it is the |
| 3337 |
* result. Otherwise, it's the next argument, but we can |
| 3338 |
* drop following arguments since they will never be |
| 3339 |
* reached. |
| 3340 |
*/ |
| 3341 |
if (IsA(e, Const)) |
| 3342 |
{ |
| 3343 |
if (((Const *) e)->constisnull) |
| 3344 |
continue; /* drop null constant */ |
| 3345 |
if (newargs == NIL) |
| 3346 |
return e; /* first expr */ |
| 3347 |
newargs = lappend(newargs, e); |
| 3348 |
break; |
| 3349 |
} |
| 3350 |
newargs = lappend(newargs, e); |
| 3351 |
} |
| 3352 |
|
| 3353 |
/* |
| 3354 |
* If all the arguments were constant null, the result is just |
| 3355 |
* null |
| 3356 |
*/ |
| 3357 |
if (newargs == NIL) |
| 3358 |
return (Node *) makeNullConst(coalesceexpr->coalescetype, |
| 3359 |
-1, |
| 3360 |
coalesceexpr->coalescecollid); |
| 3361 |
|
| 3362 |
newcoalesce = makeNode(CoalesceExpr); |
| 3363 |
newcoalesce->coalescetype = coalesceexpr->coalescetype; |
| 3364 |
newcoalesce->coalescecollid = coalesceexpr->coalescecollid; |
| 3365 |
newcoalesce->args = newargs; |
| 3366 |
newcoalesce->location = coalesceexpr->location; |
| 3367 |
return (Node *) newcoalesce; |
| 3368 |
} |
| 3369 |
case T_SQLValueFunction: |
| 3370 |
{ |
| 3371 |
/* |
| 3372 |
* All variants of SQLValueFunction are stable, so if we are |
| 3373 |
* estimating the expression's value, we should evaluate the |
| 3374 |
* current function value. Otherwise just copy. |
| 3375 |
*/ |
| 3376 |
SQLValueFunction *svf = (SQLValueFunction *) node; |
| 3377 |
|
| 3378 |
if (context->estimate) |
| 3379 |
return (Node *) evaluate_expr((Expr *) svf, |
| 3380 |
svf->type, |
| 3381 |
svf->typmod, |
| 3382 |
InvalidOid); |
| 3383 |
else |
| 3384 |
return copyObject((Node *) svf); |
| 3385 |
} |
| 3386 |
case T_FieldSelect: |
| 3387 |
{ |
| 3388 |
/* |
| 3389 |
* We can optimize field selection from a whole-row Var into a |
| 3390 |
* simple Var. (This case won't be generated directly by the |
| 3391 |
* parser, because ParseComplexProjection short-circuits it. |
| 3392 |
* But it can arise while simplifying functions.) Also, we |
| 3393 |
* can optimize field selection from a RowExpr construct, or |
| 3394 |
* of course from a constant. |
| 3395 |
* |
| 3396 |
* However, replacing a whole-row Var in this way has a |
| 3397 |
* pitfall: if we've already built the rel targetlist for the |
| 3398 |
* source relation, then the whole-row Var is scheduled to be |
| 3399 |
* produced by the relation scan, but the simple Var probably |
| 3400 |
* isn't, which will lead to a failure in setrefs.c. This is |
| 3401 |
* not a problem when handling simple single-level queries, in |
| 3402 |
* which expression simplification always happens first. It |
| 3403 |
* is a risk for lateral references from subqueries, though. |
| 3404 |
* To avoid such failures, don't optimize uplevel references. |
| 3405 |
* |
| 3406 |
* We must also check that the declared type of the field is |
| 3407 |
* still the same as when the FieldSelect was created --- this |
| 3408 |
* can change if someone did ALTER COLUMN TYPE on the rowtype. |
| 3409 |
* If it isn't, we skip the optimization; the case will |
| 3410 |
* probably fail at runtime, but that's not our problem here. |
| 3411 |
*/ |
| 3412 |
FieldSelect *fselect = (FieldSelect *) node; |
| 3413 |
FieldSelect *newfselect; |
| 3414 |
Node *arg; |
| 3415 |
|
| 3416 |
arg = eval_const_expressions_mutator((Node *) fselect->arg, |
| 3417 |
context); |
| 3418 |
if (arg && IsA(arg, Var) && |
| 3419 |
((Var *) arg)->varattno == InvalidAttrNumber && |
| 3420 |
((Var *) arg)->varlevelsup == 0) |
| 3421 |
{ |
| 3422 |
if (rowtype_field_matches(((Var *) arg)->vartype, |
| 3423 |
fselect->fieldnum, |
| 3424 |
fselect->resulttype, |
| 3425 |
fselect->resulttypmod, |
| 3426 |
fselect->resultcollid)) |
| 3427 |
{ |
| 3428 |
Var *newvar; |
| 3429 |
|
| 3430 |
newvar = makeVar(((Var *) arg)->varno, |
| 3431 |
fselect->fieldnum, |
| 3432 |
fselect->resulttype, |
| 3433 |
fselect->resulttypmod, |
| 3434 |
fselect->resultcollid, |
| 3435 |
((Var *) arg)->varlevelsup); |
| 3436 |
/* New Var has same OLD/NEW returning as old one */ |
| 3437 |
newvar->varreturningtype = ((Var *) arg)->varreturningtype; |
| 3438 |
/* New Var is nullable by same rels as the old one */ |
| 3439 |
newvar->varnullingrels = ((Var *) arg)->varnullingrels; |
| 3440 |
return (Node *) newvar; |
| 3441 |
} |
| 3442 |
} |
| 3443 |
if (arg && IsA(arg, RowExpr)) |
| 3444 |
{ |
| 3445 |
RowExpr *rowexpr = (RowExpr *) arg; |
| 3446 |
|
| 3447 |
if (fselect->fieldnum > 0 && |
| 3448 |
fselect->fieldnum <= list_length(rowexpr->args)) |
| 3449 |
{ |
| 3450 |
Node *fld = (Node *) list_nth(rowexpr->args, |
| 3451 |
fselect->fieldnum - 1); |
| 3452 |
|
| 3453 |
if (rowtype_field_matches(rowexpr->row_typeid, |
| 3454 |
fselect->fieldnum, |
| 3455 |
fselect->resulttype, |
| 3456 |
fselect->resulttypmod, |
| 3457 |
fselect->resultcollid) && |
| 3458 |
fselect->resulttype == exprType(fld) && |
| 3459 |
fselect->resulttypmod == exprTypmod(fld) && |
| 3460 |
fselect->resultcollid == exprCollation(fld)) |
| 3461 |
return fld; |
| 3462 |
} |
| 3463 |
} |
| 3464 |
newfselect = makeNode(FieldSelect); |
| 3465 |
newfselect->arg = (Expr *) arg; |
| 3466 |
newfselect->fieldnum = fselect->fieldnum; |
| 3467 |
newfselect->resulttype = fselect->resulttype; |
| 3468 |
newfselect->resulttypmod = fselect->resulttypmod; |
| 3469 |
newfselect->resultcollid = fselect->resultcollid; |
| 3470 |
if (arg && IsA(arg, Const)) |
| 3471 |
{ |
| 3472 |
Const *con = (Const *) arg; |
| 3473 |
|
| 3474 |
if (rowtype_field_matches(con->consttype, |
| 3475 |
newfselect->fieldnum, |
| 3476 |
newfselect->resulttype, |
| 3477 |
newfselect->resulttypmod, |
| 3478 |
newfselect->resultcollid)) |
| 3479 |
return ece_evaluate_expr(newfselect); |
| 3480 |
} |
| 3481 |
return (Node *) newfselect; |
| 3482 |
} |
| 3483 |
case T_NullTest: |
| 3484 |
{ |
| 3485 |
NullTest *ntest = (NullTest *) node; |
| 3486 |
NullTest *newntest; |
| 3487 |
Node *arg; |
| 3488 |
|
| 3489 |
arg = eval_const_expressions_mutator((Node *) ntest->arg, |
| 3490 |
context); |
| 3491 |
if (ntest->argisrow && arg && IsA(arg, RowExpr)) |
| 3492 |
{ |
| 3493 |
/* |
| 3494 |
* We break ROW(...) IS [NOT] NULL into separate tests on |
| 3495 |
* its component fields. This form is usually more |
| 3496 |
* efficient to evaluate, as well as being more amenable |
| 3497 |
* to optimization. |
| 3498 |
*/ |
| 3499 |
RowExpr *rarg = (RowExpr *) arg; |
| 3500 |
List *newargs = NIL; |
| 3501 |
ListCell *l; |
| 3502 |
|
| 3503 |
foreach(l, rarg->args) |
| 3504 |
{ |
| 3505 |
Node *relem = (Node *) lfirst(l); |
| 3506 |
|
| 3507 |
/* |
| 3508 |
* A constant field refutes the whole NullTest if it's |
| 3509 |
* of the wrong nullness; else we can discard it. |
| 3510 |
*/ |
| 3511 |
if (relem && IsA(relem, Const)) |
| 3512 |
{ |
| 3513 |
Const *carg = (Const *) relem; |
| 3514 |
|
| 3515 |
if (carg->constisnull ? |
| 3516 |
(ntest->nulltesttype == IS_NOT_NULL) : |
| 3517 |
(ntest->nulltesttype == IS_NULL)) |
| 3518 |
return makeBoolConst(false, false); |
| 3519 |
continue; |
| 3520 |
} |
| 3521 |
|
| 3522 |
/* |
| 3523 |
* Else, make a scalar (argisrow == false) NullTest |
| 3524 |
* for this field. Scalar semantics are required |
| 3525 |
* because IS [NOT] NULL doesn't recurse; see comments |
| 3526 |
* in ExecEvalRowNullInt(). |
| 3527 |
*/ |
| 3528 |
newntest = makeNode(NullTest); |
| 3529 |
newntest->arg = (Expr *) relem; |
| 3530 |
newntest->nulltesttype = ntest->nulltesttype; |
| 3531 |
newntest->argisrow = false; |
| 3532 |
newntest->location = ntest->location; |
| 3533 |
newargs = lappend(newargs, newntest); |
| 3534 |
} |
| 3535 |
/* If all the inputs were constants, result is TRUE */ |
| 3536 |
if (newargs == NIL) |
| 3537 |
return makeBoolConst(true, false); |
| 3538 |
/* If only one nonconst input, it's the result */ |
| 3539 |
if (list_length(newargs) == 1) |
| 3540 |
return (Node *) linitial(newargs); |
| 3541 |
/* Else we need an AND node */ |
| 3542 |
return (Node *) make_andclause(newargs); |
| 3543 |
} |
| 3544 |
if (!ntest->argisrow && arg && IsA(arg, Const)) |
| 3545 |
{ |
| 3546 |
Const *carg = (Const *) arg; |
| 3547 |
bool result; |
| 3548 |
|
| 3549 |
switch (ntest->nulltesttype) |
| 3550 |
{ |
| 3551 |
case IS_NULL: |
| 3552 |
result = carg->constisnull; |
| 3553 |
break; |
| 3554 |
case IS_NOT_NULL: |
| 3555 |
result = !carg->constisnull; |
| 3556 |
break; |
| 3557 |
default: |
| 3558 |
elog(ERROR, "unrecognized nulltesttype: %d", |
| 3559 |
(int) ntest->nulltesttype); |
| 3560 |
result = false; /* keep compiler quiet */ |
| 3561 |
break; |
| 3562 |
} |
| 3563 |
|
| 3564 |
return makeBoolConst(result, false); |
| 3565 |
} |
| 3566 |
|
| 3567 |
newntest = makeNode(NullTest); |
| 3568 |
newntest->arg = (Expr *) arg; |
| 3569 |
newntest->nulltesttype = ntest->nulltesttype; |
| 3570 |
newntest->argisrow = ntest->argisrow; |
| 3571 |
newntest->location = ntest->location; |
| 3572 |
return (Node *) newntest; |
| 3573 |
} |
| 3574 |
case T_BooleanTest: |
| 3575 |
{ |
| 3576 |
/* |
| 3577 |
* This case could be folded into the generic handling used |
| 3578 |
* for ArrayExpr etc. But because the simplification logic is |
| 3579 |
* so trivial, applying evaluate_expr() to perform it would be |
| 3580 |
* a heavy overhead. BooleanTest is probably common enough to |
| 3581 |
* justify keeping this bespoke implementation. |
| 3582 |
*/ |
| 3583 |
BooleanTest *btest = (BooleanTest *) node; |
| 3584 |
BooleanTest *newbtest; |
| 3585 |
Node *arg; |
| 3586 |
|
| 3587 |
arg = eval_const_expressions_mutator((Node *) btest->arg, |
| 3588 |
context); |
| 3589 |
if (arg && IsA(arg, Const)) |
| 3590 |
{ |
| 3591 |
Const *carg = (Const *) arg; |
| 3592 |
bool result; |
| 3593 |
|
| 3594 |
switch (btest->booltesttype) |
| 3595 |
{ |
| 3596 |
case IS_TRUE: |
| 3597 |
result = (!carg->constisnull && |
| 3598 |
DatumGetBool(carg->constvalue)); |
| 3599 |
break; |
| 3600 |
case IS_NOT_TRUE: |
| 3601 |
result = (carg->constisnull || |
| 3602 |
!DatumGetBool(carg->constvalue)); |
| 3603 |
break; |
| 3604 |
case IS_FALSE: |
| 3605 |
result = (!carg->constisnull && |
| 3606 |
!DatumGetBool(carg->constvalue)); |
| 3607 |
break; |
| 3608 |
case IS_NOT_FALSE: |
| 3609 |
result = (carg->constisnull || |
| 3610 |
DatumGetBool(carg->constvalue)); |
| 3611 |
break; |
| 3612 |
case IS_UNKNOWN: |
| 3613 |
result = carg->constisnull; |
| 3614 |
break; |
| 3615 |
case IS_NOT_UNKNOWN: |
| 3616 |
result = !carg->constisnull; |
| 3617 |
break; |
| 3618 |
default: |
| 3619 |
elog(ERROR, "unrecognized booltesttype: %d", |
| 3620 |
(int) btest->booltesttype); |
| 3621 |
result = false; /* keep compiler quiet */ |
| 3622 |
break; |
| 3623 |
} |
| 3624 |
|
| 3625 |
return makeBoolConst(result, false); |
| 3626 |
} |
| 3627 |
|
| 3628 |
newbtest = makeNode(BooleanTest); |
| 3629 |
newbtest->arg = (Expr *) arg; |
| 3630 |
newbtest->booltesttype = btest->booltesttype; |
| 3631 |
newbtest->location = btest->location; |
| 3632 |
return (Node *) newbtest; |
| 3633 |
} |
| 3634 |
case T_CoerceToDomain: |
| 3635 |
{ |
| 3636 |
/* |
| 3637 |
* If the domain currently has no constraints, we replace the |
| 3638 |
* CoerceToDomain node with a simple RelabelType, which is |
| 3639 |
* both far faster to execute and more amenable to later |
| 3640 |
* optimization. We must then mark the plan as needing to be |
| 3641 |
* rebuilt if the domain's constraints change. |
| 3642 |
* |
| 3643 |
* Also, in estimation mode, always replace CoerceToDomain |
| 3644 |
* nodes, effectively assuming that the coercion will succeed. |
| 3645 |
*/ |
| 3646 |
CoerceToDomain *cdomain = (CoerceToDomain *) node; |
| 3647 |
CoerceToDomain *newcdomain; |
| 3648 |
Node *arg; |
| 3649 |
|
| 3650 |
arg = eval_const_expressions_mutator((Node *) cdomain->arg, |
| 3651 |
context); |
| 3652 |
if (context->estimate || |
| 3653 |
!DomainHasConstraints(cdomain->resulttype)) |
| 3654 |
{ |
| 3655 |
/* Record dependency, if this isn't estimation mode */ |
| 3656 |
if (context->root && !context->estimate) |
| 3657 |
record_plan_type_dependency(context->root, |
| 3658 |
cdomain->resulttype); |
| 3659 |
|
| 3660 |
/* Generate RelabelType to substitute for CoerceToDomain */ |
| 3661 |
return applyRelabelType(arg, |
| 3662 |
cdomain->resulttype, |
| 3663 |
cdomain->resulttypmod, |
| 3664 |
cdomain->resultcollid, |
| 3665 |
cdomain->coercionformat, |
| 3666 |
cdomain->location, |
| 3667 |
true); |
| 3668 |
} |
| 3669 |
|
| 3670 |
newcdomain = makeNode(CoerceToDomain); |
| 3671 |
newcdomain->arg = (Expr *) arg; |
| 3672 |
newcdomain->resulttype = cdomain->resulttype; |
| 3673 |
newcdomain->resulttypmod = cdomain->resulttypmod; |
| 3674 |
newcdomain->resultcollid = cdomain->resultcollid; |
| 3675 |
newcdomain->coercionformat = cdomain->coercionformat; |
| 3676 |
newcdomain->location = cdomain->location; |
| 3677 |
return (Node *) newcdomain; |
| 3678 |
} |
| 3679 |
case T_PlaceHolderVar: |
| 3680 |
|
| 3681 |
/* |
| 3682 |
* In estimation mode, just strip the PlaceHolderVar node |
| 3683 |
* altogether; this amounts to estimating that the contained value |
| 3684 |
* won't be forced to null by an outer join. In regular mode we |
| 3685 |
* just use the default behavior (ie, simplify the expression but |
| 3686 |
* leave the PlaceHolderVar node intact). |
| 3687 |
*/ |
| 3688 |
if (context->estimate) |
| 3689 |
{ |
| 3690 |
PlaceHolderVar *phv = (PlaceHolderVar *) node; |
| 3691 |
|
| 3692 |
return eval_const_expressions_mutator((Node *) phv->phexpr, |
| 3693 |
context); |
| 3694 |
} |
| 3695 |
break; |
| 3696 |
case T_ConvertRowtypeExpr: |
| 3697 |
{ |
| 3698 |
ConvertRowtypeExpr *cre = castNode(ConvertRowtypeExpr, node); |
| 3699 |
Node *arg; |
| 3700 |
ConvertRowtypeExpr *newcre; |
| 3701 |
|
| 3702 |
arg = eval_const_expressions_mutator((Node *) cre->arg, |
| 3703 |
context); |
| 3704 |
|
| 3705 |
newcre = makeNode(ConvertRowtypeExpr); |
| 3706 |
newcre->resulttype = cre->resulttype; |
| 3707 |
newcre->convertformat = cre->convertformat; |
| 3708 |
newcre->location = cre->location; |
| 3709 |
|
| 3710 |
/* |
| 3711 |
* In case of a nested ConvertRowtypeExpr, we can convert the |
| 3712 |
* leaf row directly to the topmost row format without any |
| 3713 |
* intermediate conversions. (This works because |
| 3714 |
* ConvertRowtypeExpr is used only for child->parent |
| 3715 |
* conversion in inheritance trees, which works by exact match |
| 3716 |
* of column name, and a column absent in an intermediate |
| 3717 |
* result can't be present in the final result.) |
| 3718 |
* |
| 3719 |
* No need to check more than one level deep, because the |
| 3720 |
* above recursion will have flattened anything else. |
| 3721 |
*/ |
| 3722 |
if (arg != NULL && IsA(arg, ConvertRowtypeExpr)) |
| 3723 |
{ |
| 3724 |
ConvertRowtypeExpr *argcre = (ConvertRowtypeExpr *) arg; |
| 3725 |
|
| 3726 |
arg = (Node *) argcre->arg; |
| 3727 |
|
| 3728 |
/* |
| 3729 |
* Make sure an outer implicit conversion can't hide an |
| 3730 |
* inner explicit one. |
| 3731 |
*/ |
| 3732 |
if (newcre->convertformat == COERCE_IMPLICIT_CAST) |
| 3733 |
newcre->convertformat = argcre->convertformat; |
| 3734 |
} |
| 3735 |
|
| 3736 |
newcre->arg = (Expr *) arg; |
| 3737 |
|
| 3738 |
if (arg != NULL && IsA(arg, Const)) |
| 3739 |
return ece_evaluate_expr((Node *) newcre); |
| 3740 |
return (Node *) newcre; |
| 3741 |
} |
| 3742 |
default: |
| 3743 |
break; |
| 3744 |
} |
| 3745 |
|
| 3746 |
/* |
| 3747 |
* For any node type not handled above, copy the node unchanged but |
| 3748 |
* const-simplify its subexpressions. This is the correct thing for node |
| 3749 |
* types whose behavior might change between planning and execution, such |
| 3750 |
* as CurrentOfExpr. It's also a safe default for new node types not |
| 3751 |
* known to this routine. |
| 3752 |
*/ |
| 3753 |
return ece_generic_processing(node); |
| 3754 |
} |
| 3755 |
|
| 3756 |
/* |
| 3757 |
* Subroutine for eval_const_expressions: check for non-Const nodes. |
| 3758 |
* |
| 3759 |
* We can abort recursion immediately on finding a non-Const node. This is |
| 3760 |
* critical for performance, else eval_const_expressions_mutator would take |
| 3761 |
* O(N^2) time on non-simplifiable trees. However, we do need to descend |
| 3762 |
* into List nodes since expression_tree_walker sometimes invokes the walker |
| 3763 |
* function directly on List subtrees. |
| 3764 |
*/ |
| 3765 |
static bool |
| 3766 |
contain_non_const_walker(Node *node, void *context) |
| 3767 |
{ |
| 3768 |
if (node == NULL) |
| 3769 |
return false; |
| 3770 |
if (IsA(node, Const)) |
| 3771 |
return false; |
| 3772 |
if (IsA(node, List)) |
| 3773 |
return expression_tree_walker(node, contain_non_const_walker, context); |
| 3774 |
/* Otherwise, abort the tree traversal and return true */ |
| 3775 |
return true; |
| 3776 |
} |
| 3777 |
|
| 3778 |
/* |
| 3779 |
* Subroutine for eval_const_expressions: check if a function is OK to evaluate |
| 3780 |
*/ |
| 3781 |
static bool |
| 3782 |
ece_function_is_safe(Oid funcid, eval_const_expressions_context *context) |
| 3783 |
{ |
| 3784 |
char provolatile = func_volatile(funcid); |
| 3785 |
|
| 3786 |
/* |
| 3787 |
* Ordinarily we are only allowed to simplify immutable functions. But for |
| 3788 |
* purposes of estimation, we consider it okay to simplify functions that |
| 3789 |
* are merely stable; the risk that the result might change from planning |
| 3790 |
* time to execution time is worth taking in preference to not being able |
| 3791 |
* to estimate the value at all. |
| 3792 |
*/ |
| 3793 |
if (provolatile == PROVOLATILE_IMMUTABLE) |
| 3794 |
return true; |
| 3795 |
if (context->estimate && provolatile == PROVOLATILE_STABLE) |
| 3796 |
return true; |
| 3797 |
return false; |
| 3798 |
} |
| 3799 |
|
| 3800 |
/* |
| 3801 |
* Subroutine for eval_const_expressions: process arguments of an OR clause |
| 3802 |
* |
| 3803 |
* This includes flattening of nested ORs as well as recursion to |
| 3804 |
* eval_const_expressions to simplify the OR arguments. |
| 3805 |
* |
| 3806 |
* After simplification, OR arguments are handled as follows: |
| 3807 |
* non constant: keep |
| 3808 |
* FALSE: drop (does not affect result) |
| 3809 |
* TRUE: force result to TRUE |
| 3810 |
* NULL: keep only one |
| 3811 |
* We must keep one NULL input because OR expressions evaluate to NULL when no |
| 3812 |
* input is TRUE and at least one is NULL. We don't actually include the NULL |
| 3813 |
* here, that's supposed to be done by the caller. |
| 3814 |
* |
| 3815 |
* The output arguments *haveNull and *forceTrue must be initialized false |
| 3816 |
* by the caller. They will be set true if a NULL constant or TRUE constant, |
| 3817 |
* respectively, is detected anywhere in the argument list. |
| 3818 |
*/ |
| 3819 |
static List * |
| 3820 |
simplify_or_arguments(List *args, |
| 3821 |
eval_const_expressions_context *context, |
| 3822 |
bool *haveNull, bool *forceTrue) |
| 3823 |
{ |
| 3824 |
List *newargs = NIL; |
| 3825 |
List *unprocessed_args; |
| 3826 |
|
| 3827 |
/* |
| 3828 |
* We want to ensure that any OR immediately beneath another OR gets |
| 3829 |
* flattened into a single OR-list, so as to simplify later reasoning. |
| 3830 |
* |
| 3831 |
* To avoid stack overflow from recursion of eval_const_expressions, we |
| 3832 |
* resort to some tenseness here: we keep a list of not-yet-processed |
| 3833 |
* inputs, and handle flattening of nested ORs by prepending to the to-do |
| 3834 |
* list instead of recursing. Now that the parser generates N-argument |
| 3835 |
* ORs from simple lists, this complexity is probably less necessary than |
| 3836 |
* it once was, but we might as well keep the logic. |
| 3837 |
*/ |
| 3838 |
unprocessed_args = list_copy(args); |
| 3839 |
while (unprocessed_args) |
| 3840 |
{ |
| 3841 |
Node *arg = (Node *) linitial(unprocessed_args); |
| 3842 |
|
| 3843 |
unprocessed_args = list_delete_first(unprocessed_args); |
| 3844 |
|
| 3845 |
/* flatten nested ORs as per above comment */ |
| 3846 |
if (is_orclause(arg)) |
| 3847 |
{ |
| 3848 |
List *subargs = ((BoolExpr *) arg)->args; |
| 3849 |
List *oldlist = unprocessed_args; |
| 3850 |
|
| 3851 |
unprocessed_args = list_concat_copy(subargs, unprocessed_args); |
| 3852 |
/* perhaps-overly-tense code to avoid leaking old lists */ |
| 3853 |
list_free(oldlist); |
| 3854 |
continue; |
| 3855 |
} |
| 3856 |
|
| 3857 |
/* If it's not an OR, simplify it */ |
| 3858 |
arg = eval_const_expressions_mutator(arg, context); |
| 3859 |
|
| 3860 |
/* |
| 3861 |
* It is unlikely but not impossible for simplification of a non-OR |
| 3862 |
* clause to produce an OR. Recheck, but don't be too tense about it |
| 3863 |
* since it's not a mainstream case. In particular we don't worry |
| 3864 |
* about const-simplifying the input twice, nor about list leakage. |
| 3865 |
*/ |
| 3866 |
if (is_orclause(arg)) |
| 3867 |
{ |
| 3868 |
List *subargs = ((BoolExpr *) arg)->args; |
| 3869 |
|
| 3870 |
unprocessed_args = list_concat_copy(subargs, unprocessed_args); |
| 3871 |
continue; |
| 3872 |
} |
| 3873 |
|
| 3874 |
/* |
| 3875 |
* OK, we have a const-simplified non-OR argument. Process it per |
| 3876 |
* comments above. |
| 3877 |
*/ |
| 3878 |
if (IsA(arg, Const)) |
| 3879 |
{ |
| 3880 |
Const *const_input = (Const *) arg; |
| 3881 |
|
| 3882 |
if (const_input->constisnull) |
| 3883 |
*haveNull = true; |
| 3884 |
else if (DatumGetBool(const_input->constvalue)) |
| 3885 |
{ |
| 3886 |
*forceTrue = true; |
| 3887 |
|
| 3888 |
/* |
| 3889 |
* Once we detect a TRUE result we can just exit the loop |
| 3890 |
* immediately. However, if we ever add a notion of |
| 3891 |
* non-removable functions, we'd need to keep scanning. |
| 3892 |
*/ |
| 3893 |
return NIL; |
| 3894 |
} |
| 3895 |
/* otherwise, we can drop the constant-false input */ |
| 3896 |
continue; |
| 3897 |
} |
| 3898 |
|
| 3899 |
/* else emit the simplified arg into the result list */ |
| 3900 |
newargs = lappend(newargs, arg); |
| 3901 |
} |
| 3902 |
|
| 3903 |
return newargs; |
| 3904 |
} |
| 3905 |
|
| 3906 |
/* |
| 3907 |
* Subroutine for eval_const_expressions: process arguments of an AND clause |
| 3908 |
* |
| 3909 |
* This includes flattening of nested ANDs as well as recursion to |
| 3910 |
* eval_const_expressions to simplify the AND arguments. |
| 3911 |
* |
| 3912 |
* After simplification, AND arguments are handled as follows: |
| 3913 |
* non constant: keep |
| 3914 |
* TRUE: drop (does not affect result) |
| 3915 |
* FALSE: force result to FALSE |
| 3916 |
* NULL: keep only one |
| 3917 |
* We must keep one NULL input because AND expressions evaluate to NULL when |
| 3918 |
* no input is FALSE and at least one is NULL. We don't actually include the |
| 3919 |
* NULL here, that's supposed to be done by the caller. |
| 3920 |
* |
| 3921 |
* The output arguments *haveNull and *forceFalse must be initialized false |
| 3922 |
* by the caller. They will be set true if a null constant or false constant, |
| 3923 |
* respectively, is detected anywhere in the argument list. |
| 3924 |
*/ |
| 3925 |
static List * |
| 3926 |
simplify_and_arguments(List *args, |
| 3927 |
eval_const_expressions_context *context, |
| 3928 |
bool *haveNull, bool *forceFalse) |
| 3929 |
{ |
| 3930 |
List *newargs = NIL; |
| 3931 |
List *unprocessed_args; |
| 3932 |
|
| 3933 |
/* See comments in simplify_or_arguments */ |
| 3934 |
unprocessed_args = list_copy(args); |
| 3935 |
while (unprocessed_args) |
| 3936 |
{ |
| 3937 |
Node *arg = (Node *) linitial(unprocessed_args); |
| 3938 |
|
| 3939 |
unprocessed_args = list_delete_first(unprocessed_args); |
| 3940 |
|
| 3941 |
/* flatten nested ANDs as per above comment */ |
| 3942 |
if (is_andclause(arg)) |
| 3943 |
{ |
| 3944 |
List *subargs = ((BoolExpr *) arg)->args; |
| 3945 |
List *oldlist = unprocessed_args; |
| 3946 |
|
| 3947 |
unprocessed_args = list_concat_copy(subargs, unprocessed_args); |
| 3948 |
/* perhaps-overly-tense code to avoid leaking old lists */ |
| 3949 |
list_free(oldlist); |
| 3950 |
continue; |
| 3951 |
} |
| 3952 |
|
| 3953 |
/* If it's not an AND, simplify it */ |
| 3954 |
arg = eval_const_expressions_mutator(arg, context); |
| 3955 |
|
| 3956 |
/* |
| 3957 |
* It is unlikely but not impossible for simplification of a non-AND |
| 3958 |
* clause to produce an AND. Recheck, but don't be too tense about it |
| 3959 |
* since it's not a mainstream case. In particular we don't worry |
| 3960 |
* about const-simplifying the input twice, nor about list leakage. |
| 3961 |
*/ |
| 3962 |
if (is_andclause(arg)) |
| 3963 |
{ |
| 3964 |
List *subargs = ((BoolExpr *) arg)->args; |
| 3965 |
|
| 3966 |
unprocessed_args = list_concat_copy(subargs, unprocessed_args); |
| 3967 |
continue; |
| 3968 |
} |
| 3969 |
|
| 3970 |
/* |
| 3971 |
* OK, we have a const-simplified non-AND argument. Process it per |
| 3972 |
* comments above. |
| 3973 |
*/ |
| 3974 |
if (IsA(arg, Const)) |
| 3975 |
{ |
| 3976 |
Const *const_input = (Const *) arg; |
| 3977 |
|
| 3978 |
if (const_input->constisnull) |
| 3979 |
*haveNull = true; |
| 3980 |
else if (!DatumGetBool(const_input->constvalue)) |
| 3981 |
{ |
| 3982 |
*forceFalse = true; |
| 3983 |
|
| 3984 |
/* |
| 3985 |
* Once we detect a FALSE result we can just exit the loop |
| 3986 |
* immediately. However, if we ever add a notion of |
| 3987 |
* non-removable functions, we'd need to keep scanning. |
| 3988 |
*/ |
| 3989 |
return NIL; |
| 3990 |
} |
| 3991 |
/* otherwise, we can drop the constant-true input */ |
| 3992 |
continue; |
| 3993 |
} |
| 3994 |
|
| 3995 |
/* else emit the simplified arg into the result list */ |
| 3996 |
newargs = lappend(newargs, arg); |
| 3997 |
} |
| 3998 |
|
| 3999 |
return newargs; |
| 4000 |
} |
| 4001 |
|
| 4002 |
/* |
| 4003 |
* Subroutine for eval_const_expressions: try to simplify boolean equality |
| 4004 |
* or inequality condition |
| 4005 |
* |
| 4006 |
* Inputs are the operator OID and the simplified arguments to the operator. |
| 4007 |
* Returns a simplified expression if successful, or NULL if cannot |
| 4008 |
* simplify the expression. |
| 4009 |
* |
| 4010 |
* The idea here is to reduce "x = true" to "x" and "x = false" to "NOT x", |
| 4011 |
* or similarly "x <> true" to "NOT x" and "x <> false" to "x". |
| 4012 |
* This is only marginally useful in itself, but doing it in constant folding |
| 4013 |
* ensures that we will recognize these forms as being equivalent in, for |
| 4014 |
* example, partial index matching. |
| 4015 |
* |
| 4016 |
* We come here only if simplify_function has failed; therefore we cannot |
| 4017 |
* see two constant inputs, nor a constant-NULL input. |
| 4018 |
*/ |
| 4019 |
static Node * |
| 4020 |
simplify_boolean_equality(Oid opno, List *args) |
| 4021 |
{ |
| 4022 |
Node *leftop; |
| 4023 |
Node *rightop; |
| 4024 |
|
| 4025 |
Assert(list_length(args) == 2); |
| 4026 |
leftop = linitial(args); |
| 4027 |
rightop = lsecond(args); |
| 4028 |
if (leftop && IsA(leftop, Const)) |
| 4029 |
{ |
| 4030 |
Assert(!((Const *) leftop)->constisnull); |
| 4031 |
if (opno == BooleanEqualOperator) |
| 4032 |
{ |
| 4033 |
if (DatumGetBool(((Const *) leftop)->constvalue)) |
| 4034 |
return rightop; /* true = foo */ |
| 4035 |
else |
| 4036 |
return negate_clause(rightop); /* false = foo */ |
| 4037 |
} |
| 4038 |
else |
| 4039 |
{ |
| 4040 |
if (DatumGetBool(((Const *) leftop)->constvalue)) |
| 4041 |
return negate_clause(rightop); /* true <> foo */ |
| 4042 |
else |
| 4043 |
return rightop; /* false <> foo */ |
| 4044 |
} |
| 4045 |
} |
| 4046 |
if (rightop && IsA(rightop, Const)) |
| 4047 |
{ |
| 4048 |
Assert(!((Const *) rightop)->constisnull); |
| 4049 |
if (opno == BooleanEqualOperator) |
| 4050 |
{ |
| 4051 |
if (DatumGetBool(((Const *) rightop)->constvalue)) |
| 4052 |
return leftop; /* foo = true */ |
| 4053 |
else |
| 4054 |
return negate_clause(leftop); /* foo = false */ |
| 4055 |
} |
| 4056 |
else |
| 4057 |
{ |
| 4058 |
if (DatumGetBool(((Const *) rightop)->constvalue)) |
| 4059 |
return negate_clause(leftop); /* foo <> true */ |
| 4060 |
else |
| 4061 |
return leftop; /* foo <> false */ |
| 4062 |
} |
| 4063 |
} |
| 4064 |
return NULL; |
| 4065 |
} |
| 4066 |
|
| 4067 |
/* |
| 4068 |
* Subroutine for eval_const_expressions: try to simplify a function call |
| 4069 |
* (which might originally have been an operator; we don't care) |
| 4070 |
* |
| 4071 |
* Inputs are the function OID, actual result type OID (which is needed for |
| 4072 |
* polymorphic functions), result typmod, result collation, the input |
| 4073 |
* collation to use for the function, the original argument list (not |
| 4074 |
* const-simplified yet, unless process_args is false), and some flags; |
| 4075 |
* also the context data for eval_const_expressions. |
| 4076 |
* |
| 4077 |
* Returns a simplified expression if successful, or NULL if cannot |
| 4078 |
* simplify the function call. |
| 4079 |
* |
| 4080 |
* This function is also responsible for converting named-notation argument |
| 4081 |
* lists into positional notation and/or adding any needed default argument |
| 4082 |
* expressions; which is a bit grotty, but it avoids extra fetches of the |
| 4083 |
* function's pg_proc tuple. For this reason, the args list is |
| 4084 |
* pass-by-reference. Conversion and const-simplification of the args list |
| 4085 |
* will be done even if simplification of the function call itself is not |
| 4086 |
* possible. |
| 4087 |
*/ |
| 4088 |
static Expr * |
| 4089 |
simplify_function(Oid funcid, Oid result_type, int32 result_typmod, |
| 4090 |
Oid result_collid, Oid input_collid, List **args_p, |
| 4091 |
bool funcvariadic, bool process_args, bool allow_non_const, |
| 4092 |
eval_const_expressions_context *context) |
| 4093 |
{ |
| 4094 |
List *args = *args_p; |
| 4095 |
HeapTuple func_tuple; |
| 4096 |
Form_pg_proc func_form; |
| 4097 |
Expr *newexpr; |
| 4098 |
|
| 4099 |
/* |
| 4100 |
* We have three strategies for simplification: execute the function to |
| 4101 |
* deliver a constant result, use a transform function to generate a |
| 4102 |
* substitute node tree, or expand in-line the body of the function |
| 4103 |
* definition (which only works for simple SQL-language functions, but |
| 4104 |
* that is a common case). Each case needs access to the function's |
| 4105 |
* pg_proc tuple, so fetch it just once. |
| 4106 |
* |
| 4107 |
* Note: the allow_non_const flag suppresses both the second and third |
| 4108 |
* strategies; so if !allow_non_const, simplify_function can only return a |
| 4109 |
* Const or NULL. Argument-list rewriting happens anyway, though. |
| 4110 |
*/ |
| 4111 |
func_tuple = SearchSysCache1(PROCOID, ObjectIdGetDatum(funcid)); |
| 4112 |
if (!HeapTupleIsValid(func_tuple)) |
| 4113 |
elog(ERROR, "cache lookup failed for function %u", funcid); |
| 4114 |
func_form = (Form_pg_proc) GETSTRUCT(func_tuple); |
| 4115 |
|
| 4116 |
/* |
| 4117 |
* Process the function arguments, unless the caller did it already. |
| 4118 |
* |
| 4119 |
* Here we must deal with named or defaulted arguments, and then |
| 4120 |
* recursively apply eval_const_expressions to the whole argument list. |
| 4121 |
*/ |
| 4122 |
if (process_args) |
| 4123 |
{ |
| 4124 |
args = expand_function_arguments(args, false, result_type, func_tuple); |
| 4125 |
args = (List *) expression_tree_mutator((Node *) args, |
| 4126 |
eval_const_expressions_mutator, |
| 4127 |
context); |
| 4128 |
/* Argument processing done, give it back to the caller */ |
| 4129 |
*args_p = args; |
| 4130 |
} |
| 4131 |
|
| 4132 |
/* Now attempt simplification of the function call proper. */ |
| 4133 |
|
| 4134 |
newexpr = evaluate_function(funcid, result_type, result_typmod, |
| 4135 |
result_collid, input_collid, |
| 4136 |
args, funcvariadic, |
| 4137 |
func_tuple, context); |
| 4138 |
|
| 4139 |
if (!newexpr && allow_non_const && OidIsValid(func_form->prosupport)) |
| 4140 |
{ |
| 4141 |
/* |
| 4142 |
* Build a SupportRequestSimplify node to pass to the support |
| 4143 |
* function, pointing to a dummy FuncExpr node containing the |
| 4144 |
* simplified arg list. We use this approach to present a uniform |
| 4145 |
* interface to the support function regardless of how the target |
| 4146 |
* function is actually being invoked. |
| 4147 |
*/ |
| 4148 |
SupportRequestSimplify req; |
| 4149 |
FuncExpr fexpr; |
| 4150 |
|
| 4151 |
fexpr.xpr.type = T_FuncExpr; |
| 4152 |
fexpr.funcid = funcid; |
| 4153 |
fexpr.funcresulttype = result_type; |
| 4154 |
fexpr.funcretset = func_form->proretset; |
| 4155 |
fexpr.funcvariadic = funcvariadic; |
| 4156 |
fexpr.funcformat = COERCE_EXPLICIT_CALL; |
| 4157 |
fexpr.funccollid = result_collid; |
| 4158 |
fexpr.inputcollid = input_collid; |
| 4159 |
fexpr.args = args; |
| 4160 |
fexpr.location = -1; |
| 4161 |
|
| 4162 |
req.type = T_SupportRequestSimplify; |
| 4163 |
req.root = context->root; |
| 4164 |
req.fcall = &fexpr; |
| 4165 |
|
| 4166 |
newexpr = (Expr *) |
| 4167 |
DatumGetPointer(OidFunctionCall1(func_form->prosupport, |
| 4168 |
PointerGetDatum(&req))); |
| 4169 |
|
| 4170 |
/* catch a possible API misunderstanding */ |
| 4171 |
Assert(newexpr != (Expr *) &fexpr); |
| 4172 |
} |
| 4173 |
|
| 4174 |
if (!newexpr && allow_non_const) |
| 4175 |
newexpr = inline_function(funcid, result_type, result_collid, |
| 4176 |
input_collid, args, funcvariadic, |
| 4177 |
func_tuple, context); |
| 4178 |
|
| 4179 |
ReleaseSysCache(func_tuple); |
| 4180 |
|
| 4181 |
return newexpr; |
| 4182 |
} |
| 4183 |
|
| 4184 |
/* |
| 4185 |
* expand_function_arguments: convert named-notation args to positional args |
| 4186 |
* and/or insert default args, as needed |
| 4187 |
* |
| 4188 |
* Returns a possibly-transformed version of the args list. |
| 4189 |
* |
| 4190 |
* If include_out_arguments is true, then the args list and the result |
| 4191 |
* include OUT arguments. |
| 4192 |
* |
| 4193 |
* The expected result type of the call must be given, for sanity-checking |
| 4194 |
* purposes. Also, we ask the caller to provide the function's actual |
| 4195 |
* pg_proc tuple, not just its OID. |
| 4196 |
* |
| 4197 |
* If we need to change anything, the input argument list is copied, not |
| 4198 |
* modified. |
| 4199 |
* |
| 4200 |
* Note: this gets applied to operator argument lists too, even though the |
| 4201 |
* cases it handles should never occur there. This should be OK since it |
| 4202 |
* will fall through very quickly if there's nothing to do. |
| 4203 |
*/ |
| 4204 |
List * |
| 4205 |
expand_function_arguments(List *args, bool include_out_arguments, |
| 4206 |
Oid result_type, HeapTuple func_tuple) |
| 4207 |
{ |
| 4208 |
Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple); |
| 4209 |
Oid *proargtypes = funcform->proargtypes.values; |
| 4210 |
int pronargs = funcform->pronargs; |
| 4211 |
bool has_named_args = false; |
| 4212 |
ListCell *lc; |
| 4213 |
|
| 4214 |
/* |
| 4215 |
* If we are asked to match to OUT arguments, then use the proallargtypes |
| 4216 |
* array (which includes those); otherwise use proargtypes (which |
| 4217 |
* doesn't). Of course, if proallargtypes is null, we always use |
| 4218 |
* proargtypes. (Fetching proallargtypes is annoyingly expensive |
| 4219 |
* considering that we may have nothing to do here, but fortunately the |
| 4220 |
* common case is include_out_arguments == false.) |
| 4221 |
*/ |
| 4222 |
if (include_out_arguments) |
| 4223 |
{ |
| 4224 |
Datum proallargtypes; |
| 4225 |
bool isNull; |
| 4226 |
|
| 4227 |
proallargtypes = SysCacheGetAttr(PROCOID, func_tuple, |
| 4228 |
Anum_pg_proc_proallargtypes, |
| 4229 |
&isNull); |
| 4230 |
if (!isNull) |
| 4231 |
{ |
| 4232 |
ArrayType *arr = DatumGetArrayTypeP(proallargtypes); |
| 4233 |
|
| 4234 |
pronargs = ARR_DIMS(arr)[0]; |
| 4235 |
if (ARR_NDIM(arr) != 1 || |
| 4236 |
pronargs < 0 || |
| 4237 |
ARR_HASNULL(arr) || |
| 4238 |
ARR_ELEMTYPE(arr) != OIDOID) |
| 4239 |
elog(ERROR, "proallargtypes is not a 1-D Oid array or it contains nulls"); |
| 4240 |
Assert(pronargs >= funcform->pronargs); |
| 4241 |
proargtypes = (Oid *) ARR_DATA_PTR(arr); |
| 4242 |
} |
| 4243 |
} |
| 4244 |
|
| 4245 |
/* Do we have any named arguments? */ |
| 4246 |
foreach(lc, args) |
| 4247 |
{ |
| 4248 |
Node *arg = (Node *) lfirst(lc); |
| 4249 |
|
| 4250 |
if (IsA(arg, NamedArgExpr)) |
| 4251 |
{ |
| 4252 |
has_named_args = true; |
| 4253 |
break; |
| 4254 |
} |
| 4255 |
} |
| 4256 |
|
| 4257 |
/* If so, we must apply reorder_function_arguments */ |
| 4258 |
if (has_named_args) |
| 4259 |
{ |
| 4260 |
args = reorder_function_arguments(args, pronargs, func_tuple); |
| 4261 |
/* Recheck argument types and add casts if needed */ |
| 4262 |
recheck_cast_function_args(args, result_type, |
| 4263 |
proargtypes, pronargs, |
| 4264 |
func_tuple); |
| 4265 |
} |
| 4266 |
else if (list_length(args) < pronargs) |
| 4267 |
{ |
| 4268 |
/* No named args, but we seem to be short some defaults */ |
| 4269 |
args = add_function_defaults(args, pronargs, func_tuple); |
| 4270 |
/* Recheck argument types and add casts if needed */ |
| 4271 |
recheck_cast_function_args(args, result_type, |
| 4272 |
proargtypes, pronargs, |
| 4273 |
func_tuple); |
| 4274 |
} |
| 4275 |
|
| 4276 |
return args; |
| 4277 |
} |
| 4278 |
|
| 4279 |
/* |
| 4280 |
* reorder_function_arguments: convert named-notation args to positional args |
| 4281 |
* |
| 4282 |
* This function also inserts default argument values as needed, since it's |
| 4283 |
* impossible to form a truly valid positional call without that. |
| 4284 |
*/ |
| 4285 |
static List * |
| 4286 |
reorder_function_arguments(List *args, int pronargs, HeapTuple func_tuple) |
| 4287 |
{ |
| 4288 |
Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple); |
| 4289 |
int nargsprovided = list_length(args); |
| 4290 |
Node *argarray[FUNC_MAX_ARGS]; |
| 4291 |
ListCell *lc; |
| 4292 |
int i; |
| 4293 |
|
| 4294 |
Assert(nargsprovided <= pronargs); |
| 4295 |
if (pronargs < 0 || pronargs > FUNC_MAX_ARGS) |
| 4296 |
elog(ERROR, "too many function arguments"); |
| 4297 |
memset(argarray, 0, pronargs * sizeof(Node *)); |
| 4298 |
|
| 4299 |
/* Deconstruct the argument list into an array indexed by argnumber */ |
| 4300 |
i = 0; |
| 4301 |
foreach(lc, args) |
| 4302 |
{ |
| 4303 |
Node *arg = (Node *) lfirst(lc); |
| 4304 |
|
| 4305 |
if (!IsA(arg, NamedArgExpr)) |
| 4306 |
{ |
| 4307 |
/* positional argument, assumed to precede all named args */ |
| 4308 |
Assert(argarray[i] == NULL); |
| 4309 |
argarray[i++] = arg; |
| 4310 |
} |
| 4311 |
else |
| 4312 |
{ |
| 4313 |
NamedArgExpr *na = (NamedArgExpr *) arg; |
| 4314 |
|
| 4315 |
Assert(na->argnumber >= 0 && na->argnumber < pronargs); |
| 4316 |
Assert(argarray[na->argnumber] == NULL); |
| 4317 |
argarray[na->argnumber] = (Node *) na->arg; |
| 4318 |
} |
| 4319 |
} |
| 4320 |
|
| 4321 |
/* |
| 4322 |
* Fetch default expressions, if needed, and insert into array at proper |
| 4323 |
* locations (they aren't necessarily consecutive or all used) |
| 4324 |
*/ |
| 4325 |
if (nargsprovided < pronargs) |
| 4326 |
{ |
| 4327 |
List *defaults = fetch_function_defaults(func_tuple); |
| 4328 |
|
| 4329 |
i = pronargs - funcform->pronargdefaults; |
| 4330 |
foreach(lc, defaults) |
| 4331 |
{ |
| 4332 |
if (argarray[i] == NULL) |
| 4333 |
argarray[i] = (Node *) lfirst(lc); |
| 4334 |
i++; |
| 4335 |
} |
| 4336 |
} |
| 4337 |
|
| 4338 |
/* Now reconstruct the args list in proper order */ |
| 4339 |
args = NIL; |
| 4340 |
for (i = 0; i < pronargs; i++) |
| 4341 |
{ |
| 4342 |
Assert(argarray[i] != NULL); |
| 4343 |
args = lappend(args, argarray[i]); |
| 4344 |
} |
| 4345 |
|
| 4346 |
return args; |
| 4347 |
} |
| 4348 |
|
| 4349 |
/* |
| 4350 |
* add_function_defaults: add missing function arguments from its defaults |
| 4351 |
* |
| 4352 |
* This is used only when the argument list was positional to begin with, |
| 4353 |
* and so we know we just need to add defaults at the end. |
| 4354 |
*/ |
| 4355 |
static List * |
| 4356 |
add_function_defaults(List *args, int pronargs, HeapTuple func_tuple) |
| 4357 |
{ |
| 4358 |
int nargsprovided = list_length(args); |
| 4359 |
List *defaults; |
| 4360 |
int ndelete; |
| 4361 |
|
| 4362 |
/* Get all the default expressions from the pg_proc tuple */ |
| 4363 |
defaults = fetch_function_defaults(func_tuple); |
| 4364 |
|
| 4365 |
/* Delete any unused defaults from the list */ |
| 4366 |
ndelete = nargsprovided + list_length(defaults) - pronargs; |
| 4367 |
if (ndelete < 0) |
| 4368 |
elog(ERROR, "not enough default arguments"); |
| 4369 |
if (ndelete > 0) |
| 4370 |
defaults = list_delete_first_n(defaults, ndelete); |
| 4371 |
|
| 4372 |
/* And form the combined argument list, not modifying the input list */ |
| 4373 |
return list_concat_copy(args, defaults); |
| 4374 |
} |
| 4375 |
|
| 4376 |
/* |
| 4377 |
* fetch_function_defaults: get function's default arguments as expression list |
| 4378 |
*/ |
| 4379 |
static List * |
| 4380 |
fetch_function_defaults(HeapTuple func_tuple) |
| 4381 |
{ |
| 4382 |
List *defaults; |
| 4383 |
Datum proargdefaults; |
| 4384 |
char *str; |
| 4385 |
|
| 4386 |
proargdefaults = SysCacheGetAttrNotNull(PROCOID, func_tuple, |
| 4387 |
Anum_pg_proc_proargdefaults); |
| 4388 |
str = TextDatumGetCString(proargdefaults); |
| 4389 |
defaults = castNode(List, stringToNode(str)); |
| 4390 |
pfree(str); |
| 4391 |
return defaults; |
| 4392 |
} |
| 4393 |
|
| 4394 |
/* |
| 4395 |
* recheck_cast_function_args: recheck function args and typecast as needed |
| 4396 |
* after adding defaults. |
| 4397 |
* |
| 4398 |
* It is possible for some of the defaulted arguments to be polymorphic; |
| 4399 |
* therefore we can't assume that the default expressions have the correct |
| 4400 |
* data types already. We have to re-resolve polymorphics and do coercion |
| 4401 |
* just like the parser did. |
| 4402 |
* |
| 4403 |
* This should be a no-op if there are no polymorphic arguments, |
| 4404 |
* but we do it anyway to be sure. |
| 4405 |
* |
| 4406 |
* Note: if any casts are needed, the args list is modified in-place; |
| 4407 |
* caller should have already copied the list structure. |
| 4408 |
*/ |
| 4409 |
static void |
| 4410 |
recheck_cast_function_args(List *args, Oid result_type, |
| 4411 |
Oid *proargtypes, int pronargs, |
| 4412 |
HeapTuple func_tuple) |
| 4413 |
{ |
| 4414 |
Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple); |
| 4415 |
int nargs; |
| 4416 |
Oid actual_arg_types[FUNC_MAX_ARGS] = {0}; |
| 4417 |
Oid declared_arg_types[FUNC_MAX_ARGS]; |
| 4418 |
Oid rettype; |
| 4419 |
ListCell *lc; |
| 4420 |
|
| 4421 |
if (list_length(args) > FUNC_MAX_ARGS) |
| 4422 |
elog(ERROR, "too many function arguments"); |
| 4423 |
nargs = 0; |
| 4424 |
foreach(lc, args) |
| 4425 |
{ |
| 4426 |
actual_arg_types[nargs++] = exprType((Node *) lfirst(lc)); |
| 4427 |
} |
| 4428 |
Assert(nargs == pronargs); |
| 4429 |
memcpy(declared_arg_types, proargtypes, pronargs * sizeof(Oid)); |
| 4430 |
rettype = enforce_generic_type_consistency(actual_arg_types, |
| 4431 |
declared_arg_types, |
| 4432 |
nargs, |
| 4433 |
funcform->prorettype, |
| 4434 |
false); |
| 4435 |
/* let's just check we got the same answer as the parser did ... */ |
| 4436 |
if (rettype != result_type) |
| 4437 |
elog(ERROR, "function's resolved result type changed during planning"); |
| 4438 |
|
| 4439 |
/* perform any necessary typecasting of arguments */ |
| 4440 |
make_fn_arguments(NULL, args, actual_arg_types, declared_arg_types); |
| 4441 |
} |
| 4442 |
|
| 4443 |
/* |
| 4444 |
* evaluate_function: try to pre-evaluate a function call |
| 4445 |
* |
| 4446 |
* We can do this if the function is strict and has any constant-null inputs |
| 4447 |
* (just return a null constant), or if the function is immutable and has all |
| 4448 |
* constant inputs (call it and return the result as a Const node). In |
| 4449 |
* estimation mode we are willing to pre-evaluate stable functions too. |
| 4450 |
* |
| 4451 |
* Returns a simplified expression if successful, or NULL if cannot |
| 4452 |
* simplify the function. |
| 4453 |
*/ |
| 4454 |
static Expr * |
| 4455 |
evaluate_function(Oid funcid, Oid result_type, int32 result_typmod, |
| 4456 |
Oid result_collid, Oid input_collid, List *args, |
| 4457 |
bool funcvariadic, |
| 4458 |
HeapTuple func_tuple, |
| 4459 |
eval_const_expressions_context *context) |
| 4460 |
{ |
| 4461 |
Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple); |
| 4462 |
bool has_nonconst_input = false; |
| 4463 |
bool has_null_input = false; |
| 4464 |
ListCell *arg; |
| 4465 |
FuncExpr *newexpr; |
| 4466 |
|
| 4467 |
/* |
| 4468 |
* Can't simplify if it returns a set. |
| 4469 |
*/ |
| 4470 |
if (funcform->proretset) |
| 4471 |
return NULL; |
| 4472 |
|
| 4473 |
/* |
| 4474 |
* Can't simplify if it returns RECORD. The immediate problem is that it |
| 4475 |
* will be needing an expected tupdesc which we can't supply here. |
| 4476 |
* |
| 4477 |
* In the case where it has OUT parameters, we could build an expected |
| 4478 |
* tupdesc from those, but there may be other gotchas lurking. In |
| 4479 |
* particular, if the function were to return NULL, we would produce a |
| 4480 |
* null constant with no remaining indication of which concrete record |
| 4481 |
* type it is. For now, seems best to leave the function call unreduced. |
| 4482 |
*/ |
| 4483 |
if (funcform->prorettype == RECORDOID) |
| 4484 |
return NULL; |
| 4485 |
|
| 4486 |
/* |
| 4487 |
* Check for constant inputs and especially constant-NULL inputs. |
| 4488 |
*/ |
| 4489 |
foreach(arg, args) |
| 4490 |
{ |
| 4491 |
if (IsA(lfirst(arg), Const)) |
| 4492 |
has_null_input |= ((Const *) lfirst(arg))->constisnull; |
| 4493 |
else |
| 4494 |
has_nonconst_input = true; |
| 4495 |
} |
| 4496 |
|
| 4497 |
/* |
| 4498 |
* If the function is strict and has a constant-NULL input, it will never |
| 4499 |
* be called at all, so we can replace the call by a NULL constant, even |
| 4500 |
* if there are other inputs that aren't constant, and even if the |
| 4501 |
* function is not otherwise immutable. |
| 4502 |
*/ |
| 4503 |
if (funcform->proisstrict && has_null_input) |
| 4504 |
return (Expr *) makeNullConst(result_type, result_typmod, |
| 4505 |
result_collid); |
| 4506 |
|
| 4507 |
/* |
| 4508 |
* Otherwise, can simplify only if all inputs are constants. (For a |
| 4509 |
* non-strict function, constant NULL inputs are treated the same as |
| 4510 |
* constant non-NULL inputs.) |
| 4511 |
*/ |
| 4512 |
if (has_nonconst_input) |
| 4513 |
return NULL; |
| 4514 |
|
| 4515 |
/* |
| 4516 |
* Ordinarily we are only allowed to simplify immutable functions. But for |
| 4517 |
* purposes of estimation, we consider it okay to simplify functions that |
| 4518 |
* are merely stable; the risk that the result might change from planning |
| 4519 |
* time to execution time is worth taking in preference to not being able |
| 4520 |
* to estimate the value at all. |
| 4521 |
*/ |
| 4522 |
if (funcform->provolatile == PROVOLATILE_IMMUTABLE) |
| 4523 |
/* okay */ ; |
| 4524 |
else if (context->estimate && funcform->provolatile == PROVOLATILE_STABLE) |
| 4525 |
/* okay */ ; |
| 4526 |
else |
| 4527 |
return NULL; |
| 4528 |
|
| 4529 |
/* |
| 4530 |
* OK, looks like we can simplify this operator/function. |
| 4531 |
* |
| 4532 |
* Build a new FuncExpr node containing the already-simplified arguments. |
| 4533 |
*/ |
| 4534 |
newexpr = makeNode(FuncExpr); |
| 4535 |
newexpr->funcid = funcid; |
| 4536 |
newexpr->funcresulttype = result_type; |
| 4537 |
newexpr->funcretset = false; |
| 4538 |
newexpr->funcvariadic = funcvariadic; |
| 4539 |
newexpr->funcformat = COERCE_EXPLICIT_CALL; /* doesn't matter */ |
| 4540 |
newexpr->funccollid = result_collid; /* doesn't matter */ |
| 4541 |
newexpr->inputcollid = input_collid; |
| 4542 |
newexpr->args = args; |
| 4543 |
newexpr->location = -1; |
| 4544 |
|
| 4545 |
return evaluate_expr((Expr *) newexpr, result_type, result_typmod, |
| 4546 |
result_collid); |
| 4547 |
} |
| 4548 |
|
| 4549 |
/* |
| 4550 |
* inline_function: try to expand a function call inline |
| 4551 |
* |
| 4552 |
* If the function is a sufficiently simple SQL-language function |
| 4553 |
* (just "SELECT expression"), then we can inline it and avoid the rather |
| 4554 |
* high per-call overhead of SQL functions. Furthermore, this can expose |
| 4555 |
* opportunities for constant-folding within the function expression. |
| 4556 |
* |
| 4557 |
* We have to beware of some special cases however. A directly or |
| 4558 |
* indirectly recursive function would cause us to recurse forever, |
| 4559 |
* so we keep track of which functions we are already expanding and |
| 4560 |
* do not re-expand them. Also, if a parameter is used more than once |
| 4561 |
* in the SQL-function body, we require it not to contain any volatile |
| 4562 |
* functions (volatiles might deliver inconsistent answers) nor to be |
| 4563 |
* unreasonably expensive to evaluate. The expensiveness check not only |
| 4564 |
* prevents us from doing multiple evaluations of an expensive parameter |
| 4565 |
* at runtime, but is a safety value to limit growth of an expression due |
| 4566 |
* to repeated inlining. |
| 4567 |
* |
| 4568 |
* We must also beware of changing the volatility or strictness status of |
| 4569 |
* functions by inlining them. |
| 4570 |
* |
| 4571 |
* Also, at the moment we can't inline functions returning RECORD. This |
| 4572 |
* doesn't work in the general case because it discards information such |
| 4573 |
* as OUT-parameter declarations. |
| 4574 |
* |
| 4575 |
* Also, context-dependent expression nodes in the argument list are trouble. |
| 4576 |
* |
| 4577 |
* Returns a simplified expression if successful, or NULL if cannot |
| 4578 |
* simplify the function. |
| 4579 |
*/ |
| 4580 |
static Expr * |
| 4581 |
inline_function(Oid funcid, Oid result_type, Oid result_collid, |
| 4582 |
Oid input_collid, List *args, |
| 4583 |
bool funcvariadic, |
| 4584 |
HeapTuple func_tuple, |
| 4585 |
eval_const_expressions_context *context) |
| 4586 |
{ |
| 4587 |
Form_pg_proc funcform = (Form_pg_proc) GETSTRUCT(func_tuple); |
| 4588 |
char *src; |
| 4589 |
Datum tmp; |
| 4590 |
bool isNull; |
| 4591 |
MemoryContext oldcxt; |
| 4592 |
MemoryContext mycxt; |
| 4593 |
inline_error_callback_arg callback_arg; |
| 4594 |
ErrorContextCallback sqlerrcontext; |
| 4595 |
FuncExpr *fexpr; |
| 4596 |
SQLFunctionParseInfoPtr pinfo; |
| 4597 |
TupleDesc rettupdesc; |
| 4598 |
ParseState *pstate; |
| 4599 |
List *raw_parsetree_list; |
| 4600 |
List *querytree_list; |
| 4601 |
Query *querytree; |
| 4602 |
Node *newexpr; |
| 4603 |
int *usecounts; |
| 4604 |
ListCell *arg; |
| 4605 |
int i; |
| 4606 |
|
| 4607 |
/* |
| 4608 |
* Forget it if the function is not SQL-language or has other showstopper |
| 4609 |
* properties. (The prokind and nargs checks are just paranoia.) |
| 4610 |
*/ |
| 4611 |
if (funcform->prolang != SQLlanguageId || |
| 4612 |
funcform->prokind != PROKIND_FUNCTION || |
| 4613 |
funcform->prosecdef || |
| 4614 |
funcform->proretset || |
| 4615 |
funcform->prorettype == RECORDOID || |
| 4616 |
!heap_attisnull(func_tuple, Anum_pg_proc_proconfig, NULL) || |
| 4617 |
funcform->pronargs != list_length(args)) |
| 4618 |
return NULL; |
| 4619 |
|
| 4620 |
/* Check for recursive function, and give up trying to expand if so */ |
| 4621 |
if (list_member_oid(context->active_fns, funcid)) |
| 4622 |
return NULL; |
| 4623 |
|
| 4624 |
/* Check permission to call function (fail later, if not) */ |
| 4625 |
if (object_aclcheck(ProcedureRelationId, funcid, GetUserId(), ACL_EXECUTE) != ACLCHECK_OK) |
| 4626 |
return NULL; |
| 4627 |
|
| 4628 |
/* Check whether a plugin wants to hook function entry/exit */ |
| 4629 |
if (FmgrHookIsNeeded(funcid)) |
| 4630 |
return NULL; |
| 4631 |
|
| 4632 |
/* |
| 4633 |
* Make a temporary memory context, so that we don't leak all the stuff |
| 4634 |
* that parsing might create. |
| 4635 |
*/ |
| 4636 |
mycxt = AllocSetContextCreate(CurrentMemoryContext, |
| 4637 |
"inline_function", |
| 4638 |
ALLOCSET_DEFAULT_SIZES); |
| 4639 |
oldcxt = MemoryContextSwitchTo(mycxt); |
| 4640 |
|
| 4641 |
/* |
| 4642 |
* We need a dummy FuncExpr node containing the already-simplified |
| 4643 |
* arguments. (In some cases we don't really need it, but building it is |
| 4644 |
* cheap enough that it's not worth contortions to avoid.) |
| 4645 |
*/ |
| 4646 |
fexpr = makeNode(FuncExpr); |
| 4647 |
fexpr->funcid = funcid; |
| 4648 |
fexpr->funcresulttype = result_type; |
| 4649 |
fexpr->funcretset = false; |
| 4650 |
fexpr->funcvariadic = funcvariadic; |
| 4651 |
fexpr->funcformat = COERCE_EXPLICIT_CALL; /* doesn't matter */ |
| 4652 |
fexpr->funccollid = result_collid; /* doesn't matter */ |
| 4653 |
fexpr->inputcollid = input_collid; |
| 4654 |
fexpr->args = args; |
| 4655 |
fexpr->location = -1; |
| 4656 |
|
| 4657 |
/* Fetch the function body */ |
| 4658 |
tmp = SysCacheGetAttrNotNull(PROCOID, func_tuple, Anum_pg_proc_prosrc); |
| 4659 |
src = TextDatumGetCString(tmp); |
| 4660 |
|
| 4661 |
/* |
| 4662 |
* Setup error traceback support for ereport(). This is so that we can |
| 4663 |
* finger the function that bad information came from. |
| 4664 |
*/ |
| 4665 |
callback_arg.proname = NameStr(funcform->proname); |
| 4666 |
callback_arg.prosrc = src; |
| 4667 |
|
| 4668 |
sqlerrcontext.callback = sql_inline_error_callback; |
| 4669 |
sqlerrcontext.arg = &callback_arg; |
| 4670 |
sqlerrcontext.previous = error_context_stack; |
| 4671 |
error_context_stack = &sqlerrcontext; |
| 4672 |
|
| 4673 |
/* If we have prosqlbody, pay attention to that not prosrc */ |
| 4674 |
tmp = SysCacheGetAttr(PROCOID, |
| 4675 |
func_tuple, |
| 4676 |
Anum_pg_proc_prosqlbody, |
| 4677 |
&isNull); |
| 4678 |
if (!isNull) |
| 4679 |
{ |
| 4680 |
Node *n; |
| 4681 |
List *query_list; |
| 4682 |
|
| 4683 |
n = stringToNode(TextDatumGetCString(tmp)); |
| 4684 |
if (IsA(n, List)) |
| 4685 |
query_list = linitial_node(List, castNode(List, n)); |
| 4686 |
else |
| 4687 |
query_list = list_make1(n); |
| 4688 |
if (list_length(query_list) != 1) |
| 4689 |
goto fail; |
| 4690 |
querytree = linitial(query_list); |
| 4691 |
|
| 4692 |
/* |
| 4693 |
* Because we'll insist below that the querytree have an empty rtable |
| 4694 |
* and no sublinks, it cannot have any relation references that need |
| 4695 |
* to be locked or rewritten. So we can omit those steps. |
| 4696 |
*/ |
| 4697 |
} |
| 4698 |
else |
| 4699 |
{ |
| 4700 |
/* Set up to handle parameters while parsing the function body. */ |
| 4701 |
pinfo = prepare_sql_fn_parse_info(func_tuple, |
| 4702 |
(Node *) fexpr, |
| 4703 |
input_collid); |
| 4704 |
|
| 4705 |
/* |
| 4706 |
* We just do parsing and parse analysis, not rewriting, because |
| 4707 |
* rewriting will not affect table-free-SELECT-only queries, which is |
| 4708 |
* all that we care about. Also, we can punt as soon as we detect |
| 4709 |
* more than one command in the function body. |
| 4710 |
*/ |
| 4711 |
raw_parsetree_list = pg_parse_query(src); |
| 4712 |
if (list_length(raw_parsetree_list) != 1) |
| 4713 |
goto fail; |
| 4714 |
|
| 4715 |
pstate = make_parsestate(NULL); |
| 4716 |
pstate->p_sourcetext = src; |
| 4717 |
sql_fn_parser_setup(pstate, pinfo); |
| 4718 |
|
| 4719 |
querytree = transformTopLevelStmt(pstate, linitial(raw_parsetree_list)); |
| 4720 |
|
| 4721 |
free_parsestate(pstate); |
| 4722 |
} |
| 4723 |
|
| 4724 |
/* |
| 4725 |
* The single command must be a simple "SELECT expression". |
| 4726 |
* |
| 4727 |
* Note: if you change the tests involved in this, see also plpgsql's |
| 4728 |
* exec_simple_check_plan(). That generally needs to have the same idea |
| 4729 |
* of what's a "simple expression", so that inlining a function that |
| 4730 |
* previously wasn't inlined won't change plpgsql's conclusion. |
| 4731 |
*/ |
| 4732 |
if (!IsA(querytree, Query) || |
| 4733 |
querytree->commandType != CMD_SELECT || |
| 4734 |
querytree->hasAggs || |
| 4735 |
querytree->hasWindowFuncs || |
| 4736 |
querytree->hasTargetSRFs || |
| 4737 |
querytree->hasSubLinks || |
| 4738 |
querytree->cteList || |
| 4739 |
querytree->rtable || |
| 4740 |
querytree->jointree->fromlist || |
| 4741 |
querytree->jointree->quals || |
| 4742 |
querytree->groupClause || |
| 4743 |
querytree->groupingSets || |
| 4744 |
querytree->havingQual || |
| 4745 |
querytree->windowClause || |
| 4746 |
querytree->distinctClause || |
| 4747 |
querytree->sortClause || |
| 4748 |
querytree->limitOffset || |
| 4749 |
querytree->limitCount || |
| 4750 |
querytree->setOperations || |
| 4751 |
list_length(querytree->targetList) != 1) |
| 4752 |
goto fail; |
| 4753 |
|
| 4754 |
/* If the function result is composite, resolve it */ |
| 4755 |
(void) get_expr_result_type((Node *) fexpr, |
| 4756 |
NULL, |
| 4757 |
&rettupdesc); |
| 4758 |
|
| 4759 |
/* |
| 4760 |
* Make sure the function (still) returns what it's declared to. This |
| 4761 |
* will raise an error if wrong, but that's okay since the function would |
| 4762 |
* fail at runtime anyway. Note that check_sql_fn_retval will also insert |
| 4763 |
* a coercion if needed to make the tlist expression match the declared |
| 4764 |
* type of the function. |
| 4765 |
* |
| 4766 |
* Note: we do not try this until we have verified that no rewriting was |
| 4767 |
* needed; that's probably not important, but let's be careful. |
| 4768 |
*/ |
| 4769 |
querytree_list = list_make1(querytree); |
| 4770 |
if (check_sql_fn_retval(list_make1(querytree_list), |
| 4771 |
result_type, rettupdesc, |
| 4772 |
funcform->prokind, |
| 4773 |
false)) |
| 4774 |
goto fail; /* reject whole-tuple-result cases */ |
| 4775 |
|
| 4776 |
/* |
| 4777 |
* Given the tests above, check_sql_fn_retval shouldn't have decided to |
| 4778 |
* inject a projection step, but let's just make sure. |
| 4779 |
*/ |
| 4780 |
if (querytree != linitial(querytree_list)) |
| 4781 |
goto fail; |
| 4782 |
|
| 4783 |
/* Now we can grab the tlist expression */ |
| 4784 |
newexpr = (Node *) ((TargetEntry *) linitial(querytree->targetList))->expr; |
| 4785 |
|
| 4786 |
/* |
| 4787 |
* If the SQL function returns VOID, we can only inline it if it is a |
| 4788 |
* SELECT of an expression returning VOID (ie, it's just a redirection to |
| 4789 |
* another VOID-returning function). In all non-VOID-returning cases, |
| 4790 |
* check_sql_fn_retval should ensure that newexpr returns the function's |
| 4791 |
* declared result type, so this test shouldn't fail otherwise; but we may |
| 4792 |
* as well cope gracefully if it does. |
| 4793 |
*/ |
| 4794 |
if (exprType(newexpr) != result_type) |
| 4795 |
goto fail; |
| 4796 |
|
| 4797 |
/* |
| 4798 |
* Additional validity checks on the expression. It mustn't be more |
| 4799 |
* volatile than the surrounding function (this is to avoid breaking hacks |
| 4800 |
* that involve pretending a function is immutable when it really ain't). |
| 4801 |
* If the surrounding function is declared strict, then the expression |
| 4802 |
* must contain only strict constructs and must use all of the function |
| 4803 |
* parameters (this is overkill, but an exact analysis is hard). |
| 4804 |
*/ |
| 4805 |
if (funcform->provolatile == PROVOLATILE_IMMUTABLE && |
| 4806 |
contain_mutable_functions(newexpr)) |
| 4807 |
goto fail; |
| 4808 |
else if (funcform->provolatile == PROVOLATILE_STABLE && |
| 4809 |
contain_volatile_functions(newexpr)) |
| 4810 |
goto fail; |
| 4811 |
|
| 4812 |
if (funcform->proisstrict && |
| 4813 |
contain_nonstrict_functions(newexpr)) |
| 4814 |
goto fail; |
| 4815 |
|
| 4816 |
/* |
| 4817 |
* If any parameter expression contains a context-dependent node, we can't |
| 4818 |
* inline, for fear of putting such a node into the wrong context. |
| 4819 |
*/ |
| 4820 |
if (contain_context_dependent_node((Node *) args)) |
| 4821 |
goto fail; |
| 4822 |
|
| 4823 |
/* |
| 4824 |
* We may be able to do it; there are still checks on parameter usage to |
| 4825 |
* make, but those are most easily done in combination with the actual |
| 4826 |
* substitution of the inputs. So start building expression with inputs |
| 4827 |
* substituted. |
| 4828 |
*/ |
| 4829 |
usecounts = (int *) palloc0(funcform->pronargs * sizeof(int)); |
| 4830 |
newexpr = substitute_actual_parameters(newexpr, funcform->pronargs, |
| 4831 |
args, usecounts); |
| 4832 |
|
| 4833 |
/* Now check for parameter usage */ |
| 4834 |
i = 0; |
| 4835 |
foreach(arg, args) |
| 4836 |
{ |
| 4837 |
Node *param = lfirst(arg); |
| 4838 |
|
| 4839 |
if (usecounts[i] == 0) |
| 4840 |
{ |
| 4841 |
/* Param not used at all: uncool if func is strict */ |
| 4842 |
if (funcform->proisstrict) |
| 4843 |
goto fail; |
| 4844 |
} |
| 4845 |
else if (usecounts[i] != 1) |
| 4846 |
{ |
| 4847 |
/* Param used multiple times: uncool if expensive or volatile */ |
| 4848 |
QualCost eval_cost; |
| 4849 |
|
| 4850 |
/* |
| 4851 |
* We define "expensive" as "contains any subplan or more than 10 |
| 4852 |
* operators". Note that the subplan search has to be done |
| 4853 |
* explicitly, since cost_qual_eval() will barf on unplanned |
| 4854 |
* subselects. |
| 4855 |
*/ |
| 4856 |
if (contain_subplans(param)) |
| 4857 |
goto fail; |
| 4858 |
cost_qual_eval(&eval_cost, list_make1(param), NULL); |
| 4859 |
if (eval_cost.startup + eval_cost.per_tuple > |
| 4860 |
10 * cpu_operator_cost) |
| 4861 |
goto fail; |
| 4862 |
|
| 4863 |
/* |
| 4864 |
* Check volatility last since this is more expensive than the |
| 4865 |
* above tests |
| 4866 |
*/ |
| 4867 |
if (contain_volatile_functions(param)) |
| 4868 |
goto fail; |
| 4869 |
} |
| 4870 |
i++; |
| 4871 |
} |
| 4872 |
|
| 4873 |
/* |
| 4874 |
* Whew --- we can make the substitution. Copy the modified expression |
| 4875 |
* out of the temporary memory context, and clean up. |
| 4876 |
*/ |
| 4877 |
MemoryContextSwitchTo(oldcxt); |
| 4878 |
|
| 4879 |
newexpr = copyObject(newexpr); |
| 4880 |
|
| 4881 |
MemoryContextDelete(mycxt); |
| 4882 |
|
| 4883 |
/* |
| 4884 |
* If the result is of a collatable type, force the result to expose the |
| 4885 |
* correct collation. In most cases this does not matter, but it's |
| 4886 |
* possible that the function result is used directly as a sort key or in |
| 4887 |
* other places where we expect exprCollation() to tell the truth. |
| 4888 |
*/ |
| 4889 |
if (OidIsValid(result_collid)) |
| 4890 |
{ |
| 4891 |
Oid exprcoll = exprCollation(newexpr); |
| 4892 |
|
| 4893 |
if (OidIsValid(exprcoll) && exprcoll != result_collid) |
| 4894 |
{ |
| 4895 |
CollateExpr *newnode = makeNode(CollateExpr); |
| 4896 |
|
| 4897 |
newnode->arg = (Expr *) newexpr; |
| 4898 |
newnode->collOid = result_collid; |
| 4899 |
newnode->location = -1; |
| 4900 |
|
| 4901 |
newexpr = (Node *) newnode; |
| 4902 |
} |
| 4903 |
} |
| 4904 |
|
| 4905 |
/* |
| 4906 |
* Since there is now no trace of the function in the plan tree, we must |
| 4907 |
* explicitly record the plan's dependency on the function. |
| 4908 |
*/ |
| 4909 |
if (context->root) |
| 4910 |
record_plan_function_dependency(context->root, funcid); |
| 4911 |
|
| 4912 |
/* |
| 4913 |
* Recursively try to simplify the modified expression. Here we must add |
| 4914 |
* the current function to the context list of active functions. |
| 4915 |
*/ |
| 4916 |
context->active_fns = lappend_oid(context->active_fns, funcid); |
| 4917 |
newexpr = eval_const_expressions_mutator(newexpr, context); |
| 4918 |
context->active_fns = list_delete_last(context->active_fns); |
| 4919 |
|
| 4920 |
error_context_stack = sqlerrcontext.previous; |
| 4921 |
|
| 4922 |
return (Expr *) newexpr; |
| 4923 |
|
| 4924 |
/* Here if func is not inlinable: release temp memory and return NULL */ |
| 4925 |
fail: |
| 4926 |
MemoryContextSwitchTo(oldcxt); |
| 4927 |
MemoryContextDelete(mycxt); |
| 4928 |
error_context_stack = sqlerrcontext.previous; |
| 4929 |
|
| 4930 |
return NULL; |
| 4931 |
} |
| 4932 |
|
| 4933 |
/* |
| 4934 |
* Replace Param nodes by appropriate actual parameters |
| 4935 |
*/ |
| 4936 |
static Node * |
| 4937 |
substitute_actual_parameters(Node *expr, int nargs, List *args, |
| 4938 |
int *usecounts) |
| 4939 |
{ |
| 4940 |
substitute_actual_parameters_context context; |
| 4941 |
|
| 4942 |
context.nargs = nargs; |
| 4943 |
context.args = args; |
| 4944 |
context.usecounts = usecounts; |
| 4945 |
|
| 4946 |
return substitute_actual_parameters_mutator(expr, &context); |
| 4947 |
} |
| 4948 |
|
| 4949 |
static Node * |
| 4950 |
substitute_actual_parameters_mutator(Node *node, |
| 4951 |
substitute_actual_parameters_context *context) |
| 4952 |
{ |
| 4953 |
if (node == NULL) |
| 4954 |
return NULL; |
| 4955 |
if (IsA(node, Param)) |
| 4956 |
{ |
| 4957 |
Param *param = (Param *) node; |
| 4958 |
|
| 4959 |
if (param->paramkind != PARAM_EXTERN) |
| 4960 |
elog(ERROR, "unexpected paramkind: %d", (int) param->paramkind); |
| 4961 |
if (param->paramid <= 0 || param->paramid > context->nargs) |
| 4962 |
elog(ERROR, "invalid paramid: %d", param->paramid); |
| 4963 |
|
| 4964 |
/* Count usage of parameter */ |
| 4965 |
context->usecounts[param->paramid - 1]++; |
| 4966 |
|
| 4967 |
/* Select the appropriate actual arg and replace the Param with it */ |
| 4968 |
/* We don't need to copy at this time (it'll get done later) */ |
| 4969 |
return list_nth(context->args, param->paramid - 1); |
| 4970 |
} |
| 4971 |
return expression_tree_mutator(node, substitute_actual_parameters_mutator, context); |
| 4972 |
} |
| 4973 |
|
| 4974 |
/* |
| 4975 |
* error context callback to let us supply a call-stack traceback |
| 4976 |
*/ |
| 4977 |
static void |
| 4978 |
sql_inline_error_callback(void *arg) |
| 4979 |
{ |
| 4980 |
inline_error_callback_arg *callback_arg = (inline_error_callback_arg *) arg; |
| 4981 |
int syntaxerrposition; |
| 4982 |
|
| 4983 |
/* If it's a syntax error, convert to internal syntax error report */ |
| 4984 |
syntaxerrposition = geterrposition(); |
| 4985 |
if (syntaxerrposition > 0) |
| 4986 |
{ |
| 4987 |
errposition(0); |
| 4988 |
internalerrposition(syntaxerrposition); |
| 4989 |
internalerrquery(callback_arg->prosrc); |
| 4990 |
} |
| 4991 |
|
| 4992 |
errcontext("SQL function \"%s\" during inlining", callback_arg->proname); |
| 4993 |
} |
| 4994 |
|
| 4995 |
/* |
| 4996 |
* evaluate_expr: pre-evaluate a constant expression |
| 4997 |
* |
| 4998 |
* We use the executor's routine ExecEvalExpr() to avoid duplication of |
| 4999 |
* code and ensure we get the same result as the executor would get. |
| 5000 |
*/ |
| 5001 |
Expr * |
| 5002 |
evaluate_expr(Expr *expr, Oid result_type, int32 result_typmod, |
| 5003 |
Oid result_collation) |
| 5004 |
{ |
| 5005 |
EState *estate; |
| 5006 |
ExprState *exprstate; |
| 5007 |
MemoryContext oldcontext; |
| 5008 |
Datum const_val; |
| 5009 |
bool const_is_null; |
| 5010 |
int16 resultTypLen; |
| 5011 |
bool resultTypByVal; |
| 5012 |
|
| 5013 |
/* |
| 5014 |
* To use the executor, we need an EState. |
| 5015 |
*/ |
| 5016 |
estate = CreateExecutorState(); |
| 5017 |
|
| 5018 |
/* We can use the estate's working context to avoid memory leaks. */ |
| 5019 |
oldcontext = MemoryContextSwitchTo(estate->es_query_cxt); |
| 5020 |
|
| 5021 |
/* Make sure any opfuncids are filled in. */ |
| 5022 |
fix_opfuncids((Node *) expr); |
| 5023 |
|
| 5024 |
/* |
| 5025 |
* Prepare expr for execution. (Note: we can't use ExecPrepareExpr |
| 5026 |
* because it'd result in recursively invoking eval_const_expressions.) |
| 5027 |
*/ |
| 5028 |
exprstate = ExecInitExpr(expr, NULL); |
| 5029 |
|
| 5030 |
/* |
| 5031 |
* And evaluate it. |
| 5032 |
* |
| 5033 |
* It is OK to use a default econtext because none of the ExecEvalExpr() |
| 5034 |
* code used in this situation will use econtext. That might seem |
| 5035 |
* fortuitous, but it's not so unreasonable --- a constant expression does |
| 5036 |
* not depend on context, by definition, n'est ce pas? |
| 5037 |
*/ |
| 5038 |
const_val = ExecEvalExprSwitchContext(exprstate, |
| 5039 |
GetPerTupleExprContext(estate), |
| 5040 |
&const_is_null); |
| 5041 |
|
| 5042 |
/* Get info needed about result datatype */ |
| 5043 |
get_typlenbyval(result_type, &resultTypLen, &resultTypByVal); |
| 5044 |
|
| 5045 |
/* Get back to outer memory context */ |
| 5046 |
MemoryContextSwitchTo(oldcontext); |
| 5047 |
|
| 5048 |
/* |
| 5049 |
* Must copy result out of sub-context used by expression eval. |
| 5050 |
* |
| 5051 |
* Also, if it's varlena, forcibly detoast it. This protects us against |
| 5052 |
* storing TOAST pointers into plans that might outlive the referenced |
| 5053 |
* data. (makeConst would handle detoasting anyway, but it's worth a few |
| 5054 |
* extra lines here so that we can do the copy and detoast in one step.) |
| 5055 |
*/ |
| 5056 |
if (!const_is_null) |
| 5057 |
{ |
| 5058 |
if (resultTypLen == -1) |
| 5059 |
const_val = PointerGetDatum(PG_DETOAST_DATUM_COPY(const_val)); |
| 5060 |
else |
| 5061 |
const_val = datumCopy(const_val, resultTypByVal, resultTypLen); |
| 5062 |
} |
| 5063 |
|
| 5064 |
/* Release all the junk we just created */ |
| 5065 |
FreeExecutorState(estate); |
| 5066 |
|
| 5067 |
/* |
| 5068 |
* Make the constant result node. |
| 5069 |
*/ |
| 5070 |
return (Expr *) makeConst(result_type, result_typmod, result_collation, |
| 5071 |
resultTypLen, |
| 5072 |
const_val, const_is_null, |
| 5073 |
resultTypByVal); |
| 5074 |
} |
| 5075 |
|
| 5076 |
|
| 5077 |
/* |
| 5078 |
* inline_set_returning_function |
| 5079 |
* Attempt to "inline" a set-returning function in the FROM clause. |
| 5080 |
* |
| 5081 |
* "rte" is an RTE_FUNCTION rangetable entry. If it represents a call of a |
| 5082 |
* set-returning SQL function that can safely be inlined, expand the function |
| 5083 |
* and return the substitute Query structure. Otherwise, return NULL. |
| 5084 |
* |
| 5085 |
* We assume that the RTE's expression has already been put through |
| 5086 |
* eval_const_expressions(), which among other things will take care of |
| 5087 |
* default arguments and named-argument notation. |
| 5088 |
* |
| 5089 |
* This has a good deal of similarity to inline_function(), but that's |
| 5090 |
* for the non-set-returning case, and there are enough differences to |
| 5091 |
* justify separate functions. |
| 5092 |
*/ |
| 5093 |
Query * |
| 5094 |
inline_set_returning_function(PlannerInfo *root, RangeTblEntry *rte) |
| 5095 |
{ |
| 5096 |
RangeTblFunction *rtfunc; |
| 5097 |
FuncExpr *fexpr; |
| 5098 |
Oid func_oid; |
| 5099 |
HeapTuple func_tuple; |
| 5100 |
Form_pg_proc funcform; |
| 5101 |
char *src; |
| 5102 |
Datum tmp; |
| 5103 |
bool isNull; |
| 5104 |
MemoryContext oldcxt; |
| 5105 |
MemoryContext mycxt; |
| 5106 |
inline_error_callback_arg callback_arg; |
| 5107 |
ErrorContextCallback sqlerrcontext; |
| 5108 |
SQLFunctionParseInfoPtr pinfo; |
| 5109 |
TypeFuncClass functypclass; |
| 5110 |
TupleDesc rettupdesc; |
| 5111 |
List *raw_parsetree_list; |
| 5112 |
List *querytree_list; |
| 5113 |
Query *querytree; |
| 5114 |
|
| 5115 |
Assert(rte->rtekind == RTE_FUNCTION); |
| 5116 |
|
| 5117 |
/* |
| 5118 |
* It doesn't make a lot of sense for a SQL SRF to refer to itself in its |
| 5119 |
* own FROM clause, since that must cause infinite recursion at runtime. |
| 5120 |
* It will cause this code to recurse too, so check for stack overflow. |
| 5121 |
* (There's no need to do more.) |
| 5122 |
*/ |
| 5123 |
check_stack_depth(); |
| 5124 |
|
| 5125 |
/* Fail if the RTE has ORDINALITY - we don't implement that here. */ |
| 5126 |
if (rte->funcordinality) |
| 5127 |
return NULL; |
| 5128 |
|
| 5129 |
/* Fail if RTE isn't a single, simple FuncExpr */ |
| 5130 |
if (list_length(rte->functions) != 1) |
| 5131 |
return NULL; |
| 5132 |
rtfunc = (RangeTblFunction *) linitial(rte->functions); |
| 5133 |
|
| 5134 |
if (!IsA(rtfunc->funcexpr, FuncExpr)) |
| 5135 |
return NULL; |
| 5136 |
fexpr = (FuncExpr *) rtfunc->funcexpr; |
| 5137 |
|
| 5138 |
func_oid = fexpr->funcid; |
| 5139 |
|
| 5140 |
/* |
| 5141 |
* The function must be declared to return a set, else inlining would |
| 5142 |
* change the results if the contained SELECT didn't return exactly one |
| 5143 |
* row. |
| 5144 |
*/ |
| 5145 |
if (!fexpr->funcretset) |
| 5146 |
return NULL; |
| 5147 |
|
| 5148 |
/* |
| 5149 |
* Refuse to inline if the arguments contain any volatile functions or |
| 5150 |
* sub-selects. Volatile functions are rejected because inlining may |
| 5151 |
* result in the arguments being evaluated multiple times, risking a |
| 5152 |
* change in behavior. Sub-selects are rejected partly for implementation |
| 5153 |
* reasons (pushing them down another level might change their behavior) |
| 5154 |
* and partly because they're likely to be expensive and so multiple |
| 5155 |
* evaluation would be bad. |
| 5156 |
*/ |
| 5157 |
if (contain_volatile_functions((Node *) fexpr->args) || |
| 5158 |
contain_subplans((Node *) fexpr->args)) |
| 5159 |
return NULL; |
| 5160 |
|
| 5161 |
/* Check permission to call function (fail later, if not) */ |
| 5162 |
if (object_aclcheck(ProcedureRelationId, func_oid, GetUserId(), ACL_EXECUTE) != ACLCHECK_OK) |
| 5163 |
return NULL; |
| 5164 |
|
| 5165 |
/* Check whether a plugin wants to hook function entry/exit */ |
| 5166 |
if (FmgrHookIsNeeded(func_oid)) |
| 5167 |
return NULL; |
| 5168 |
|
| 5169 |
/* |
| 5170 |
* OK, let's take a look at the function's pg_proc entry. |
| 5171 |
*/ |
| 5172 |
func_tuple = SearchSysCache1(PROCOID, ObjectIdGetDatum(func_oid)); |
| 5173 |
if (!HeapTupleIsValid(func_tuple)) |
| 5174 |
elog(ERROR, "cache lookup failed for function %u", func_oid); |
| 5175 |
funcform = (Form_pg_proc) GETSTRUCT(func_tuple); |
| 5176 |
|
| 5177 |
/* |
| 5178 |
* Forget it if the function is not SQL-language or has other showstopper |
| 5179 |
* properties. In particular it mustn't be declared STRICT, since we |
| 5180 |
* couldn't enforce that. It also mustn't be VOLATILE, because that is |
| 5181 |
* supposed to cause it to be executed with its own snapshot, rather than |
| 5182 |
* sharing the snapshot of the calling query. We also disallow returning |
| 5183 |
* SETOF VOID, because inlining would result in exposing the actual result |
| 5184 |
* of the function's last SELECT, which should not happen in that case. |
| 5185 |
* (Rechecking prokind, proretset, and pronargs is just paranoia.) |
| 5186 |
*/ |
| 5187 |
if (funcform->prolang != SQLlanguageId || |
| 5188 |
funcform->prokind != PROKIND_FUNCTION || |
| 5189 |
funcform->proisstrict || |
| 5190 |
funcform->provolatile == PROVOLATILE_VOLATILE || |
| 5191 |
funcform->prorettype == VOIDOID || |
| 5192 |
funcform->prosecdef || |
| 5193 |
!funcform->proretset || |
| 5194 |
list_length(fexpr->args) != funcform->pronargs || |
| 5195 |
!heap_attisnull(func_tuple, Anum_pg_proc_proconfig, NULL)) |
| 5196 |
{ |
| 5197 |
ReleaseSysCache(func_tuple); |
| 5198 |
return NULL; |
| 5199 |
} |
| 5200 |
|
| 5201 |
/* |
| 5202 |
* Make a temporary memory context, so that we don't leak all the stuff |
| 5203 |
* that parsing might create. |
| 5204 |
*/ |
| 5205 |
mycxt = AllocSetContextCreate(CurrentMemoryContext, |
| 5206 |
"inline_set_returning_function", |
| 5207 |
ALLOCSET_DEFAULT_SIZES); |
| 5208 |
oldcxt = MemoryContextSwitchTo(mycxt); |
| 5209 |
|
| 5210 |
/* Fetch the function body */ |
| 5211 |
tmp = SysCacheGetAttrNotNull(PROCOID, func_tuple, Anum_pg_proc_prosrc); |
| 5212 |
src = TextDatumGetCString(tmp); |
| 5213 |
|
| 5214 |
/* |
| 5215 |
* Setup error traceback support for ereport(). This is so that we can |
| 5216 |
* finger the function that bad information came from. |
| 5217 |
*/ |
| 5218 |
callback_arg.proname = NameStr(funcform->proname); |
| 5219 |
callback_arg.prosrc = src; |
| 5220 |
|
| 5221 |
sqlerrcontext.callback = sql_inline_error_callback; |
| 5222 |
sqlerrcontext.arg = &callback_arg; |
| 5223 |
sqlerrcontext.previous = error_context_stack; |
| 5224 |
error_context_stack = &sqlerrcontext; |
| 5225 |
|
| 5226 |
/* If we have prosqlbody, pay attention to that not prosrc */ |
| 5227 |
tmp = SysCacheGetAttr(PROCOID, |
| 5228 |
func_tuple, |
| 5229 |
Anum_pg_proc_prosqlbody, |
| 5230 |
&isNull); |
| 5231 |
if (!isNull) |
| 5232 |
{ |
| 5233 |
Node *n; |
| 5234 |
|
| 5235 |
n = stringToNode(TextDatumGetCString(tmp)); |
| 5236 |
if (IsA(n, List)) |
| 5237 |
querytree_list = linitial_node(List, castNode(List, n)); |
| 5238 |
else |
| 5239 |
querytree_list = list_make1(n); |
| 5240 |
if (list_length(querytree_list) != 1) |
| 5241 |
goto fail; |
| 5242 |
querytree = linitial(querytree_list); |
| 5243 |
|
| 5244 |
/* Acquire necessary locks, then apply rewriter. */ |
| 5245 |
AcquireRewriteLocks(querytree, true, false); |
| 5246 |
querytree_list = pg_rewrite_query(querytree); |
| 5247 |
if (list_length(querytree_list) != 1) |
| 5248 |
goto fail; |
| 5249 |
querytree = linitial(querytree_list); |
| 5250 |
} |
| 5251 |
else |
| 5252 |
{ |
| 5253 |
/* |
| 5254 |
* Set up to handle parameters while parsing the function body. We |
| 5255 |
* can use the FuncExpr just created as the input for |
| 5256 |
* prepare_sql_fn_parse_info. |
| 5257 |
*/ |
| 5258 |
pinfo = prepare_sql_fn_parse_info(func_tuple, |
| 5259 |
(Node *) fexpr, |
| 5260 |
fexpr->inputcollid); |
| 5261 |
|
| 5262 |
/* |
| 5263 |
* Parse, analyze, and rewrite (unlike inline_function(), we can't |
| 5264 |
* skip rewriting here). We can fail as soon as we find more than one |
| 5265 |
* query, though. |
| 5266 |
*/ |
| 5267 |
raw_parsetree_list = pg_parse_query(src); |
| 5268 |
if (list_length(raw_parsetree_list) != 1) |
| 5269 |
goto fail; |
| 5270 |
|
| 5271 |
querytree_list = pg_analyze_and_rewrite_withcb(linitial(raw_parsetree_list), |
| 5272 |
src, |
| 5273 |
(ParserSetupHook) sql_fn_parser_setup, |
| 5274 |
pinfo, NULL); |
| 5275 |
if (list_length(querytree_list) != 1) |
| 5276 |
goto fail; |
| 5277 |
querytree = linitial(querytree_list); |
| 5278 |
} |
| 5279 |
|
| 5280 |
/* |
| 5281 |
* Also resolve the actual function result tupdesc, if composite. If we |
| 5282 |
* have a coldeflist, believe that; otherwise use get_expr_result_type. |
| 5283 |
* (This logic should match ExecInitFunctionScan.) |
| 5284 |
*/ |
| 5285 |
if (rtfunc->funccolnames != NIL) |
| 5286 |
{ |
| 5287 |
functypclass = TYPEFUNC_RECORD; |
| 5288 |
rettupdesc = BuildDescFromLists(rtfunc->funccolnames, |
| 5289 |
rtfunc->funccoltypes, |
| 5290 |
rtfunc->funccoltypmods, |
| 5291 |
rtfunc->funccolcollations); |
| 5292 |
} |
| 5293 |
else |
| 5294 |
functypclass = get_expr_result_type((Node *) fexpr, NULL, &rettupdesc); |
| 5295 |
|
| 5296 |
/* |
| 5297 |
* The single command must be a plain SELECT. |
| 5298 |
*/ |
| 5299 |
if (!IsA(querytree, Query) || |
| 5300 |
querytree->commandType != CMD_SELECT) |
| 5301 |
goto fail; |
| 5302 |
|
| 5303 |
/* |
| 5304 |
* Make sure the function (still) returns what it's declared to. This |
| 5305 |
* will raise an error if wrong, but that's okay since the function would |
| 5306 |
* fail at runtime anyway. Note that check_sql_fn_retval will also insert |
| 5307 |
* coercions if needed to make the tlist expression(s) match the declared |
| 5308 |
* type of the function. We also ask it to insert dummy NULL columns for |
| 5309 |
* any dropped columns in rettupdesc, so that the elements of the modified |
| 5310 |
* tlist match up to the attribute numbers. |
| 5311 |
* |
| 5312 |
* If the function returns a composite type, don't inline unless the check |
| 5313 |
* shows it's returning a whole tuple result; otherwise what it's |
| 5314 |
* returning is a single composite column which is not what we need. |
| 5315 |
*/ |
| 5316 |
if (!check_sql_fn_retval(list_make1(querytree_list), |
| 5317 |
fexpr->funcresulttype, rettupdesc, |
| 5318 |
funcform->prokind, |
| 5319 |
true) && |
| 5320 |
(functypclass == TYPEFUNC_COMPOSITE || |
| 5321 |
functypclass == TYPEFUNC_COMPOSITE_DOMAIN || |
| 5322 |
functypclass == TYPEFUNC_RECORD)) |
| 5323 |
goto fail; /* reject not-whole-tuple-result cases */ |
| 5324 |
|
| 5325 |
/* |
| 5326 |
* check_sql_fn_retval might've inserted a projection step, but that's |
| 5327 |
* fine; just make sure we use the upper Query. |
| 5328 |
*/ |
| 5329 |
querytree = linitial_node(Query, querytree_list); |
| 5330 |
|
| 5331 |
/* |
| 5332 |
* Looks good --- substitute parameters into the query. |
| 5333 |
*/ |
| 5334 |
querytree = substitute_actual_srf_parameters(querytree, |
| 5335 |
funcform->pronargs, |
| 5336 |
fexpr->args); |
| 5337 |
|
| 5338 |
/* |
| 5339 |
* Copy the modified query out of the temporary memory context, and clean |
| 5340 |
* up. |
| 5341 |
*/ |
| 5342 |
MemoryContextSwitchTo(oldcxt); |
| 5343 |
|
| 5344 |
querytree = copyObject(querytree); |
| 5345 |
|
| 5346 |
MemoryContextDelete(mycxt); |
| 5347 |
error_context_stack = sqlerrcontext.previous; |
| 5348 |
ReleaseSysCache(func_tuple); |
| 5349 |
|
| 5350 |
/* |
| 5351 |
* We don't have to fix collations here because the upper query is already |
| 5352 |
* parsed, ie, the collations in the RTE are what count. |
| 5353 |
*/ |
| 5354 |
|
| 5355 |
/* |
| 5356 |
* Since there is now no trace of the function in the plan tree, we must |
| 5357 |
* explicitly record the plan's dependency on the function. |
| 5358 |
*/ |
| 5359 |
record_plan_function_dependency(root, func_oid); |
| 5360 |
|
| 5361 |
/* |
| 5362 |
* We must also notice if the inserted query adds a dependency on the |
| 5363 |
* calling role due to RLS quals. |
| 5364 |
*/ |
| 5365 |
if (querytree->hasRowSecurity) |
| 5366 |
root->glob->dependsOnRole = true; |
| 5367 |
|
| 5368 |
return querytree; |
| 5369 |
|
| 5370 |
/* Here if func is not inlinable: release temp memory and return NULL */ |
| 5371 |
fail: |
| 5372 |
MemoryContextSwitchTo(oldcxt); |
| 5373 |
MemoryContextDelete(mycxt); |
| 5374 |
error_context_stack = sqlerrcontext.previous; |
| 5375 |
ReleaseSysCache(func_tuple); |
| 5376 |
|
| 5377 |
return NULL; |
| 5378 |
} |
| 5379 |
|
| 5380 |
/* |
| 5381 |
* Replace Param nodes by appropriate actual parameters |
| 5382 |
* |
| 5383 |
* This is just enough different from substitute_actual_parameters() |
| 5384 |
* that it needs its own code. |
| 5385 |
*/ |
| 5386 |
static Query * |
| 5387 |
substitute_actual_srf_parameters(Query *expr, int nargs, List *args) |
| 5388 |
{ |
| 5389 |
substitute_actual_srf_parameters_context context; |
| 5390 |
|
| 5391 |
context.nargs = nargs; |
| 5392 |
context.args = args; |
| 5393 |
context.sublevels_up = 1; |
| 5394 |
|
| 5395 |
return query_tree_mutator(expr, |
| 5396 |
substitute_actual_srf_parameters_mutator, |
| 5397 |
&context, |
| 5398 |
0); |
| 5399 |
} |
| 5400 |
|
| 5401 |
static Node * |
| 5402 |
substitute_actual_srf_parameters_mutator(Node *node, |
| 5403 |
substitute_actual_srf_parameters_context *context) |
| 5404 |
{ |
| 5405 |
Node *result; |
| 5406 |
|
| 5407 |
if (node == NULL) |
| 5408 |
return NULL; |
| 5409 |
if (IsA(node, Query)) |
| 5410 |
{ |
| 5411 |
context->sublevels_up++; |
| 5412 |
result = (Node *) query_tree_mutator((Query *) node, |
| 5413 |
substitute_actual_srf_parameters_mutator, |
| 5414 |
context, |
| 5415 |
0); |
| 5416 |
context->sublevels_up--; |
| 5417 |
return result; |
| 5418 |
} |
| 5419 |
if (IsA(node, Param)) |
| 5420 |
{ |
| 5421 |
Param *param = (Param *) node; |
| 5422 |
|
| 5423 |
if (param->paramkind == PARAM_EXTERN) |
| 5424 |
{ |
| 5425 |
if (param->paramid <= 0 || param->paramid > context->nargs) |
| 5426 |
elog(ERROR, "invalid paramid: %d", param->paramid); |
| 5427 |
|
| 5428 |
/* |
| 5429 |
* Since the parameter is being inserted into a subquery, we must |
| 5430 |
* adjust levels. |
| 5431 |
*/ |
| 5432 |
result = copyObject(list_nth(context->args, param->paramid - 1)); |
| 5433 |
IncrementVarSublevelsUp(result, context->sublevels_up, 0); |
| 5434 |
return result; |
| 5435 |
} |
| 5436 |
} |
| 5437 |
return expression_tree_mutator(node, |
| 5438 |
substitute_actual_srf_parameters_mutator, |
| 5439 |
context); |
| 5440 |
} |
| 5441 |
|
| 5442 |
/* |
| 5443 |
* pull_paramids |
| 5444 |
* Returns a Bitmapset containing the paramids of all Params in 'expr'. |
| 5445 |
*/ |
| 5446 |
Bitmapset * |
| 5447 |
pull_paramids(Expr *expr) |
| 5448 |
{ |
| 5449 |
Bitmapset *result = NULL; |
| 5450 |
|
| 5451 |
(void) pull_paramids_walker((Node *) expr, &result); |
| 5452 |
|
| 5453 |
return result; |
| 5454 |
} |
| 5455 |
|
| 5456 |
static bool |
| 5457 |
pull_paramids_walker(Node *node, Bitmapset **context) |
| 5458 |
{ |
| 5459 |
if (node == NULL) |
| 5460 |
return false; |
| 5461 |
if (IsA(node, Param)) |
| 5462 |
{ |
| 5463 |
Param *param = (Param *) node; |
| 5464 |
|
| 5465 |
*context = bms_add_member(*context, param->paramid); |
| 5466 |
return false; |
| 5467 |
} |
| 5468 |
return expression_tree_walker(node, pull_paramids_walker, context); |
| 5469 |
} |
| 5470 |
|
| 5471 |
/* |
| 5472 |
* expression_has_grouping_conflict |
| 5473 |
* Detect whether 'expr' would distinguish rows that a grouping mechanism |
| 5474 |
* (GROUP BY, DISTINCT, DISTINCT ON, window PARTITION BY, or set operation) |
| 5475 |
* considers equal. |
| 5476 |
* |
| 5477 |
* The caller supplies a get_eqop callback (see clauses.h) so the same walker |
| 5478 |
* serves every grouping context. The callback identifies a grouping column by |
| 5479 |
* returning a valid eqop for its Var. A grouping column is safe to reference |
| 5480 |
* only if the reference yields the same result for every value the grouping |
| 5481 |
* treats as equal. Otherwise, pushing the clause past the grouping could |
| 5482 |
* discard rows that the grouping would have combined into a single group. |
| 5483 |
* |
| 5484 |
* The reference is provably safe only when the grouping column is a direct |
| 5485 |
* operand of a comparison that tests the grouping's own equality. Such an |
| 5486 |
* operand is rejected when the comparison's operator does not have equality |
| 5487 |
* semantics compatible with the grouping eqop, or, for a nondeterministic |
| 5488 |
* collation, when the comparison applies a collation other than the column's. |
| 5489 |
* |
| 5490 |
* For a nondeterministic collation, every other reference is rejected: a |
| 5491 |
* comparison under a different collation, and any function or operator over |
| 5492 |
* the column, because we cannot tell whether the function yields the same |
| 5493 |
* result for values the grouping treats as equal, and many do not. A column |
| 5494 |
* with a deterministic collation is not restricted this way. |
| 5495 |
* |
| 5496 |
* This leaves one case uncaught: with a deterministic collation, a function |
| 5497 |
* over the column can still feed a finer comparison than the direct-operand |
| 5498 |
* check sees, for example record_image_ops over a rebuilt record, or scale() |
| 5499 |
* over numeric where two equal values differ in scale. Catching it would |
| 5500 |
* require knowing that a type's equality is bitwise, which we do not test |
| 5501 |
* here. |
| 5502 |
* |
| 5503 |
* Returns true if any such conflict exists. |
| 5504 |
*/ |
| 5505 |
bool |
| 5506 |
expression_has_grouping_conflict(Node *expr, |
| 5507 |
grouping_eqop_callback get_eqop, |
| 5508 |
void *context) |
| 5509 |
{ |
| 5510 |
grouping_walker_ctx ctx; |
| 5511 |
|
| 5512 |
if (expr == NULL) |
| 5513 |
return false; |
| 5514 |
|
| 5515 |
ctx.get_eqop = get_eqop; |
| 5516 |
ctx.cb_context = context; |
| 5517 |
|
| 5518 |
return grouping_conflict_walker(expr, &ctx); |
| 5519 |
} |
| 5520 |
|
| 5521 |
/* |
| 5522 |
* Walker function for expression_has_grouping_conflict. |
| 5523 |
* |
| 5524 |
* A comparison node checks its direct operands with grouping_check_operand, |
| 5525 |
* which does not recurse into a grouping-column operand. A grouping column |
| 5526 |
* therefore reaches the Var branch only when it is referenced in some other |
| 5527 |
* way: wrapped in a function or other expression, used as the whole qual (a |
| 5528 |
* bare boolean column), or used as an operand of an operator that is not a |
| 5529 |
* btree/hash member and so is not treated as a comparison here. |
| 5530 |
* |
| 5531 |
* Comparison nodes are OpExpr/ScalarArrayOpExpr whose operator is a btree/hash |
| 5532 |
* member, and RowCompareExpr (one operator and collation per column). A |
| 5533 |
* simple CASE (CaseExpr with a non-NULL arg) is a comparison in disguise: |
| 5534 |
* parse analysis builds each WHEN as "OpExpr(CaseTestExpr op val)", with the |
| 5535 |
* CaseTestExpr standing in for the arg, so the arg is effectively an operand |
| 5536 |
* of each WHEN's comparison. Those WHEN operators are always the type-default |
| 5537 |
* "=", matching the grouping eqop, so only a collation conflict is possible |
| 5538 |
* there. |
| 5539 |
*/ |
| 5540 |
static bool |
| 5541 |
grouping_conflict_walker(Node *node, grouping_walker_ctx *ctx) |
| 5542 |
{ |
| 5543 |
if (node == NULL) |
| 5544 |
return false; |
| 5545 |
|
| 5546 |
if (IsA(node, Var)) |
| 5547 |
{ |
| 5548 |
Var *var = (Var *) node; |
| 5549 |
|
| 5550 |
/* |
| 5551 |
* A grouping column reaches here when it was not handled as a direct |
| 5552 |
* operand by a comparison node above (see the function header). That |
| 5553 |
* is safe for a deterministic collation, but not for a |
| 5554 |
* nondeterministic one, where the reference may distinguish values |
| 5555 |
* the grouping considers equal. A bare boolean qual is safe too: |
| 5556 |
* boolean is not collatable, so it takes the deterministic path here. |
| 5557 |
*/ |
| 5558 |
if (OidIsValid(ctx->get_eqop(var, ctx->cb_context)) && |
| 5559 |
OidIsValid(var->varcollid) && |
| 5560 |
!get_collation_isdeterministic(var->varcollid)) |
| 5561 |
return true; |
| 5562 |
return false; |
| 5563 |
} |
| 5564 |
else if (IsA(node, OpExpr)) |
| 5565 |
{ |
| 5566 |
OpExpr *opexpr = (OpExpr *) node; |
| 5567 |
|
| 5568 |
if (op_is_safe_index_member(opexpr->opno)) |
| 5569 |
return grouping_check_operands(opexpr->opno, opexpr->inputcollid, |
| 5570 |
opexpr->args, ctx); |
| 5571 |
/* fall through */ |
| 5572 |
} |
| 5573 |
else if (IsA(node, ScalarArrayOpExpr)) |
| 5574 |
{ |
| 5575 |
ScalarArrayOpExpr *saop = (ScalarArrayOpExpr *) node; |
| 5576 |
|
| 5577 |
if (op_is_safe_index_member(saop->opno)) |
| 5578 |
return grouping_check_operands(saop->opno, saop->inputcollid, |
| 5579 |
saop->args, ctx); |
| 5580 |
/* fall through */ |
| 5581 |
} |
| 5582 |
else if (IsA(node, RowCompareExpr)) |
| 5583 |
{ |
| 5584 |
RowCompareExpr *rcexpr = (RowCompareExpr *) node; |
| 5585 |
ListCell *lc_l; |
| 5586 |
ListCell *lc_r; |
| 5587 |
ListCell *lc_o; |
| 5588 |
ListCell *lc_c; |
| 5589 |
|
| 5590 |
/* Each column is compared under its own operator and inputcollid. */ |
| 5591 |
forfour(lc_l, rcexpr->largs, |
| 5592 |
lc_r, rcexpr->rargs, |
| 5593 |
lc_o, rcexpr->opnos, |
| 5594 |
lc_c, rcexpr->inputcollids) |
| 5595 |
{ |
| 5596 |
Oid opno = lfirst_oid(lc_o); |
| 5597 |
Oid collid = lfirst_oid(lc_c); |
| 5598 |
|
| 5599 |
if (grouping_check_operand((Node *) lfirst(lc_l), opno, collid, ctx) || |
| 5600 |
grouping_check_operand((Node *) lfirst(lc_r), opno, collid, ctx)) |
| 5601 |
return true; |
| 5602 |
} |
| 5603 |
return false; |
| 5604 |
} |
| 5605 |
else if (IsA(node, CaseExpr) && ((CaseExpr *) node)->arg != NULL) |
| 5606 |
{ |
| 5607 |
CaseExpr *cexpr = (CaseExpr *) node; |
| 5608 |
Node *arg = (Node *) cexpr->arg; |
| 5609 |
|
| 5610 |
/* Look through RelabelType to find a direct Var arg. */ |
| 5611 |
while (arg && IsA(arg, RelabelType)) |
| 5612 |
arg = (Node *) ((RelabelType *) arg)->arg; |
| 5613 |
|
| 5614 |
if (arg && IsA(arg, Var)) |
| 5615 |
{ |
| 5616 |
Var *var = (Var *) arg; |
| 5617 |
|
| 5618 |
/* |
| 5619 |
* The arg is a grouping column compared by every WHEN. For a |
| 5620 |
* nondeterministic collation, reject if any WHEN applies a |
| 5621 |
* different collation. |
| 5622 |
*/ |
| 5623 |
if (OidIsValid(ctx->get_eqop(var, ctx->cb_context)) && |
| 5624 |
OidIsValid(var->varcollid) && |
| 5625 |
!get_collation_isdeterministic(var->varcollid)) |
| 5626 |
{ |
| 5627 |
foreach_node(CaseWhen, cw, cexpr->args) |
| 5628 |
{ |
| 5629 |
Oid collid = exprInputCollation((Node *) cw->expr); |
| 5630 |
|
| 5631 |
if (OidIsValid(collid) && collid != var->varcollid) |
| 5632 |
return true; |
| 5633 |
} |
| 5634 |
} |
| 5635 |
} |
| 5636 |
else if (grouping_conflict_walker((Node *) cexpr->arg, ctx)) |
| 5637 |
{ |
| 5638 |
/* arg is a complex expression; walked as a non-operand */ |
| 5639 |
return true; |
| 5640 |
} |
| 5641 |
|
| 5642 |
/* |
| 5643 |
* Walk the WHEN conditions, their results, and the default result as |
| 5644 |
* non-operands. The WHEN conditions hold a CaseTestExpr in place of |
| 5645 |
* the arg, so they contribute no grouping operand of their own, but |
| 5646 |
* the condition expression or the substitution result may reference |
| 5647 |
* another grouping column. |
| 5648 |
*/ |
| 5649 |
foreach_node(CaseWhen, cw, cexpr->args) |
| 5650 |
{ |
| 5651 |
if (grouping_conflict_walker((Node *) cw->expr, ctx) || |
| 5652 |
grouping_conflict_walker((Node *) cw->result, ctx)) |
| 5653 |
return true; |
| 5654 |
} |
| 5655 |
return grouping_conflict_walker((Node *) cexpr->defresult, ctx); |
| 5656 |
} |
| 5657 |
|
| 5658 |
return expression_tree_walker(node, grouping_conflict_walker, ctx); |
| 5659 |
} |
| 5660 |
|
| 5661 |
/* |
| 5662 |
* grouping_check_operands |
| 5663 |
* Check every argument of a comparison node as a direct operand of the |
| 5664 |
* comparison's operator 'opno' and collation 'inputcollid'. |
| 5665 |
*/ |
| 5666 |
static bool |
| 5667 |
grouping_check_operands(Oid opno, Oid inputcollid, List *args, |
| 5668 |
grouping_walker_ctx *ctx) |
| 5669 |
{ |
| 5670 |
ListCell *lc; |
| 5671 |
|
| 5672 |
foreach(lc, args) |
| 5673 |
{ |
| 5674 |
if (grouping_check_operand((Node *) lfirst(lc), opno, inputcollid, ctx)) |
| 5675 |
return true; |
| 5676 |
} |
| 5677 |
return false; |
| 5678 |
} |
| 5679 |
|
| 5680 |
/* |
| 5681 |
* grouping_check_operand |
| 5682 |
* Handle one operand 'arg' of a comparison with operator 'opno' and |
| 5683 |
* collation 'inputcollid'. |
| 5684 |
* |
| 5685 |
* If 'arg' is a grouping column (after looking through RelabelType), verify |
| 5686 |
* that comparison's operator has equality semantics compatible with the |
| 5687 |
* grouping eqop and, for a nondeterministic collation, that it uses the same |
| 5688 |
* collation; such a direct operand is then fully handled and is not recursed |
| 5689 |
* into. Any other operand is walked normally, so a grouping column buried |
| 5690 |
* inside it is seen as a non-operand reference. |
| 5691 |
*/ |
| 5692 |
static bool |
| 5693 |
grouping_check_operand(Node *arg, Oid opno, Oid inputcollid, |
| 5694 |
grouping_walker_ctx *ctx) |
| 5695 |
{ |
| 5696 |
Node *node = arg; |
| 5697 |
|
| 5698 |
while (node && IsA(node, RelabelType)) |
| 5699 |
node = (Node *) ((RelabelType *) node)->arg; |
| 5700 |
|
| 5701 |
if (node && IsA(node, Var)) |
| 5702 |
{ |
| 5703 |
Var *var = (Var *) node; |
| 5704 |
Oid grouping_eqop = ctx->get_eqop(var, ctx->cb_context); |
| 5705 |
|
| 5706 |
if (OidIsValid(grouping_eqop)) |
| 5707 |
{ |
| 5708 |
/* incompatible equality semantics */ |
| 5709 |
if (!equality_ops_are_compatible(opno, grouping_eqop)) |
| 5710 |
return true; |
| 5711 |
/* nondeterministic collation compared under a different collation */ |
| 5712 |
if (OidIsValid(var->varcollid) && |
| 5713 |
!get_collation_isdeterministic(var->varcollid) && |
| 5714 |
inputcollid != var->varcollid) |
| 5715 |
return true; |
| 5716 |
} |
| 5717 |
return false; /* direct operand handled; do not recurse */ |
| 5718 |
} |
| 5719 |
|
| 5720 |
return grouping_conflict_walker(arg, ctx); |
| 5721 |
} |
| 5722 |
|
| 5723 |
/* |
| 5724 |
* Build ScalarArrayOpExpr on top of 'exprs.' 'haveNonConst' indicates |
| 5725 |
* whether at least one of the expressions is not Const. When it's false, |
| 5726 |
* the array constant is built directly; otherwise, we have to build a child |
| 5727 |
* ArrayExpr. The 'exprs' list gets freed if not directly used in the output |
| 5728 |
* expression tree. |
| 5729 |
*/ |
| 5730 |
ScalarArrayOpExpr * |
| 5731 |
make_SAOP_expr(Oid oper, Node *leftexpr, Oid coltype, Oid arraycollid, |
| 5732 |
Oid inputcollid, List *exprs, bool haveNonConst) |
| 5733 |
{ |
| 5734 |
Node *arrayNode = NULL; |
| 5735 |
ScalarArrayOpExpr *saopexpr = NULL; |
| 5736 |
Oid arraytype = get_array_type(coltype); |
| 5737 |
|
| 5738 |
if (!OidIsValid(arraytype)) |
| 5739 |
return NULL; |
| 5740 |
|
| 5741 |
/* |
| 5742 |
* Assemble an array from the list of constants. It seems more profitable |
| 5743 |
* to build a const array. But in the presence of other nodes, we don't |
| 5744 |
* have a specific value here and must employ an ArrayExpr instead. |
| 5745 |
*/ |
| 5746 |
if (haveNonConst) |
| 5747 |
{ |
| 5748 |
ArrayExpr *arrayExpr = makeNode(ArrayExpr); |
| 5749 |
|
| 5750 |
/* array_collid will be set by parse_collate.c */ |
| 5751 |
arrayExpr->element_typeid = coltype; |
| 5752 |
arrayExpr->array_typeid = arraytype; |
| 5753 |
arrayExpr->multidims = false; |
| 5754 |
arrayExpr->elements = exprs; |
| 5755 |
arrayExpr->location = -1; |
| 5756 |
|
| 5757 |
arrayNode = (Node *) arrayExpr; |
| 5758 |
} |
| 5759 |
else |
| 5760 |
{ |
| 5761 |
int16 typlen; |
| 5762 |
bool typbyval; |
| 5763 |
char typalign; |
| 5764 |
Datum *elems; |
| 5765 |
bool *nulls; |
| 5766 |
int i = 0; |
| 5767 |
ArrayType *arrayConst; |
| 5768 |
int dims[1] = {list_length(exprs)}; |
| 5769 |
int lbs[1] = {1}; |
| 5770 |
|
| 5771 |
get_typlenbyvalalign(coltype, &typlen, &typbyval, &typalign); |
| 5772 |
|
| 5773 |
elems = (Datum *) palloc(sizeof(Datum) * list_length(exprs)); |
| 5774 |
nulls = (bool *) palloc(sizeof(bool) * list_length(exprs)); |
| 5775 |
foreach_node(Const, value, exprs) |
| 5776 |
{ |
| 5777 |
elems[i] = value->constvalue; |
| 5778 |
nulls[i++] = value->constisnull; |
| 5779 |
} |
| 5780 |
|
| 5781 |
arrayConst = construct_md_array(elems, nulls, 1, dims, lbs, |
| 5782 |
coltype, typlen, typbyval, typalign); |
| 5783 |
arrayNode = (Node *) makeConst(arraytype, -1, arraycollid, |
| 5784 |
-1, PointerGetDatum(arrayConst), |
| 5785 |
false, false); |
| 5786 |
|
| 5787 |
pfree(elems); |
| 5788 |
pfree(nulls); |
| 5789 |
list_free(exprs); |
| 5790 |
} |
| 5791 |
|
| 5792 |
/* Build the SAOP expression node */ |
| 5793 |
saopexpr = makeNode(ScalarArrayOpExpr); |
| 5794 |
saopexpr->opno = oper; |
| 5795 |
saopexpr->opfuncid = get_opcode(oper); |
| 5796 |
saopexpr->hashfuncid = InvalidOid; |
| 5797 |
saopexpr->negfuncid = InvalidOid; |
| 5798 |
saopexpr->useOr = true; |
| 5799 |
saopexpr->inputcollid = inputcollid; |
| 5800 |
saopexpr->args = list_make2(leftexpr, arrayNode); |
| 5801 |
saopexpr->location = -1; |
| 5802 |
|
| 5803 |
return saopexpr; |
| 5804 |
} |