| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* tlist.c |
| 4 |
* Target list manipulation routines |
| 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/tlist.c |
| 12 |
* |
| 13 |
*------------------------------------------------------------------------- |
| 14 |
*/ |
| 15 |
#include "postgres.h" |
| 16 |
|
| 17 |
#include "nodes/makefuncs.h" |
| 18 |
#include "nodes/nodeFuncs.h" |
| 19 |
#include "optimizer/cost.h" |
| 20 |
#include "optimizer/optimizer.h" |
| 21 |
#include "optimizer/tlist.h" |
| 22 |
#include "rewrite/rewriteManip.h" |
| 23 |
|
| 24 |
|
| 25 |
/* |
| 26 |
* Test if an expression node represents a SRF call. Beware multiple eval! |
| 27 |
* |
| 28 |
* Please note that this is only meant for use in split_pathtarget_at_srfs(); |
| 29 |
* if you use it anywhere else, your code is almost certainly wrong for SRFs |
| 30 |
* nested within expressions. Use expression_returns_set() instead. |
| 31 |
*/ |
| 32 |
#define IS_SRF_CALL(node) \ |
| 33 |
((IsA(node, FuncExpr) && ((FuncExpr *) (node))->funcretset) || \ |
| 34 |
(IsA(node, OpExpr) && ((OpExpr *) (node))->opretset)) |
| 35 |
|
| 36 |
/* |
| 37 |
* Data structures for split_pathtarget_at_srfs(). To preserve the identity |
| 38 |
* of sortgroupref items even if they are textually equal(), what we track is |
| 39 |
* not just bare expressions but expressions plus their sortgroupref indexes. |
| 40 |
*/ |
| 41 |
typedef struct |
| 42 |
{ |
| 43 |
Node *expr; /* some subexpression of a PathTarget */ |
| 44 |
Index sortgroupref; /* its sortgroupref, or 0 if none */ |
| 45 |
} split_pathtarget_item; |
| 46 |
|
| 47 |
typedef struct |
| 48 |
{ |
| 49 |
PlannerInfo *root; |
| 50 |
bool is_grouping_target; /* true if processing grouping target */ |
| 51 |
/* This is a List of bare expressions: */ |
| 52 |
List *input_target_exprs; /* exprs available from input */ |
| 53 |
/* These are Lists of Lists of split_pathtarget_items: */ |
| 54 |
List *level_srfs; /* SRF exprs to evaluate at each level */ |
| 55 |
List *level_input_vars; /* input vars needed at each level */ |
| 56 |
List *level_input_srfs; /* input SRFs needed at each level */ |
| 57 |
/* These are Lists of split_pathtarget_items: */ |
| 58 |
List *current_input_vars; /* vars needed in current subexpr */ |
| 59 |
List *current_input_srfs; /* SRFs needed in current subexpr */ |
| 60 |
/* Auxiliary data for current split_pathtarget_walker traversal: */ |
| 61 |
int current_depth; /* max SRF depth in current subexpr */ |
| 62 |
Index current_sgref; /* current subexpr's sortgroupref, or 0 */ |
| 63 |
} split_pathtarget_context; |
| 64 |
|
| 65 |
static void split_pathtarget_at_srfs_extended(PlannerInfo *root, |
| 66 |
PathTarget *target, |
| 67 |
PathTarget *input_target, |
| 68 |
List **targets, |
| 69 |
List **targets_contain_srfs, |
| 70 |
bool is_grouping_target); |
| 71 |
static bool split_pathtarget_walker(Node *node, |
| 72 |
split_pathtarget_context *context); |
| 73 |
static void add_sp_item_to_pathtarget(PathTarget *target, |
| 74 |
split_pathtarget_item *item); |
| 75 |
static void add_sp_items_to_pathtarget(PathTarget *target, List *items); |
| 76 |
|
| 77 |
|
| 78 |
/***************************************************************************** |
| 79 |
* Target list creation and searching utilities |
| 80 |
*****************************************************************************/ |
| 81 |
|
| 82 |
/* |
| 83 |
* tlist_member |
| 84 |
* Finds the (first) member of the given tlist whose expression is |
| 85 |
* equal() to the given expression. Result is NULL if no such member. |
| 86 |
*/ |
| 87 |
TargetEntry * |
| 88 |
tlist_member(Expr *node, List *targetlist) |
| 89 |
{ |
| 90 |
ListCell *temp; |
| 91 |
|
| 92 |
foreach(temp, targetlist) |
| 93 |
{ |
| 94 |
TargetEntry *tlentry = (TargetEntry *) lfirst(temp); |
| 95 |
|
| 96 |
if (equal(node, tlentry->expr)) |
| 97 |
return tlentry; |
| 98 |
} |
| 99 |
return NULL; |
| 100 |
} |
| 101 |
|
| 102 |
/* |
| 103 |
* tlist_member_match_var |
| 104 |
* Same as above, except that we match the provided Var on the basis |
| 105 |
* of varno/varattno/varlevelsup/vartype only, rather than full equal(). |
| 106 |
* |
| 107 |
* This is needed in some cases where we can't be sure of an exact typmod |
| 108 |
* match. For safety, though, we insist on vartype match. |
| 109 |
*/ |
| 110 |
static TargetEntry * |
| 111 |
tlist_member_match_var(Var *var, List *targetlist) |
| 112 |
{ |
| 113 |
ListCell *temp; |
| 114 |
|
| 115 |
foreach(temp, targetlist) |
| 116 |
{ |
| 117 |
TargetEntry *tlentry = (TargetEntry *) lfirst(temp); |
| 118 |
Var *tlvar = (Var *) tlentry->expr; |
| 119 |
|
| 120 |
if (!tlvar || !IsA(tlvar, Var)) |
| 121 |
continue; |
| 122 |
if (var->varno == tlvar->varno && |
| 123 |
var->varattno == tlvar->varattno && |
| 124 |
var->varlevelsup == tlvar->varlevelsup && |
| 125 |
var->vartype == tlvar->vartype) |
| 126 |
return tlentry; |
| 127 |
} |
| 128 |
return NULL; |
| 129 |
} |
| 130 |
|
| 131 |
/* |
| 132 |
* add_to_flat_tlist |
| 133 |
* Add more items to a flattened tlist (if they're not already in it) |
| 134 |
* |
| 135 |
* 'tlist' is the flattened tlist |
| 136 |
* 'exprs' is a list of expressions (usually, but not necessarily, Vars) |
| 137 |
* |
| 138 |
* Returns the extended tlist. |
| 139 |
*/ |
| 140 |
List * |
| 141 |
add_to_flat_tlist(List *tlist, List *exprs) |
| 142 |
{ |
| 143 |
int next_resno = list_length(tlist) + 1; |
| 144 |
ListCell *lc; |
| 145 |
|
| 146 |
foreach(lc, exprs) |
| 147 |
{ |
| 148 |
Expr *expr = (Expr *) lfirst(lc); |
| 149 |
|
| 150 |
if (!tlist_member(expr, tlist)) |
| 151 |
{ |
| 152 |
TargetEntry *tle; |
| 153 |
|
| 154 |
tle = makeTargetEntry(copyObject(expr), /* copy needed?? */ |
| 155 |
next_resno++, |
| 156 |
NULL, |
| 157 |
false); |
| 158 |
tlist = lappend(tlist, tle); |
| 159 |
} |
| 160 |
} |
| 161 |
return tlist; |
| 162 |
} |
| 163 |
|
| 164 |
|
| 165 |
/* |
| 166 |
* get_tlist_exprs |
| 167 |
* Get just the expression subtrees of a tlist |
| 168 |
* |
| 169 |
* Resjunk columns are ignored unless includeJunk is true |
| 170 |
*/ |
| 171 |
List * |
| 172 |
get_tlist_exprs(List *tlist, bool includeJunk) |
| 173 |
{ |
| 174 |
List *result = NIL; |
| 175 |
ListCell *l; |
| 176 |
|
| 177 |
foreach(l, tlist) |
| 178 |
{ |
| 179 |
TargetEntry *tle = (TargetEntry *) lfirst(l); |
| 180 |
|
| 181 |
if (tle->resjunk && !includeJunk) |
| 182 |
continue; |
| 183 |
|
| 184 |
result = lappend(result, tle->expr); |
| 185 |
} |
| 186 |
return result; |
| 187 |
} |
| 188 |
|
| 189 |
|
| 190 |
/* |
| 191 |
* count_nonjunk_tlist_entries |
| 192 |
* What it says ... |
| 193 |
*/ |
| 194 |
int |
| 195 |
count_nonjunk_tlist_entries(List *tlist) |
| 196 |
{ |
| 197 |
int len = 0; |
| 198 |
ListCell *l; |
| 199 |
|
| 200 |
foreach(l, tlist) |
| 201 |
{ |
| 202 |
TargetEntry *tle = (TargetEntry *) lfirst(l); |
| 203 |
|
| 204 |
if (!tle->resjunk) |
| 205 |
len++; |
| 206 |
} |
| 207 |
return len; |
| 208 |
} |
| 209 |
|
| 210 |
|
| 211 |
/* |
| 212 |
* tlist_same_exprs |
| 213 |
* Check whether two target lists contain the same expressions |
| 214 |
* |
| 215 |
* Note: this function is used to decide whether it's safe to jam a new tlist |
| 216 |
* into a non-projection-capable plan node. Obviously we can't do that unless |
| 217 |
* the node's tlist shows it already returns the column values we want. |
| 218 |
* However, we can ignore the TargetEntry attributes resname, ressortgroupref, |
| 219 |
* resorigtbl, resorigcol, and resjunk, because those are only labelings that |
| 220 |
* don't affect the row values computed by the node. (Moreover, if we didn't |
| 221 |
* ignore them, we'd frequently fail to make the desired optimization, since |
| 222 |
* the planner tends to not bother to make resname etc. valid in intermediate |
| 223 |
* plan nodes.) Note that on success, the caller must still jam the desired |
| 224 |
* tlist into the plan node, else it won't have the desired labeling fields. |
| 225 |
*/ |
| 226 |
bool |
| 227 |
tlist_same_exprs(List *tlist1, List *tlist2) |
| 228 |
{ |
| 229 |
ListCell *lc1, |
| 230 |
*lc2; |
| 231 |
|
| 232 |
if (list_length(tlist1) != list_length(tlist2)) |
| 233 |
return false; /* not same length, so can't match */ |
| 234 |
|
| 235 |
forboth(lc1, tlist1, lc2, tlist2) |
| 236 |
{ |
| 237 |
TargetEntry *tle1 = (TargetEntry *) lfirst(lc1); |
| 238 |
TargetEntry *tle2 = (TargetEntry *) lfirst(lc2); |
| 239 |
|
| 240 |
if (!equal(tle1->expr, tle2->expr)) |
| 241 |
return false; |
| 242 |
} |
| 243 |
|
| 244 |
return true; |
| 245 |
} |
| 246 |
|
| 247 |
|
| 248 |
/* |
| 249 |
* Does tlist have same output datatypes as listed in colTypes? |
| 250 |
* |
| 251 |
* Resjunk columns are ignored if junkOK is true; otherwise presence of |
| 252 |
* a resjunk column will always cause a 'false' result. |
| 253 |
* |
| 254 |
* Note: currently no callers care about comparing typmods. |
| 255 |
*/ |
| 256 |
bool |
| 257 |
tlist_same_datatypes(List *tlist, List *colTypes, bool junkOK) |
| 258 |
{ |
| 259 |
ListCell *l; |
| 260 |
ListCell *curColType = list_head(colTypes); |
| 261 |
|
| 262 |
foreach(l, tlist) |
| 263 |
{ |
| 264 |
TargetEntry *tle = (TargetEntry *) lfirst(l); |
| 265 |
|
| 266 |
if (tle->resjunk) |
| 267 |
{ |
| 268 |
if (!junkOK) |
| 269 |
return false; |
| 270 |
} |
| 271 |
else |
| 272 |
{ |
| 273 |
if (curColType == NULL) |
| 274 |
return false; /* tlist longer than colTypes */ |
| 275 |
if (exprType((Node *) tle->expr) != lfirst_oid(curColType)) |
| 276 |
return false; |
| 277 |
curColType = lnext(colTypes, curColType); |
| 278 |
} |
| 279 |
} |
| 280 |
if (curColType != NULL) |
| 281 |
return false; /* tlist shorter than colTypes */ |
| 282 |
return true; |
| 283 |
} |
| 284 |
|
| 285 |
/* |
| 286 |
* Does tlist have same exposed collations as listed in colCollations? |
| 287 |
* |
| 288 |
* Identical logic to the above, but for collations. |
| 289 |
*/ |
| 290 |
bool |
| 291 |
tlist_same_collations(List *tlist, List *colCollations, bool junkOK) |
| 292 |
{ |
| 293 |
ListCell *l; |
| 294 |
ListCell *curColColl = list_head(colCollations); |
| 295 |
|
| 296 |
foreach(l, tlist) |
| 297 |
{ |
| 298 |
TargetEntry *tle = (TargetEntry *) lfirst(l); |
| 299 |
|
| 300 |
if (tle->resjunk) |
| 301 |
{ |
| 302 |
if (!junkOK) |
| 303 |
return false; |
| 304 |
} |
| 305 |
else |
| 306 |
{ |
| 307 |
if (curColColl == NULL) |
| 308 |
return false; /* tlist longer than colCollations */ |
| 309 |
if (exprCollation((Node *) tle->expr) != lfirst_oid(curColColl)) |
| 310 |
return false; |
| 311 |
curColColl = lnext(colCollations, curColColl); |
| 312 |
} |
| 313 |
} |
| 314 |
if (curColColl != NULL) |
| 315 |
return false; /* tlist shorter than colCollations */ |
| 316 |
return true; |
| 317 |
} |
| 318 |
|
| 319 |
/* |
| 320 |
* apply_tlist_labeling |
| 321 |
* Apply the TargetEntry labeling attributes of src_tlist to dest_tlist |
| 322 |
* |
| 323 |
* This is useful for reattaching column names etc to a plan's final output |
| 324 |
* targetlist. |
| 325 |
*/ |
| 326 |
void |
| 327 |
apply_tlist_labeling(List *dest_tlist, List *src_tlist) |
| 328 |
{ |
| 329 |
ListCell *ld, |
| 330 |
*ls; |
| 331 |
|
| 332 |
Assert(list_length(dest_tlist) == list_length(src_tlist)); |
| 333 |
forboth(ld, dest_tlist, ls, src_tlist) |
| 334 |
{ |
| 335 |
TargetEntry *dest_tle = (TargetEntry *) lfirst(ld); |
| 336 |
TargetEntry *src_tle = (TargetEntry *) lfirst(ls); |
| 337 |
|
| 338 |
Assert(dest_tle->resno == src_tle->resno); |
| 339 |
dest_tle->resname = src_tle->resname; |
| 340 |
dest_tle->ressortgroupref = src_tle->ressortgroupref; |
| 341 |
dest_tle->resorigtbl = src_tle->resorigtbl; |
| 342 |
dest_tle->resorigcol = src_tle->resorigcol; |
| 343 |
dest_tle->resjunk = src_tle->resjunk; |
| 344 |
} |
| 345 |
} |
| 346 |
|
| 347 |
|
| 348 |
/* |
| 349 |
* get_sortgroupref_tle |
| 350 |
* Find the targetlist entry matching the given SortGroupRef index, |
| 351 |
* and return it. |
| 352 |
*/ |
| 353 |
TargetEntry * |
| 354 |
get_sortgroupref_tle(Index sortref, List *targetList) |
| 355 |
{ |
| 356 |
ListCell *l; |
| 357 |
|
| 358 |
foreach(l, targetList) |
| 359 |
{ |
| 360 |
TargetEntry *tle = (TargetEntry *) lfirst(l); |
| 361 |
|
| 362 |
if (tle->ressortgroupref == sortref) |
| 363 |
return tle; |
| 364 |
} |
| 365 |
|
| 366 |
elog(ERROR, "ORDER/GROUP BY expression not found in targetlist"); |
| 367 |
return NULL; /* keep compiler quiet */ |
| 368 |
} |
| 369 |
|
| 370 |
/* |
| 371 |
* get_sortgroupclause_tle |
| 372 |
* Find the targetlist entry matching the given SortGroupClause |
| 373 |
* by ressortgroupref, and return it. |
| 374 |
*/ |
| 375 |
TargetEntry * |
| 376 |
get_sortgroupclause_tle(SortGroupClause *sgClause, |
| 377 |
List *targetList) |
| 378 |
{ |
| 379 |
return get_sortgroupref_tle(sgClause->tleSortGroupRef, targetList); |
| 380 |
} |
| 381 |
|
| 382 |
/* |
| 383 |
* get_sortgroupclause_expr |
| 384 |
* Find the targetlist entry matching the given SortGroupClause |
| 385 |
* by ressortgroupref, and return its expression. |
| 386 |
*/ |
| 387 |
Node * |
| 388 |
get_sortgroupclause_expr(SortGroupClause *sgClause, List *targetList) |
| 389 |
{ |
| 390 |
TargetEntry *tle = get_sortgroupclause_tle(sgClause, targetList); |
| 391 |
|
| 392 |
return (Node *) tle->expr; |
| 393 |
} |
| 394 |
|
| 395 |
/* |
| 396 |
* get_sortgrouplist_exprs |
| 397 |
* Given a list of SortGroupClauses, build a list |
| 398 |
* of the referenced targetlist expressions. |
| 399 |
*/ |
| 400 |
List * |
| 401 |
get_sortgrouplist_exprs(List *sgClauses, List *targetList) |
| 402 |
{ |
| 403 |
List *result = NIL; |
| 404 |
ListCell *l; |
| 405 |
|
| 406 |
foreach(l, sgClauses) |
| 407 |
{ |
| 408 |
SortGroupClause *sortcl = (SortGroupClause *) lfirst(l); |
| 409 |
Node *sortexpr; |
| 410 |
|
| 411 |
sortexpr = get_sortgroupclause_expr(sortcl, targetList); |
| 412 |
result = lappend(result, sortexpr); |
| 413 |
} |
| 414 |
return result; |
| 415 |
} |
| 416 |
|
| 417 |
|
| 418 |
/***************************************************************************** |
| 419 |
* Functions to extract data from a list of SortGroupClauses |
| 420 |
* |
| 421 |
* These don't really belong in tlist.c, but they are sort of related to the |
| 422 |
* functions just above, and they don't seem to deserve their own file. |
| 423 |
*****************************************************************************/ |
| 424 |
|
| 425 |
/* |
| 426 |
* get_sortgroupref_clause |
| 427 |
* Find the SortGroupClause matching the given SortGroupRef index, |
| 428 |
* and return it. |
| 429 |
*/ |
| 430 |
SortGroupClause * |
| 431 |
get_sortgroupref_clause(Index sortref, List *clauses) |
| 432 |
{ |
| 433 |
ListCell *l; |
| 434 |
|
| 435 |
foreach(l, clauses) |
| 436 |
{ |
| 437 |
SortGroupClause *cl = (SortGroupClause *) lfirst(l); |
| 438 |
|
| 439 |
if (cl->tleSortGroupRef == sortref) |
| 440 |
return cl; |
| 441 |
} |
| 442 |
|
| 443 |
elog(ERROR, "ORDER/GROUP BY expression not found in list"); |
| 444 |
return NULL; /* keep compiler quiet */ |
| 445 |
} |
| 446 |
|
| 447 |
/* |
| 448 |
* get_sortgroupref_clause_noerr |
| 449 |
* As above, but return NULL rather than throwing an error if not found. |
| 450 |
*/ |
| 451 |
SortGroupClause * |
| 452 |
get_sortgroupref_clause_noerr(Index sortref, List *clauses) |
| 453 |
{ |
| 454 |
ListCell *l; |
| 455 |
|
| 456 |
foreach(l, clauses) |
| 457 |
{ |
| 458 |
SortGroupClause *cl = (SortGroupClause *) lfirst(l); |
| 459 |
|
| 460 |
if (cl->tleSortGroupRef == sortref) |
| 461 |
return cl; |
| 462 |
} |
| 463 |
|
| 464 |
return NULL; |
| 465 |
} |
| 466 |
|
| 467 |
/* |
| 468 |
* extract_grouping_ops - make an array of the equality operator OIDs |
| 469 |
* for a SortGroupClause list |
| 470 |
*/ |
| 471 |
Oid * |
| 472 |
extract_grouping_ops(List *groupClause) |
| 473 |
{ |
| 474 |
int numCols = list_length(groupClause); |
| 475 |
int colno = 0; |
| 476 |
Oid *groupOperators; |
| 477 |
ListCell *glitem; |
| 478 |
|
| 479 |
groupOperators = (Oid *) palloc(sizeof(Oid) * numCols); |
| 480 |
|
| 481 |
foreach(glitem, groupClause) |
| 482 |
{ |
| 483 |
SortGroupClause *groupcl = (SortGroupClause *) lfirst(glitem); |
| 484 |
|
| 485 |
groupOperators[colno] = groupcl->eqop; |
| 486 |
Assert(OidIsValid(groupOperators[colno])); |
| 487 |
colno++; |
| 488 |
} |
| 489 |
|
| 490 |
return groupOperators; |
| 491 |
} |
| 492 |
|
| 493 |
/* |
| 494 |
* extract_grouping_collations - make an array of the grouping column collations |
| 495 |
* for a SortGroupClause list |
| 496 |
*/ |
| 497 |
Oid * |
| 498 |
extract_grouping_collations(List *groupClause, List *tlist) |
| 499 |
{ |
| 500 |
int numCols = list_length(groupClause); |
| 501 |
int colno = 0; |
| 502 |
Oid *grpCollations; |
| 503 |
ListCell *glitem; |
| 504 |
|
| 505 |
grpCollations = (Oid *) palloc(sizeof(Oid) * numCols); |
| 506 |
|
| 507 |
foreach(glitem, groupClause) |
| 508 |
{ |
| 509 |
SortGroupClause *groupcl = (SortGroupClause *) lfirst(glitem); |
| 510 |
TargetEntry *tle = get_sortgroupclause_tle(groupcl, tlist); |
| 511 |
|
| 512 |
grpCollations[colno++] = exprCollation((Node *) tle->expr); |
| 513 |
} |
| 514 |
|
| 515 |
return grpCollations; |
| 516 |
} |
| 517 |
|
| 518 |
/* |
| 519 |
* extract_grouping_cols - make an array of the grouping column resnos |
| 520 |
* for a SortGroupClause list |
| 521 |
*/ |
| 522 |
AttrNumber * |
| 523 |
extract_grouping_cols(List *groupClause, List *tlist) |
| 524 |
{ |
| 525 |
AttrNumber *grpColIdx; |
| 526 |
int numCols = list_length(groupClause); |
| 527 |
int colno = 0; |
| 528 |
ListCell *glitem; |
| 529 |
|
| 530 |
grpColIdx = (AttrNumber *) palloc(sizeof(AttrNumber) * numCols); |
| 531 |
|
| 532 |
foreach(glitem, groupClause) |
| 533 |
{ |
| 534 |
SortGroupClause *groupcl = (SortGroupClause *) lfirst(glitem); |
| 535 |
TargetEntry *tle = get_sortgroupclause_tle(groupcl, tlist); |
| 536 |
|
| 537 |
grpColIdx[colno++] = tle->resno; |
| 538 |
} |
| 539 |
|
| 540 |
return grpColIdx; |
| 541 |
} |
| 542 |
|
| 543 |
/* |
| 544 |
* grouping_is_sortable - is it possible to implement grouping list by sorting? |
| 545 |
* |
| 546 |
* This is easy since the parser will have included a sortop if one exists. |
| 547 |
*/ |
| 548 |
bool |
| 549 |
grouping_is_sortable(List *groupClause) |
| 550 |
{ |
| 551 |
ListCell *glitem; |
| 552 |
|
| 553 |
foreach(glitem, groupClause) |
| 554 |
{ |
| 555 |
SortGroupClause *groupcl = (SortGroupClause *) lfirst(glitem); |
| 556 |
|
| 557 |
if (!OidIsValid(groupcl->sortop)) |
| 558 |
return false; |
| 559 |
} |
| 560 |
return true; |
| 561 |
} |
| 562 |
|
| 563 |
/* |
| 564 |
* grouping_is_hashable - is it possible to implement grouping list by hashing? |
| 565 |
* |
| 566 |
* We rely on the parser to have set the hashable flag correctly. |
| 567 |
*/ |
| 568 |
bool |
| 569 |
grouping_is_hashable(List *groupClause) |
| 570 |
{ |
| 571 |
ListCell *glitem; |
| 572 |
|
| 573 |
foreach(glitem, groupClause) |
| 574 |
{ |
| 575 |
SortGroupClause *groupcl = (SortGroupClause *) lfirst(glitem); |
| 576 |
|
| 577 |
if (!groupcl->hashable) |
| 578 |
return false; |
| 579 |
} |
| 580 |
return true; |
| 581 |
} |
| 582 |
|
| 583 |
|
| 584 |
/***************************************************************************** |
| 585 |
* PathTarget manipulation functions |
| 586 |
* |
| 587 |
* PathTarget is a somewhat stripped-down version of a full targetlist; it |
| 588 |
* omits all the TargetEntry decoration except (optionally) sortgroupref data, |
| 589 |
* and it adds evaluation cost and output data width info. |
| 590 |
*****************************************************************************/ |
| 591 |
|
| 592 |
/* |
| 593 |
* make_pathtarget_from_tlist |
| 594 |
* Construct a PathTarget equivalent to the given targetlist. |
| 595 |
* |
| 596 |
* This leaves the cost and width fields as zeroes. Most callers will want |
| 597 |
* to use create_pathtarget(), so as to get those set. |
| 598 |
*/ |
| 599 |
PathTarget * |
| 600 |
make_pathtarget_from_tlist(List *tlist) |
| 601 |
{ |
| 602 |
PathTarget *target = makeNode(PathTarget); |
| 603 |
int i; |
| 604 |
ListCell *lc; |
| 605 |
|
| 606 |
target->sortgrouprefs = (Index *) palloc(list_length(tlist) * sizeof(Index)); |
| 607 |
|
| 608 |
i = 0; |
| 609 |
foreach(lc, tlist) |
| 610 |
{ |
| 611 |
TargetEntry *tle = (TargetEntry *) lfirst(lc); |
| 612 |
|
| 613 |
target->exprs = lappend(target->exprs, tle->expr); |
| 614 |
target->sortgrouprefs[i] = tle->ressortgroupref; |
| 615 |
i++; |
| 616 |
} |
| 617 |
|
| 618 |
/* |
| 619 |
* Mark volatility as unknown. The contain_volatile_functions function |
| 620 |
* will determine if there are any volatile functions when called for the |
| 621 |
* first time with this PathTarget. |
| 622 |
*/ |
| 623 |
target->has_volatile_expr = VOLATILITY_UNKNOWN; |
| 624 |
|
| 625 |
return target; |
| 626 |
} |
| 627 |
|
| 628 |
/* |
| 629 |
* make_tlist_from_pathtarget |
| 630 |
* Construct a targetlist from a PathTarget. |
| 631 |
*/ |
| 632 |
List * |
| 633 |
make_tlist_from_pathtarget(PathTarget *target) |
| 634 |
{ |
| 635 |
List *tlist = NIL; |
| 636 |
int i; |
| 637 |
ListCell *lc; |
| 638 |
|
| 639 |
i = 0; |
| 640 |
foreach(lc, target->exprs) |
| 641 |
{ |
| 642 |
Expr *expr = (Expr *) lfirst(lc); |
| 643 |
TargetEntry *tle; |
| 644 |
|
| 645 |
tle = makeTargetEntry(expr, |
| 646 |
i + 1, |
| 647 |
NULL, |
| 648 |
false); |
| 649 |
if (target->sortgrouprefs) |
| 650 |
tle->ressortgroupref = target->sortgrouprefs[i]; |
| 651 |
tlist = lappend(tlist, tle); |
| 652 |
i++; |
| 653 |
} |
| 654 |
|
| 655 |
return tlist; |
| 656 |
} |
| 657 |
|
| 658 |
/* |
| 659 |
* copy_pathtarget |
| 660 |
* Copy a PathTarget. |
| 661 |
* |
| 662 |
* The new PathTarget has its own exprs List, but shares the underlying |
| 663 |
* target expression trees with the old one. |
| 664 |
*/ |
| 665 |
PathTarget * |
| 666 |
copy_pathtarget(PathTarget *src) |
| 667 |
{ |
| 668 |
PathTarget *dst = makeNode(PathTarget); |
| 669 |
|
| 670 |
/* Copy scalar fields */ |
| 671 |
memcpy(dst, src, sizeof(PathTarget)); |
| 672 |
/* Shallow-copy the expression list */ |
| 673 |
dst->exprs = list_copy(src->exprs); |
| 674 |
/* Duplicate sortgrouprefs if any (if not, the memcpy handled this) */ |
| 675 |
if (src->sortgrouprefs) |
| 676 |
{ |
| 677 |
Size nbytes = list_length(src->exprs) * sizeof(Index); |
| 678 |
|
| 679 |
dst->sortgrouprefs = (Index *) palloc(nbytes); |
| 680 |
memcpy(dst->sortgrouprefs, src->sortgrouprefs, nbytes); |
| 681 |
} |
| 682 |
return dst; |
| 683 |
} |
| 684 |
|
| 685 |
/* |
| 686 |
* create_empty_pathtarget |
| 687 |
* Create an empty (zero columns, zero cost) PathTarget. |
| 688 |
*/ |
| 689 |
PathTarget * |
| 690 |
create_empty_pathtarget(void) |
| 691 |
{ |
| 692 |
/* This is easy, but we don't want callers to hard-wire this ... */ |
| 693 |
return makeNode(PathTarget); |
| 694 |
} |
| 695 |
|
| 696 |
/* |
| 697 |
* add_column_to_pathtarget |
| 698 |
* Append a target column to the PathTarget. |
| 699 |
* |
| 700 |
* As with make_pathtarget_from_tlist, we leave it to the caller to update |
| 701 |
* the cost and width fields. |
| 702 |
*/ |
| 703 |
void |
| 704 |
add_column_to_pathtarget(PathTarget *target, Expr *expr, Index sortgroupref) |
| 705 |
{ |
| 706 |
/* Updating the exprs list is easy ... */ |
| 707 |
target->exprs = lappend(target->exprs, expr); |
| 708 |
/* ... the sortgroupref data, a bit less so */ |
| 709 |
if (target->sortgrouprefs) |
| 710 |
{ |
| 711 |
int nexprs = list_length(target->exprs); |
| 712 |
|
| 713 |
/* This might look inefficient, but actually it's usually cheap */ |
| 714 |
target->sortgrouprefs = (Index *) |
| 715 |
repalloc(target->sortgrouprefs, nexprs * sizeof(Index)); |
| 716 |
target->sortgrouprefs[nexprs - 1] = sortgroupref; |
| 717 |
} |
| 718 |
else if (sortgroupref) |
| 719 |
{ |
| 720 |
/* Adding sortgroupref labeling to a previously unlabeled target */ |
| 721 |
int nexprs = list_length(target->exprs); |
| 722 |
|
| 723 |
target->sortgrouprefs = (Index *) palloc0(nexprs * sizeof(Index)); |
| 724 |
target->sortgrouprefs[nexprs - 1] = sortgroupref; |
| 725 |
} |
| 726 |
|
| 727 |
/* |
| 728 |
* Reset has_volatile_expr to UNKNOWN. We just leave it up to |
| 729 |
* contain_volatile_functions to set this properly again. Technically we |
| 730 |
* could save some effort here and just check the new Expr, but it seems |
| 731 |
* better to keep the logic for setting this flag in one location rather |
| 732 |
* than duplicating the logic here. |
| 733 |
*/ |
| 734 |
if (target->has_volatile_expr == VOLATILITY_NOVOLATILE) |
| 735 |
target->has_volatile_expr = VOLATILITY_UNKNOWN; |
| 736 |
} |
| 737 |
|
| 738 |
/* |
| 739 |
* add_new_column_to_pathtarget |
| 740 |
* Append a target column to the PathTarget, but only if it's not |
| 741 |
* equal() to any pre-existing target expression. |
| 742 |
* |
| 743 |
* The caller cannot specify a sortgroupref, since it would be unclear how |
| 744 |
* to merge that with a pre-existing column. |
| 745 |
* |
| 746 |
* As with make_pathtarget_from_tlist, we leave it to the caller to update |
| 747 |
* the cost and width fields. |
| 748 |
*/ |
| 749 |
void |
| 750 |
add_new_column_to_pathtarget(PathTarget *target, Expr *expr) |
| 751 |
{ |
| 752 |
if (!list_member(target->exprs, expr)) |
| 753 |
add_column_to_pathtarget(target, expr, 0); |
| 754 |
} |
| 755 |
|
| 756 |
/* |
| 757 |
* add_new_columns_to_pathtarget |
| 758 |
* Apply add_new_column_to_pathtarget() for each element of the list. |
| 759 |
*/ |
| 760 |
void |
| 761 |
add_new_columns_to_pathtarget(PathTarget *target, List *exprs) |
| 762 |
{ |
| 763 |
ListCell *lc; |
| 764 |
|
| 765 |
foreach(lc, exprs) |
| 766 |
{ |
| 767 |
Expr *expr = (Expr *) lfirst(lc); |
| 768 |
|
| 769 |
add_new_column_to_pathtarget(target, expr); |
| 770 |
} |
| 771 |
} |
| 772 |
|
| 773 |
/* |
| 774 |
* apply_pathtarget_labeling_to_tlist |
| 775 |
* Apply any sortgrouprefs in the PathTarget to matching tlist entries |
| 776 |
* |
| 777 |
* Here, we do not assume that the tlist entries are one-for-one with the |
| 778 |
* PathTarget. The intended use of this function is to deal with cases |
| 779 |
* where createplan.c has decided to use some other tlist and we have |
| 780 |
* to identify what matches exist. |
| 781 |
*/ |
| 782 |
void |
| 783 |
apply_pathtarget_labeling_to_tlist(List *tlist, PathTarget *target) |
| 784 |
{ |
| 785 |
int i; |
| 786 |
ListCell *lc; |
| 787 |
|
| 788 |
/* Nothing to do if PathTarget has no sortgrouprefs data */ |
| 789 |
if (target->sortgrouprefs == NULL) |
| 790 |
return; |
| 791 |
|
| 792 |
i = 0; |
| 793 |
foreach(lc, target->exprs) |
| 794 |
{ |
| 795 |
Expr *expr = (Expr *) lfirst(lc); |
| 796 |
TargetEntry *tle; |
| 797 |
|
| 798 |
if (target->sortgrouprefs[i]) |
| 799 |
{ |
| 800 |
/* |
| 801 |
* For Vars, use tlist_member_match_var's weakened matching rule; |
| 802 |
* this allows us to deal with some cases where a set-returning |
| 803 |
* function has been inlined, so that we now have more knowledge |
| 804 |
* about what it returns than we did when the original Var was |
| 805 |
* created. Otherwise, use regular equal() to find the matching |
| 806 |
* TLE. (In current usage, only the Var case is actually needed; |
| 807 |
* but it seems best to have sane behavior here for non-Vars too.) |
| 808 |
*/ |
| 809 |
if (expr && IsA(expr, Var)) |
| 810 |
tle = tlist_member_match_var((Var *) expr, tlist); |
| 811 |
else |
| 812 |
tle = tlist_member(expr, tlist); |
| 813 |
|
| 814 |
/* |
| 815 |
* Complain if noplace for the sortgrouprefs label, or if we'd |
| 816 |
* have to label a column twice. (The case where it already has |
| 817 |
* the desired label probably can't happen, but we may as well |
| 818 |
* allow for it.) |
| 819 |
*/ |
| 820 |
if (!tle) |
| 821 |
elog(ERROR, "ORDER/GROUP BY expression not found in targetlist"); |
| 822 |
if (tle->ressortgroupref != 0 && |
| 823 |
tle->ressortgroupref != target->sortgrouprefs[i]) |
| 824 |
elog(ERROR, "targetlist item has multiple sortgroupref labels"); |
| 825 |
|
| 826 |
tle->ressortgroupref = target->sortgrouprefs[i]; |
| 827 |
} |
| 828 |
i++; |
| 829 |
} |
| 830 |
} |
| 831 |
|
| 832 |
/* |
| 833 |
* split_pathtarget_at_srfs |
| 834 |
* Split given PathTarget into multiple levels to position SRFs safely, |
| 835 |
* performing exact matching against input_target. |
| 836 |
* |
| 837 |
* This is a wrapper for split_pathtarget_at_srfs_extended() that is used when |
| 838 |
* both targets are on the same side of the grouping boundary (i.e., both are |
| 839 |
* pre-grouping or both are post-grouping). In this case, no special handling |
| 840 |
* for the grouping nulling bit is required. |
| 841 |
* |
| 842 |
* See split_pathtarget_at_srfs_extended() for more details. |
| 843 |
*/ |
| 844 |
void |
| 845 |
split_pathtarget_at_srfs(PlannerInfo *root, |
| 846 |
PathTarget *target, PathTarget *input_target, |
| 847 |
List **targets, List **targets_contain_srfs) |
| 848 |
{ |
| 849 |
split_pathtarget_at_srfs_extended(root, target, input_target, |
| 850 |
targets, targets_contain_srfs, |
| 851 |
false); |
| 852 |
} |
| 853 |
|
| 854 |
/* |
| 855 |
* split_pathtarget_at_srfs_grouping |
| 856 |
* Split given PathTarget into multiple levels to position SRFs safely, |
| 857 |
* ignoring the grouping nulling bit when matching against input_target. |
| 858 |
* |
| 859 |
* This variant is used when the targets cross the grouping boundary (i.e., |
| 860 |
* target is post-grouping while input_target is pre-grouping). In this case, |
| 861 |
* we need to ignore the grouping nulling bit when checking for expression |
| 862 |
* availability to avoid incorrectly re-evaluating SRFs that have already been |
| 863 |
* computed in input_target. |
| 864 |
* |
| 865 |
* See split_pathtarget_at_srfs_extended() for more details. |
| 866 |
*/ |
| 867 |
void |
| 868 |
split_pathtarget_at_srfs_grouping(PlannerInfo *root, |
| 869 |
PathTarget *target, PathTarget *input_target, |
| 870 |
List **targets, List **targets_contain_srfs) |
| 871 |
{ |
| 872 |
split_pathtarget_at_srfs_extended(root, target, input_target, |
| 873 |
targets, targets_contain_srfs, |
| 874 |
true); |
| 875 |
} |
| 876 |
|
| 877 |
/* |
| 878 |
* split_pathtarget_at_srfs_extended |
| 879 |
* Split given PathTarget into multiple levels to position SRFs safely |
| 880 |
* |
| 881 |
* The executor can only handle set-returning functions that appear at the |
| 882 |
* top level of the targetlist of a ProjectSet plan node. If we have any SRFs |
| 883 |
* that are not at top level, we need to split up the evaluation into multiple |
| 884 |
* plan levels in which each level satisfies this constraint. This function |
| 885 |
* creates appropriate PathTarget(s) for each level. |
| 886 |
* |
| 887 |
* As an example, consider the tlist expression |
| 888 |
* x + srf1(srf2(y + z)) |
| 889 |
* This expression should appear as-is in the top PathTarget, but below that |
| 890 |
* we must have a PathTarget containing |
| 891 |
* x, srf1(srf2(y + z)) |
| 892 |
* and below that, another PathTarget containing |
| 893 |
* x, srf2(y + z) |
| 894 |
* and below that, another PathTarget containing |
| 895 |
* x, y, z |
| 896 |
* When these tlists are processed by setrefs.c, subexpressions that match |
| 897 |
* output expressions of the next lower tlist will be replaced by Vars, |
| 898 |
* so that what the executor gets are tlists looking like |
| 899 |
* Var1 + Var2 |
| 900 |
* Var1, srf1(Var2) |
| 901 |
* Var1, srf2(Var2 + Var3) |
| 902 |
* x, y, z |
| 903 |
* which satisfy the desired property. |
| 904 |
* |
| 905 |
* Another example is |
| 906 |
* srf1(x), srf2(srf3(y)) |
| 907 |
* That must appear as-is in the top PathTarget, but below that we need |
| 908 |
* srf1(x), srf3(y) |
| 909 |
* That is, each SRF must be computed at a level corresponding to the nesting |
| 910 |
* depth of SRFs within its arguments. |
| 911 |
* |
| 912 |
* In some cases, a SRF has already been evaluated in some previous plan level |
| 913 |
* and we shouldn't expand it again (that is, what we see in the target is |
| 914 |
* already meant as a reference to a lower subexpression). So, don't expand |
| 915 |
* any tlist expressions that appear in input_target, if that's not NULL. |
| 916 |
* |
| 917 |
* This check requires extra care when processing the grouping target |
| 918 |
* (indicated by the is_grouping_target flag). In this case input_target is |
| 919 |
* pre-grouping while target is post-grouping, so the latter may carry |
| 920 |
* nullingrels bits from the grouping step that are absent in the former. We |
| 921 |
* must ignore those bits to correctly recognize that the tlist expressions are |
| 922 |
* available in input_target. |
| 923 |
* |
| 924 |
* It's also important that we preserve any sortgroupref annotation appearing |
| 925 |
* in the given target, especially on expressions matching input_target items. |
| 926 |
* |
| 927 |
* The outputs of this function are two parallel lists, one a list of |
| 928 |
* PathTargets and the other an integer list of bool flags indicating |
| 929 |
* whether the corresponding PathTarget contains any evaluable SRFs. |
| 930 |
* The lists are given in the order they'd need to be evaluated in, with |
| 931 |
* the "lowest" PathTarget first. So the last list entry is always the |
| 932 |
* originally given PathTarget, and any entries before it indicate evaluation |
| 933 |
* levels that must be inserted below it. The first list entry must not |
| 934 |
* contain any SRFs (other than ones duplicating input_target entries), since |
| 935 |
* it will typically be attached to a plan node that cannot evaluate SRFs. |
| 936 |
* |
| 937 |
* Note: using a list for the flags may seem like overkill, since there |
| 938 |
* are only a few possible patterns for which levels contain SRFs. |
| 939 |
* But this representation decouples callers from that knowledge. |
| 940 |
*/ |
| 941 |
static void |
| 942 |
split_pathtarget_at_srfs_extended(PlannerInfo *root, |
| 943 |
PathTarget *target, PathTarget *input_target, |
| 944 |
List **targets, List **targets_contain_srfs, |
| 945 |
bool is_grouping_target) |
| 946 |
{ |
| 947 |
split_pathtarget_context context; |
| 948 |
int max_depth; |
| 949 |
bool need_extra_projection; |
| 950 |
List *prev_level_tlist; |
| 951 |
int lci; |
| 952 |
ListCell *lc, |
| 953 |
*lc1, |
| 954 |
*lc2, |
| 955 |
*lc3; |
| 956 |
|
| 957 |
/* |
| 958 |
* It's not unusual for planner.c to pass us two physically identical |
| 959 |
* targets, in which case we can conclude without further ado that all |
| 960 |
* expressions are available from the input. (The logic below would |
| 961 |
* arrive at the same conclusion, but much more tediously.) |
| 962 |
*/ |
| 963 |
if (target == input_target) |
| 964 |
{ |
| 965 |
*targets = list_make1(target); |
| 966 |
*targets_contain_srfs = list_make1_int(false); |
| 967 |
return; |
| 968 |
} |
| 969 |
|
| 970 |
/* |
| 971 |
* Pass 'root', the is_grouping_target flag, and any input_target exprs |
| 972 |
* down to split_pathtarget_walker(). |
| 973 |
*/ |
| 974 |
context.root = root; |
| 975 |
context.is_grouping_target = is_grouping_target; |
| 976 |
context.input_target_exprs = input_target ? input_target->exprs : NIL; |
| 977 |
|
| 978 |
/* |
| 979 |
* Initialize with empty level-zero lists, and no levels after that. |
| 980 |
* (Note: we could dispense with representing level zero explicitly, since |
| 981 |
* it will never receive any SRFs, but then we'd have to special-case that |
| 982 |
* level when we get to building result PathTargets. Level zero describes |
| 983 |
* the SRF-free PathTarget that will be given to the input plan node.) |
| 984 |
*/ |
| 985 |
context.level_srfs = list_make1(NIL); |
| 986 |
context.level_input_vars = list_make1(NIL); |
| 987 |
context.level_input_srfs = list_make1(NIL); |
| 988 |
|
| 989 |
/* Initialize data we'll accumulate across all the target expressions */ |
| 990 |
context.current_input_vars = NIL; |
| 991 |
context.current_input_srfs = NIL; |
| 992 |
max_depth = 0; |
| 993 |
need_extra_projection = false; |
| 994 |
|
| 995 |
/* Scan each expression in the PathTarget looking for SRFs */ |
| 996 |
lci = 0; |
| 997 |
foreach(lc, target->exprs) |
| 998 |
{ |
| 999 |
Node *node = (Node *) lfirst(lc); |
| 1000 |
|
| 1001 |
/* Tell split_pathtarget_walker about this expr's sortgroupref */ |
| 1002 |
context.current_sgref = get_pathtarget_sortgroupref(target, lci); |
| 1003 |
lci++; |
| 1004 |
|
| 1005 |
/* |
| 1006 |
* Find all SRFs and Vars (and Var-like nodes) in this expression, and |
| 1007 |
* enter them into appropriate lists within the context struct. |
| 1008 |
*/ |
| 1009 |
context.current_depth = 0; |
| 1010 |
split_pathtarget_walker(node, &context); |
| 1011 |
|
| 1012 |
/* An expression containing no SRFs is of no further interest */ |
| 1013 |
if (context.current_depth == 0) |
| 1014 |
continue; |
| 1015 |
|
| 1016 |
/* |
| 1017 |
* Track max SRF nesting depth over the whole PathTarget. Also, if |
| 1018 |
* this expression establishes a new max depth, we no longer care |
| 1019 |
* whether previous expressions contained nested SRFs; we can handle |
| 1020 |
* any required projection for them in the final ProjectSet node. |
| 1021 |
*/ |
| 1022 |
if (max_depth < context.current_depth) |
| 1023 |
{ |
| 1024 |
max_depth = context.current_depth; |
| 1025 |
need_extra_projection = false; |
| 1026 |
} |
| 1027 |
|
| 1028 |
/* |
| 1029 |
* If any maximum-depth SRF is not at the top level of its expression, |
| 1030 |
* we'll need an extra Result node to compute the top-level scalar |
| 1031 |
* expression. |
| 1032 |
*/ |
| 1033 |
if (max_depth == context.current_depth && !IS_SRF_CALL(node)) |
| 1034 |
need_extra_projection = true; |
| 1035 |
} |
| 1036 |
|
| 1037 |
/* |
| 1038 |
* If we found no SRFs needing evaluation (maybe they were all present in |
| 1039 |
* input_target, or maybe they were all removed by const-simplification), |
| 1040 |
* then no ProjectSet is needed; fall out. |
| 1041 |
*/ |
| 1042 |
if (max_depth == 0) |
| 1043 |
{ |
| 1044 |
*targets = list_make1(target); |
| 1045 |
*targets_contain_srfs = list_make1_int(false); |
| 1046 |
return; |
| 1047 |
} |
| 1048 |
|
| 1049 |
/* |
| 1050 |
* The Vars and SRF outputs needed at top level can be added to the last |
| 1051 |
* level_input lists if we don't need an extra projection step. If we do |
| 1052 |
* need one, add a SRF-free level to the lists. |
| 1053 |
*/ |
| 1054 |
if (need_extra_projection) |
| 1055 |
{ |
| 1056 |
context.level_srfs = lappend(context.level_srfs, NIL); |
| 1057 |
context.level_input_vars = lappend(context.level_input_vars, |
| 1058 |
context.current_input_vars); |
| 1059 |
context.level_input_srfs = lappend(context.level_input_srfs, |
| 1060 |
context.current_input_srfs); |
| 1061 |
} |
| 1062 |
else |
| 1063 |
{ |
| 1064 |
lc = list_nth_cell(context.level_input_vars, max_depth); |
| 1065 |
lfirst(lc) = list_concat(lfirst(lc), context.current_input_vars); |
| 1066 |
lc = list_nth_cell(context.level_input_srfs, max_depth); |
| 1067 |
lfirst(lc) = list_concat(lfirst(lc), context.current_input_srfs); |
| 1068 |
} |
| 1069 |
|
| 1070 |
/* |
| 1071 |
* Now construct the output PathTargets. The original target can be used |
| 1072 |
* as-is for the last one, but we need to construct a new SRF-free target |
| 1073 |
* representing what the preceding plan node has to emit, as well as a |
| 1074 |
* target for each intermediate ProjectSet node. |
| 1075 |
*/ |
| 1076 |
*targets = *targets_contain_srfs = NIL; |
| 1077 |
prev_level_tlist = NIL; |
| 1078 |
|
| 1079 |
forthree(lc1, context.level_srfs, |
| 1080 |
lc2, context.level_input_vars, |
| 1081 |
lc3, context.level_input_srfs) |
| 1082 |
{ |
| 1083 |
List *level_srfs = (List *) lfirst(lc1); |
| 1084 |
PathTarget *ntarget; |
| 1085 |
|
| 1086 |
if (lnext(context.level_srfs, lc1) == NULL) |
| 1087 |
{ |
| 1088 |
ntarget = target; |
| 1089 |
} |
| 1090 |
else |
| 1091 |
{ |
| 1092 |
ntarget = create_empty_pathtarget(); |
| 1093 |
|
| 1094 |
/* |
| 1095 |
* This target should actually evaluate any SRFs of the current |
| 1096 |
* level, and it needs to propagate forward any Vars needed by |
| 1097 |
* later levels, as well as SRFs computed earlier and needed by |
| 1098 |
* later levels. |
| 1099 |
*/ |
| 1100 |
add_sp_items_to_pathtarget(ntarget, level_srfs); |
| 1101 |
for_each_cell(lc, context.level_input_vars, |
| 1102 |
lnext(context.level_input_vars, lc2)) |
| 1103 |
{ |
| 1104 |
List *input_vars = (List *) lfirst(lc); |
| 1105 |
|
| 1106 |
add_sp_items_to_pathtarget(ntarget, input_vars); |
| 1107 |
} |
| 1108 |
for_each_cell(lc, context.level_input_srfs, |
| 1109 |
lnext(context.level_input_srfs, lc3)) |
| 1110 |
{ |
| 1111 |
List *input_srfs = (List *) lfirst(lc); |
| 1112 |
ListCell *lcx; |
| 1113 |
|
| 1114 |
foreach(lcx, input_srfs) |
| 1115 |
{ |
| 1116 |
split_pathtarget_item *item = lfirst(lcx); |
| 1117 |
|
| 1118 |
if (list_member(prev_level_tlist, item->expr)) |
| 1119 |
add_sp_item_to_pathtarget(ntarget, item); |
| 1120 |
} |
| 1121 |
} |
| 1122 |
set_pathtarget_cost_width(root, ntarget); |
| 1123 |
} |
| 1124 |
|
| 1125 |
/* |
| 1126 |
* Add current target and does-it-compute-SRFs flag to output lists. |
| 1127 |
*/ |
| 1128 |
*targets = lappend(*targets, ntarget); |
| 1129 |
*targets_contain_srfs = lappend_int(*targets_contain_srfs, |
| 1130 |
(level_srfs != NIL)); |
| 1131 |
|
| 1132 |
/* Remember this level's output for next pass */ |
| 1133 |
prev_level_tlist = ntarget->exprs; |
| 1134 |
} |
| 1135 |
} |
| 1136 |
|
| 1137 |
/* |
| 1138 |
* Recursively examine expressions for split_pathtarget_at_srfs. |
| 1139 |
* |
| 1140 |
* Note we make no effort here to prevent duplicate entries in the output |
| 1141 |
* lists. Duplicates will be gotten rid of later. |
| 1142 |
*/ |
| 1143 |
static bool |
| 1144 |
split_pathtarget_walker(Node *node, split_pathtarget_context *context) |
| 1145 |
{ |
| 1146 |
Node *sanitized_node = node; |
| 1147 |
|
| 1148 |
if (node == NULL) |
| 1149 |
return false; |
| 1150 |
|
| 1151 |
/* |
| 1152 |
* If we are crossing the grouping boundary (post-grouping target vs |
| 1153 |
* pre-grouping input_target), we must ignore the grouping nulling bit to |
| 1154 |
* correctly check if the subexpression is available in input_target. This |
| 1155 |
* aligns with the matching logic in set_upper_references(). |
| 1156 |
*/ |
| 1157 |
if (context->is_grouping_target && |
| 1158 |
context->root->parse->hasGroupRTE && |
| 1159 |
context->root->parse->groupingSets != NIL) |
| 1160 |
{ |
| 1161 |
sanitized_node = |
| 1162 |
remove_nulling_relids(node, |
| 1163 |
bms_make_singleton(context->root->group_rtindex), |
| 1164 |
NULL); |
| 1165 |
} |
| 1166 |
|
| 1167 |
/* |
| 1168 |
* A subexpression that matches an expression already computed in |
| 1169 |
* input_target can be treated like a Var (which indeed it will be after |
| 1170 |
* setrefs.c gets done with it), even if it's actually a SRF. Record it |
| 1171 |
* as being needed for the current expression, and ignore any |
| 1172 |
* substructure. (Note in particular that this preserves the identity of |
| 1173 |
* any expressions that appear as sortgrouprefs in input_target.) |
| 1174 |
*/ |
| 1175 |
if (list_member(context->input_target_exprs, sanitized_node)) |
| 1176 |
{ |
| 1177 |
split_pathtarget_item *item = palloc(sizeof(split_pathtarget_item)); |
| 1178 |
|
| 1179 |
item->expr = node; |
| 1180 |
item->sortgroupref = context->current_sgref; |
| 1181 |
context->current_input_vars = lappend(context->current_input_vars, |
| 1182 |
item); |
| 1183 |
return false; |
| 1184 |
} |
| 1185 |
|
| 1186 |
/* |
| 1187 |
* Vars and Var-like constructs are expected to be gotten from the input, |
| 1188 |
* too. We assume that these constructs cannot contain any SRFs (if one |
| 1189 |
* does, there will be an executor failure from a misplaced SRF). |
| 1190 |
*/ |
| 1191 |
if (IsA(node, Var) || |
| 1192 |
IsA(node, PlaceHolderVar) || |
| 1193 |
IsA(node, Aggref) || |
| 1194 |
IsA(node, GroupingFunc) || |
| 1195 |
IsA(node, WindowFunc)) |
| 1196 |
{ |
| 1197 |
split_pathtarget_item *item = palloc(sizeof(split_pathtarget_item)); |
| 1198 |
|
| 1199 |
item->expr = node; |
| 1200 |
item->sortgroupref = context->current_sgref; |
| 1201 |
context->current_input_vars = lappend(context->current_input_vars, |
| 1202 |
item); |
| 1203 |
return false; |
| 1204 |
} |
| 1205 |
|
| 1206 |
/* |
| 1207 |
* If it's a SRF, recursively examine its inputs, determine its level, and |
| 1208 |
* make appropriate entries in the output lists. |
| 1209 |
*/ |
| 1210 |
if (IS_SRF_CALL(node)) |
| 1211 |
{ |
| 1212 |
split_pathtarget_item *item = palloc(sizeof(split_pathtarget_item)); |
| 1213 |
List *save_input_vars = context->current_input_vars; |
| 1214 |
List *save_input_srfs = context->current_input_srfs; |
| 1215 |
int save_current_depth = context->current_depth; |
| 1216 |
int srf_depth; |
| 1217 |
ListCell *lc; |
| 1218 |
|
| 1219 |
item->expr = node; |
| 1220 |
item->sortgroupref = context->current_sgref; |
| 1221 |
|
| 1222 |
context->current_input_vars = NIL; |
| 1223 |
context->current_input_srfs = NIL; |
| 1224 |
context->current_depth = 0; |
| 1225 |
context->current_sgref = 0; /* subexpressions are not sortgroup items */ |
| 1226 |
|
| 1227 |
(void) expression_tree_walker(node, split_pathtarget_walker, context); |
| 1228 |
|
| 1229 |
/* Depth is one more than any SRF below it */ |
| 1230 |
srf_depth = context->current_depth + 1; |
| 1231 |
|
| 1232 |
/* If new record depth, initialize another level of output lists */ |
| 1233 |
if (srf_depth >= list_length(context->level_srfs)) |
| 1234 |
{ |
| 1235 |
context->level_srfs = lappend(context->level_srfs, NIL); |
| 1236 |
context->level_input_vars = lappend(context->level_input_vars, NIL); |
| 1237 |
context->level_input_srfs = lappend(context->level_input_srfs, NIL); |
| 1238 |
} |
| 1239 |
|
| 1240 |
/* Record this SRF as needing to be evaluated at appropriate level */ |
| 1241 |
lc = list_nth_cell(context->level_srfs, srf_depth); |
| 1242 |
lfirst(lc) = lappend(lfirst(lc), item); |
| 1243 |
|
| 1244 |
/* Record its inputs as being needed at the same level */ |
| 1245 |
lc = list_nth_cell(context->level_input_vars, srf_depth); |
| 1246 |
lfirst(lc) = list_concat(lfirst(lc), context->current_input_vars); |
| 1247 |
lc = list_nth_cell(context->level_input_srfs, srf_depth); |
| 1248 |
lfirst(lc) = list_concat(lfirst(lc), context->current_input_srfs); |
| 1249 |
|
| 1250 |
/* |
| 1251 |
* Restore caller-level state and update it for presence of this SRF. |
| 1252 |
* Notice we report the SRF itself as being needed for evaluation of |
| 1253 |
* surrounding expression. |
| 1254 |
*/ |
| 1255 |
context->current_input_vars = save_input_vars; |
| 1256 |
context->current_input_srfs = lappend(save_input_srfs, item); |
| 1257 |
context->current_depth = Max(save_current_depth, srf_depth); |
| 1258 |
|
| 1259 |
/* We're done here */ |
| 1260 |
return false; |
| 1261 |
} |
| 1262 |
|
| 1263 |
/* |
| 1264 |
* Otherwise, the node is a scalar (non-set) expression, so recurse to |
| 1265 |
* examine its inputs. |
| 1266 |
*/ |
| 1267 |
context->current_sgref = 0; /* subexpressions are not sortgroup items */ |
| 1268 |
return expression_tree_walker(node, split_pathtarget_walker, context); |
| 1269 |
} |
| 1270 |
|
| 1271 |
/* |
| 1272 |
* Add a split_pathtarget_item to the PathTarget, unless a matching item is |
| 1273 |
* already present. This is like add_new_column_to_pathtarget, but allows |
| 1274 |
* for sortgrouprefs to be handled. An item having zero sortgroupref can |
| 1275 |
* be merged with one that has a sortgroupref, acquiring the latter's |
| 1276 |
* sortgroupref. |
| 1277 |
* |
| 1278 |
* Note that we don't worry about possibly adding duplicate sortgrouprefs |
| 1279 |
* to the PathTarget. That would be bad, but it should be impossible unless |
| 1280 |
* the target passed to split_pathtarget_at_srfs already had duplicates. |
| 1281 |
* As long as it didn't, we can have at most one split_pathtarget_item with |
| 1282 |
* any particular nonzero sortgroupref. |
| 1283 |
*/ |
| 1284 |
static void |
| 1285 |
add_sp_item_to_pathtarget(PathTarget *target, split_pathtarget_item *item) |
| 1286 |
{ |
| 1287 |
int lci; |
| 1288 |
ListCell *lc; |
| 1289 |
|
| 1290 |
/* |
| 1291 |
* Look for a pre-existing entry that is equal() and does not have a |
| 1292 |
* conflicting sortgroupref already. |
| 1293 |
*/ |
| 1294 |
lci = 0; |
| 1295 |
foreach(lc, target->exprs) |
| 1296 |
{ |
| 1297 |
Node *node = (Node *) lfirst(lc); |
| 1298 |
Index sgref = get_pathtarget_sortgroupref(target, lci); |
| 1299 |
|
| 1300 |
if ((item->sortgroupref == sgref || |
| 1301 |
item->sortgroupref == 0 || |
| 1302 |
sgref == 0) && |
| 1303 |
equal(item->expr, node)) |
| 1304 |
{ |
| 1305 |
/* Found a match. Assign item's sortgroupref if it has one. */ |
| 1306 |
if (item->sortgroupref) |
| 1307 |
{ |
| 1308 |
if (target->sortgrouprefs == NULL) |
| 1309 |
{ |
| 1310 |
target->sortgrouprefs = (Index *) |
| 1311 |
palloc0(list_length(target->exprs) * sizeof(Index)); |
| 1312 |
} |
| 1313 |
target->sortgrouprefs[lci] = item->sortgroupref; |
| 1314 |
} |
| 1315 |
return; |
| 1316 |
} |
| 1317 |
lci++; |
| 1318 |
} |
| 1319 |
|
| 1320 |
/* |
| 1321 |
* No match, so add item to PathTarget. Copy the expr for safety. |
| 1322 |
*/ |
| 1323 |
add_column_to_pathtarget(target, (Expr *) copyObject(item->expr), |
| 1324 |
item->sortgroupref); |
| 1325 |
} |
| 1326 |
|
| 1327 |
/* |
| 1328 |
* Apply add_sp_item_to_pathtarget to each element of list. |
| 1329 |
*/ |
| 1330 |
static void |
| 1331 |
add_sp_items_to_pathtarget(PathTarget *target, List *items) |
| 1332 |
{ |
| 1333 |
ListCell *lc; |
| 1334 |
|
| 1335 |
foreach(lc, items) |
| 1336 |
{ |
| 1337 |
split_pathtarget_item *item = lfirst(lc); |
| 1338 |
|
| 1339 |
add_sp_item_to_pathtarget(target, item); |
| 1340 |
} |
| 1341 |
} |