| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* gistscan.c |
| 4 |
* routines to manage scans on GiST index relations |
| 5 |
* |
| 6 |
* |
| 7 |
* Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
| 8 |
* Portions Copyright (c) 1994, Regents of the University of California |
| 9 |
* |
| 10 |
* IDENTIFICATION |
| 11 |
* src/backend/access/gist/gistscan.c |
| 12 |
* |
| 13 |
*------------------------------------------------------------------------- |
| 14 |
*/ |
| 15 |
#include "postgres.h" |
| 16 |
|
| 17 |
#include "access/gist_private.h" |
| 18 |
#include "access/gistscan.h" |
| 19 |
#include "access/relscan.h" |
| 20 |
#include "utils/float.h" |
| 21 |
#include "utils/lsyscache.h" |
| 22 |
#include "utils/memutils.h" |
| 23 |
#include "utils/rel.h" |
| 24 |
|
| 25 |
|
| 26 |
/* |
| 27 |
* Pairing heap comparison function for the GISTSearchItem queue |
| 28 |
*/ |
| 29 |
static int |
| 30 |
pairingheap_GISTSearchItem_cmp(const pairingheap_node *a, const pairingheap_node *b, void *arg) |
| 31 |
{ |
| 32 |
const GISTSearchItem *sa = (const GISTSearchItem *) a; |
| 33 |
const GISTSearchItem *sb = (const GISTSearchItem *) b; |
| 34 |
IndexScanDesc scan = (IndexScanDesc) arg; |
| 35 |
int i; |
| 36 |
|
| 37 |
/* Order according to distance comparison */ |
| 38 |
for (i = 0; i < scan->numberOfOrderBys; i++) |
| 39 |
{ |
| 40 |
if (sa->distances[i].isnull) |
| 41 |
{ |
| 42 |
if (!sb->distances[i].isnull) |
| 43 |
return -1; |
| 44 |
} |
| 45 |
else if (sb->distances[i].isnull) |
| 46 |
{ |
| 47 |
return 1; |
| 48 |
} |
| 49 |
else |
| 50 |
{ |
| 51 |
int cmp = -float8_cmp_internal(sa->distances[i].value, |
| 52 |
sb->distances[i].value); |
| 53 |
|
| 54 |
if (cmp != 0) |
| 55 |
return cmp; |
| 56 |
} |
| 57 |
} |
| 58 |
|
| 59 |
/* Heap items go before inner pages, to ensure a depth-first search */ |
| 60 |
if (GISTSearchItemIsHeap(*sa) && !GISTSearchItemIsHeap(*sb)) |
| 61 |
return 1; |
| 62 |
if (!GISTSearchItemIsHeap(*sa) && GISTSearchItemIsHeap(*sb)) |
| 63 |
return -1; |
| 64 |
|
| 65 |
return 0; |
| 66 |
} |
| 67 |
|
| 68 |
|
| 69 |
/* |
| 70 |
* Index AM API functions for scanning GiST indexes |
| 71 |
*/ |
| 72 |
|
| 73 |
IndexScanDesc |
| 74 |
gistbeginscan(Relation r, int nkeys, int norderbys) |
| 75 |
{ |
| 76 |
IndexScanDesc scan; |
| 77 |
GISTSTATE *giststate; |
| 78 |
GISTScanOpaque so; |
| 79 |
MemoryContext oldCxt; |
| 80 |
|
| 81 |
scan = RelationGetIndexScan(r, nkeys, norderbys); |
| 82 |
|
| 83 |
/* First, set up a GISTSTATE with a scan-lifespan memory context */ |
| 84 |
giststate = initGISTstate(scan->indexRelation); |
| 85 |
|
| 86 |
/* |
| 87 |
* Everything made below is in the scanCxt, or is a child of the scanCxt, |
| 88 |
* so it'll all go away automatically in gistendscan. |
| 89 |
*/ |
| 90 |
oldCxt = MemoryContextSwitchTo(giststate->scanCxt); |
| 91 |
|
| 92 |
/* initialize opaque data */ |
| 93 |
so = (GISTScanOpaque) palloc0(sizeof(GISTScanOpaqueData)); |
| 94 |
so->giststate = giststate; |
| 95 |
giststate->tempCxt = createTempGistContext(); |
| 96 |
so->queue = NULL; |
| 97 |
so->queueCxt = giststate->scanCxt; /* see gistrescan */ |
| 98 |
|
| 99 |
/* workspaces with size dependent on numberOfOrderBys: */ |
| 100 |
so->distances = palloc(sizeof(so->distances[0]) * scan->numberOfOrderBys); |
| 101 |
so->qual_ok = true; /* in case there are zero keys */ |
| 102 |
if (scan->numberOfOrderBys > 0) |
| 103 |
{ |
| 104 |
scan->xs_orderbyvals = palloc0(sizeof(Datum) * scan->numberOfOrderBys); |
| 105 |
scan->xs_orderbynulls = palloc(sizeof(bool) * scan->numberOfOrderBys); |
| 106 |
memset(scan->xs_orderbynulls, true, sizeof(bool) * scan->numberOfOrderBys); |
| 107 |
} |
| 108 |
|
| 109 |
so->killedItems = NULL; /* until needed */ |
| 110 |
so->numKilled = 0; |
| 111 |
so->curBlkno = InvalidBlockNumber; |
| 112 |
so->curPageLSN = InvalidXLogRecPtr; |
| 113 |
|
| 114 |
scan->opaque = so; |
| 115 |
|
| 116 |
/* |
| 117 |
* All fields required for index-only scans are initialized in gistrescan, |
| 118 |
* as we don't know yet if we're doing an index-only scan or not. |
| 119 |
*/ |
| 120 |
|
| 121 |
MemoryContextSwitchTo(oldCxt); |
| 122 |
|
| 123 |
return scan; |
| 124 |
} |
| 125 |
|
| 126 |
void |
| 127 |
gistrescan(IndexScanDesc scan, ScanKey key, int nkeys, |
| 128 |
ScanKey orderbys, int norderbys) |
| 129 |
{ |
| 130 |
/* nkeys and norderbys arguments are ignored */ |
| 131 |
GISTScanOpaque so = (GISTScanOpaque) scan->opaque; |
| 132 |
bool first_time; |
| 133 |
int i; |
| 134 |
MemoryContext oldCxt; |
| 135 |
|
| 136 |
/* rescan an existing indexscan --- reset state */ |
| 137 |
|
| 138 |
/* |
| 139 |
* The first time through, we create the search queue in the scanCxt. |
| 140 |
* Subsequent times through, we create the queue in a separate queueCxt, |
| 141 |
* which is created on the second call and reset on later calls. Thus, in |
| 142 |
* the common case where a scan is only rescan'd once, we just put the |
| 143 |
* queue in scanCxt and don't pay the overhead of making a second memory |
| 144 |
* context. If we do rescan more than once, the first queue is just left |
| 145 |
* for dead until end of scan; this small wastage seems worth the savings |
| 146 |
* in the common case. |
| 147 |
*/ |
| 148 |
if (so->queue == NULL) |
| 149 |
{ |
| 150 |
/* first time through */ |
| 151 |
Assert(so->queueCxt == so->giststate->scanCxt); |
| 152 |
first_time = true; |
| 153 |
} |
| 154 |
else if (so->queueCxt == so->giststate->scanCxt) |
| 155 |
{ |
| 156 |
/* second time through */ |
| 157 |
so->queueCxt = AllocSetContextCreate(so->giststate->scanCxt, |
| 158 |
"GiST queue context", |
| 159 |
ALLOCSET_DEFAULT_SIZES); |
| 160 |
first_time = false; |
| 161 |
} |
| 162 |
else |
| 163 |
{ |
| 164 |
/* third or later time through */ |
| 165 |
MemoryContextReset(so->queueCxt); |
| 166 |
first_time = false; |
| 167 |
} |
| 168 |
|
| 169 |
/* |
| 170 |
* If we're doing an index-only scan, on the first call, also initialize a |
| 171 |
* tuple descriptor to represent the returned index tuples and create a |
| 172 |
* memory context to hold them during the scan. |
| 173 |
*/ |
| 174 |
if (scan->xs_want_itup && !scan->xs_hitupdesc) |
| 175 |
{ |
| 176 |
int natts; |
| 177 |
int nkeyatts; |
| 178 |
int attno; |
| 179 |
|
| 180 |
/* |
| 181 |
* The storage type of the index can be different from the original |
| 182 |
* datatype being indexed, so we cannot just grab the index's tuple |
| 183 |
* descriptor. Instead, construct a descriptor with the original data |
| 184 |
* types. |
| 185 |
*/ |
| 186 |
natts = RelationGetNumberOfAttributes(scan->indexRelation); |
| 187 |
nkeyatts = IndexRelationGetNumberOfKeyAttributes(scan->indexRelation); |
| 188 |
so->giststate->fetchTupdesc = CreateTemplateTupleDesc(natts); |
| 189 |
for (attno = 1; attno <= nkeyatts; attno++) |
| 190 |
{ |
| 191 |
TupleDescInitEntry(so->giststate->fetchTupdesc, attno, NULL, |
| 192 |
scan->indexRelation->rd_opcintype[attno - 1], |
| 193 |
-1, 0); |
| 194 |
} |
| 195 |
|
| 196 |
for (; attno <= natts; attno++) |
| 197 |
{ |
| 198 |
/* taking opcintype from giststate->tupdesc */ |
| 199 |
TupleDescInitEntry(so->giststate->fetchTupdesc, attno, NULL, |
| 200 |
TupleDescAttr(so->giststate->leafTupdesc, |
| 201 |
attno - 1)->atttypid, |
| 202 |
-1, 0); |
| 203 |
} |
| 204 |
scan->xs_hitupdesc = so->giststate->fetchTupdesc; |
| 205 |
|
| 206 |
/* Also create a memory context that will hold the returned tuples */ |
| 207 |
so->pageDataCxt = AllocSetContextCreate(so->giststate->scanCxt, |
| 208 |
"GiST page data context", |
| 209 |
ALLOCSET_DEFAULT_SIZES); |
| 210 |
} |
| 211 |
|
| 212 |
/* create new, empty pairing heap for search queue */ |
| 213 |
oldCxt = MemoryContextSwitchTo(so->queueCxt); |
| 214 |
so->queue = pairingheap_allocate(pairingheap_GISTSearchItem_cmp, scan); |
| 215 |
MemoryContextSwitchTo(oldCxt); |
| 216 |
|
| 217 |
so->firstCall = true; |
| 218 |
|
| 219 |
/* Update scan key, if a new one is given */ |
| 220 |
if (key && scan->numberOfKeys > 0) |
| 221 |
{ |
| 222 |
void **fn_extras = NULL; |
| 223 |
|
| 224 |
/* |
| 225 |
* If this isn't the first time through, preserve the fn_extra |
| 226 |
* pointers, so that if the consistentFns are using them to cache |
| 227 |
* data, that data is not leaked across a rescan. |
| 228 |
*/ |
| 229 |
if (!first_time) |
| 230 |
{ |
| 231 |
fn_extras = (void **) palloc(scan->numberOfKeys * sizeof(void *)); |
| 232 |
for (i = 0; i < scan->numberOfKeys; i++) |
| 233 |
fn_extras[i] = scan->keyData[i].sk_func.fn_extra; |
| 234 |
} |
| 235 |
|
| 236 |
memcpy(scan->keyData, key, scan->numberOfKeys * sizeof(ScanKeyData)); |
| 237 |
|
| 238 |
/* |
| 239 |
* Modify the scan key so that the Consistent method is called for all |
| 240 |
* comparisons. The original operator is passed to the Consistent |
| 241 |
* function in the form of its strategy number, which is available |
| 242 |
* from the sk_strategy field, and its subtype from the sk_subtype |
| 243 |
* field. |
| 244 |
* |
| 245 |
* Next, if any of keys is a NULL and that key is not marked with |
| 246 |
* SK_SEARCHNULL/SK_SEARCHNOTNULL then nothing can be found (ie, we |
| 247 |
* assume all indexable operators are strict). |
| 248 |
*/ |
| 249 |
so->qual_ok = true; |
| 250 |
|
| 251 |
for (i = 0; i < scan->numberOfKeys; i++) |
| 252 |
{ |
| 253 |
ScanKey skey = scan->keyData + i; |
| 254 |
|
| 255 |
/* |
| 256 |
* Copy consistent support function to ScanKey structure instead |
| 257 |
* of function implementing filtering operator. |
| 258 |
*/ |
| 259 |
fmgr_info_copy(&(skey->sk_func), |
| 260 |
&(so->giststate->consistentFn[skey->sk_attno - 1]), |
| 261 |
so->giststate->scanCxt); |
| 262 |
|
| 263 |
/* Restore prior fn_extra pointers, if not first time */ |
| 264 |
if (!first_time) |
| 265 |
skey->sk_func.fn_extra = fn_extras[i]; |
| 266 |
|
| 267 |
if (skey->sk_flags & SK_ISNULL) |
| 268 |
{ |
| 269 |
if (!(skey->sk_flags & (SK_SEARCHNULL | SK_SEARCHNOTNULL))) |
| 270 |
so->qual_ok = false; |
| 271 |
} |
| 272 |
} |
| 273 |
|
| 274 |
if (!first_time) |
| 275 |
pfree(fn_extras); |
| 276 |
} |
| 277 |
|
| 278 |
/* Update order-by key, if a new one is given */ |
| 279 |
if (orderbys && scan->numberOfOrderBys > 0) |
| 280 |
{ |
| 281 |
void **fn_extras = NULL; |
| 282 |
|
| 283 |
/* As above, preserve fn_extra if not first time through */ |
| 284 |
if (!first_time) |
| 285 |
{ |
| 286 |
fn_extras = (void **) palloc(scan->numberOfOrderBys * sizeof(void *)); |
| 287 |
for (i = 0; i < scan->numberOfOrderBys; i++) |
| 288 |
fn_extras[i] = scan->orderByData[i].sk_func.fn_extra; |
| 289 |
} |
| 290 |
|
| 291 |
memcpy(scan->orderByData, orderbys, scan->numberOfOrderBys * sizeof(ScanKeyData)); |
| 292 |
|
| 293 |
so->orderByTypes = (Oid *) palloc(scan->numberOfOrderBys * sizeof(Oid)); |
| 294 |
|
| 295 |
/* |
| 296 |
* Modify the order-by key so that the Distance method is called for |
| 297 |
* all comparisons. The original operator is passed to the Distance |
| 298 |
* function in the form of its strategy number, which is available |
| 299 |
* from the sk_strategy field, and its subtype from the sk_subtype |
| 300 |
* field. |
| 301 |
*/ |
| 302 |
for (i = 0; i < scan->numberOfOrderBys; i++) |
| 303 |
{ |
| 304 |
ScanKey skey = scan->orderByData + i; |
| 305 |
FmgrInfo *finfo = &(so->giststate->distanceFn[skey->sk_attno - 1]); |
| 306 |
|
| 307 |
/* Check we actually have a distance function ... */ |
| 308 |
if (!OidIsValid(finfo->fn_oid)) |
| 309 |
elog(ERROR, "missing support function %d for attribute %d of index \"%s\"", |
| 310 |
GIST_DISTANCE_PROC, skey->sk_attno, |
| 311 |
RelationGetRelationName(scan->indexRelation)); |
| 312 |
|
| 313 |
/* |
| 314 |
* Look up the datatype returned by the original ordering |
| 315 |
* operator. GiST always uses a float8 for the distance function, |
| 316 |
* but the ordering operator could be anything else. |
| 317 |
* |
| 318 |
* XXX: The distance function is only allowed to be lossy if the |
| 319 |
* ordering operator's result type is float4 or float8. Otherwise |
| 320 |
* we don't know how to return the distance to the executor. But |
| 321 |
* we cannot check that here, as we won't know if the distance |
| 322 |
* function is lossy until it returns *recheck = true for the |
| 323 |
* first time. |
| 324 |
*/ |
| 325 |
so->orderByTypes[i] = get_func_rettype(skey->sk_func.fn_oid); |
| 326 |
|
| 327 |
/* |
| 328 |
* Copy distance support function to ScanKey structure instead of |
| 329 |
* function implementing ordering operator. |
| 330 |
*/ |
| 331 |
fmgr_info_copy(&(skey->sk_func), finfo, so->giststate->scanCxt); |
| 332 |
|
| 333 |
/* Restore prior fn_extra pointers, if not first time */ |
| 334 |
if (!first_time) |
| 335 |
skey->sk_func.fn_extra = fn_extras[i]; |
| 336 |
} |
| 337 |
|
| 338 |
if (!first_time) |
| 339 |
pfree(fn_extras); |
| 340 |
} |
| 341 |
|
| 342 |
/* any previous xs_hitup will have been pfree'd in context resets above */ |
| 343 |
scan->xs_hitup = NULL; |
| 344 |
} |
| 345 |
|
| 346 |
void |
| 347 |
gistendscan(IndexScanDesc scan) |
| 348 |
{ |
| 349 |
GISTScanOpaque so = (GISTScanOpaque) scan->opaque; |
| 350 |
|
| 351 |
/* |
| 352 |
* freeGISTstate is enough to clean up everything made by gistbeginscan, |
| 353 |
* as well as the queueCxt if there is a separate context for it. |
| 354 |
*/ |
| 355 |
freeGISTstate(so->giststate); |
| 356 |
} |