Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * nbtutils.c
4 : * Utility code for Postgres btree implementation.
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/access/nbtree/nbtutils.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 :
16 : #include "postgres.h"
17 :
18 : #include <time.h>
19 :
20 : #include "access/nbtree.h"
21 : #include "access/reloptions.h"
22 : #include "commands/progress.h"
23 : #include "miscadmin.h"
24 : #include "utils/datum.h"
25 : #include "utils/lsyscache.h"
26 :
27 : #define LOOK_AHEAD_REQUIRED_RECHECKS 3
28 : #define LOOK_AHEAD_DEFAULT_DISTANCE 5
29 : #define NSKIPADVANCES_THRESHOLD 3
30 :
31 : static inline int32 _bt_compare_array_skey(FmgrInfo *orderproc,
32 : Datum tupdatum, bool tupnull,
33 : Datum arrdatum, ScanKey cur);
34 : static void _bt_binsrch_skiparray_skey(bool cur_elem_trig, ScanDirection dir,
35 : Datum tupdatum, bool tupnull,
36 : BTArrayKeyInfo *array, ScanKey cur,
37 : int32 *set_elem_result);
38 : static void _bt_skiparray_set_element(Relation rel, ScanKey skey, BTArrayKeyInfo *array,
39 : int32 set_elem_result, Datum tupdatum, bool tupnull);
40 : static void _bt_skiparray_set_isnull(Relation rel, ScanKey skey, BTArrayKeyInfo *array);
41 : static void _bt_array_set_low_or_high(Relation rel, ScanKey skey,
42 : BTArrayKeyInfo *array, bool low_not_high);
43 : static bool _bt_array_decrement(Relation rel, ScanKey skey, BTArrayKeyInfo *array);
44 : static bool _bt_array_increment(Relation rel, ScanKey skey, BTArrayKeyInfo *array);
45 : static bool _bt_advance_array_keys_increment(IndexScanDesc scan, ScanDirection dir,
46 : bool *skip_array_set);
47 : static void _bt_rewind_nonrequired_arrays(IndexScanDesc scan, ScanDirection dir);
48 : static bool _bt_tuple_before_array_skeys(IndexScanDesc scan, ScanDirection dir,
49 : IndexTuple tuple, TupleDesc tupdesc, int tupnatts,
50 : bool readpagetup, int sktrig, bool *scanBehind);
51 : static bool _bt_advance_array_keys(IndexScanDesc scan, BTReadPageState *pstate,
52 : IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
53 : int sktrig, bool sktrig_required);
54 : #ifdef USE_ASSERT_CHECKING
55 : static bool _bt_verify_arrays_bt_first(IndexScanDesc scan, ScanDirection dir);
56 : static bool _bt_verify_keys_with_arraykeys(IndexScanDesc scan);
57 : #endif
58 : static bool _bt_oppodir_checkkeys(IndexScanDesc scan, ScanDirection dir,
59 : IndexTuple finaltup);
60 : static bool _bt_check_compare(IndexScanDesc scan, ScanDirection dir,
61 : IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
62 : bool advancenonrequired, bool forcenonrequired,
63 : bool *continuescan, int *ikey);
64 : static bool _bt_check_rowcompare(ScanKey skey,
65 : IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
66 : ScanDirection dir, bool forcenonrequired, bool *continuescan);
67 : static void _bt_checkkeys_look_ahead(IndexScanDesc scan, BTReadPageState *pstate,
68 : int tupnatts, TupleDesc tupdesc);
69 : static int _bt_keep_natts(Relation rel, IndexTuple lastleft,
70 : IndexTuple firstright, BTScanInsert itup_key);
71 :
72 :
73 : /*
74 : * _bt_mkscankey
75 : * Build an insertion scan key that contains comparison data from itup
76 : * as well as comparator routines appropriate to the key datatypes.
77 : *
78 : * The result is intended for use with _bt_compare() and _bt_truncate().
79 : * Callers that don't need to fill out the insertion scankey arguments
80 : * (e.g. they use an ad-hoc comparison routine, or only need a scankey
81 : * for _bt_truncate()) can pass a NULL index tuple. The scankey will
82 : * be initialized as if an "all truncated" pivot tuple was passed
83 : * instead.
84 : *
85 : * Note that we may occasionally have to share lock the metapage to
86 : * determine whether or not the keys in the index are expected to be
87 : * unique (i.e. if this is a "heapkeyspace" index). We assume a
88 : * heapkeyspace index when caller passes a NULL tuple, allowing index
89 : * build callers to avoid accessing the non-existent metapage. We
90 : * also assume that the index is _not_ allequalimage when a NULL tuple
91 : * is passed; CREATE INDEX callers call _bt_allequalimage() to set the
92 : * field themselves.
93 : */
94 : BTScanInsert
95 11817834 : _bt_mkscankey(Relation rel, IndexTuple itup)
96 : {
97 : BTScanInsert key;
98 : ScanKey skey;
99 : TupleDesc itupdesc;
100 : int indnkeyatts;
101 : int16 *indoption;
102 : int tupnatts;
103 : int i;
104 :
105 11817834 : itupdesc = RelationGetDescr(rel);
106 11817834 : indnkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
107 11817834 : indoption = rel->rd_indoption;
108 11817834 : tupnatts = itup ? BTreeTupleGetNAtts(itup, rel) : 0;
109 :
110 : Assert(tupnatts <= IndexRelationGetNumberOfAttributes(rel));
111 :
112 : /*
113 : * We'll execute search using scan key constructed on key columns.
114 : * Truncated attributes and non-key attributes are omitted from the final
115 : * scan key.
116 : */
117 11817834 : key = palloc(offsetof(BTScanInsertData, scankeys) +
118 11817834 : sizeof(ScanKeyData) * indnkeyatts);
119 11817834 : if (itup)
120 11673000 : _bt_metaversion(rel, &key->heapkeyspace, &key->allequalimage);
121 : else
122 : {
123 : /* Utility statement callers can set these fields themselves */
124 144834 : key->heapkeyspace = true;
125 144834 : key->allequalimage = false;
126 : }
127 11817834 : key->anynullkeys = false; /* initial assumption */
128 11817834 : key->nextkey = false; /* usual case, required by btinsert */
129 11817834 : key->backward = false; /* usual case, required by btinsert */
130 11817834 : key->keysz = Min(indnkeyatts, tupnatts);
131 11817834 : key->scantid = key->heapkeyspace && itup ?
132 23635668 : BTreeTupleGetHeapTID(itup) : NULL;
133 11817834 : skey = key->scankeys;
134 31949802 : for (i = 0; i < indnkeyatts; i++)
135 : {
136 : FmgrInfo *procinfo;
137 : Datum arg;
138 : bool null;
139 : int flags;
140 :
141 : /*
142 : * We can use the cached (default) support procs since no cross-type
143 : * comparison can be needed.
144 : */
145 20131968 : procinfo = index_getprocinfo(rel, i + 1, BTORDER_PROC);
146 :
147 : /*
148 : * Key arguments built from truncated attributes (or when caller
149 : * provides no tuple) are defensively represented as NULL values. They
150 : * should never be used.
151 : */
152 20131968 : if (i < tupnatts)
153 19872556 : arg = index_getattr(itup, i + 1, itupdesc, &null);
154 : else
155 : {
156 259412 : arg = (Datum) 0;
157 259412 : null = true;
158 : }
159 20131968 : flags = (null ? SK_ISNULL : 0) | (indoption[i] << SK_BT_INDOPTION_SHIFT);
160 20131968 : ScanKeyEntryInitializeWithInfo(&skey[i],
161 : flags,
162 20131968 : (AttrNumber) (i + 1),
163 : InvalidStrategy,
164 : InvalidOid,
165 20131968 : rel->rd_indcollation[i],
166 : procinfo,
167 : arg);
168 : /* Record if any key attribute is NULL (or truncated) */
169 20131968 : if (null)
170 280060 : key->anynullkeys = true;
171 : }
172 :
173 : /*
174 : * In NULLS NOT DISTINCT mode, we pretend that there are no null keys, so
175 : * that full uniqueness check is done.
176 : */
177 11817834 : if (rel->rd_index->indnullsnotdistinct)
178 186 : key->anynullkeys = false;
179 :
180 11817834 : return key;
181 : }
182 :
183 : /*
184 : * free a retracement stack made by _bt_search.
185 : */
186 : void
187 21982436 : _bt_freestack(BTStack stack)
188 : {
189 : BTStack ostack;
190 :
191 40348214 : while (stack != NULL)
192 : {
193 18365778 : ostack = stack;
194 18365778 : stack = stack->bts_parent;
195 18365778 : pfree(ostack);
196 : }
197 21982436 : }
198 :
199 : /*
200 : * _bt_compare_array_skey() -- apply array comparison function
201 : *
202 : * Compares caller's tuple attribute value to a scan key/array element.
203 : * Helper function used during binary searches of SK_SEARCHARRAY arrays.
204 : *
205 : * This routine returns:
206 : * <0 if tupdatum < arrdatum;
207 : * 0 if tupdatum == arrdatum;
208 : * >0 if tupdatum > arrdatum.
209 : *
210 : * This is essentially the same interface as _bt_compare: both functions
211 : * compare the value that they're searching for to a binary search pivot.
212 : * However, unlike _bt_compare, this function's "tuple argument" comes first,
213 : * while its "array/scankey argument" comes second.
214 : */
215 : static inline int32
216 512298 : _bt_compare_array_skey(FmgrInfo *orderproc,
217 : Datum tupdatum, bool tupnull,
218 : Datum arrdatum, ScanKey cur)
219 : {
220 512298 : int32 result = 0;
221 :
222 : Assert(cur->sk_strategy == BTEqualStrategyNumber);
223 : Assert(!(cur->sk_flags & (SK_BT_MINVAL | SK_BT_MAXVAL)));
224 :
225 512298 : if (tupnull) /* NULL tupdatum */
226 : {
227 228 : if (cur->sk_flags & SK_ISNULL)
228 132 : result = 0; /* NULL "=" NULL */
229 96 : else if (cur->sk_flags & SK_BT_NULLS_FIRST)
230 0 : result = -1; /* NULL "<" NOT_NULL */
231 : else
232 96 : result = 1; /* NULL ">" NOT_NULL */
233 : }
234 512070 : else if (cur->sk_flags & SK_ISNULL) /* NOT_NULL tupdatum, NULL arrdatum */
235 : {
236 30492 : if (cur->sk_flags & SK_BT_NULLS_FIRST)
237 12 : result = 1; /* NOT_NULL ">" NULL */
238 : else
239 30480 : result = -1; /* NOT_NULL "<" NULL */
240 : }
241 : else
242 : {
243 : /*
244 : * Like _bt_compare, we need to be careful of cross-type comparisons,
245 : * so the left value has to be the value that came from an index tuple
246 : */
247 481578 : result = DatumGetInt32(FunctionCall2Coll(orderproc, cur->sk_collation,
248 : tupdatum, arrdatum));
249 :
250 : /*
251 : * We flip the sign by following the obvious rule: flip whenever the
252 : * column is a DESC column.
253 : *
254 : * _bt_compare does it the wrong way around (flip when *ASC*) in order
255 : * to compensate for passing its orderproc arguments backwards. We
256 : * don't need to play these games because we find it natural to pass
257 : * tupdatum as the left value (and arrdatum as the right value).
258 : */
259 481578 : if (cur->sk_flags & SK_BT_DESC)
260 45552 : INVERT_COMPARE_RESULT(result);
261 : }
262 :
263 512298 : return result;
264 : }
265 :
266 : /*
267 : * _bt_binsrch_array_skey() -- Binary search for next matching array key
268 : *
269 : * Returns an index to the first array element >= caller's tupdatum argument.
270 : * This convention is more natural for forwards scan callers, but that can't
271 : * really matter to backwards scan callers. Both callers require handling for
272 : * the case where the match we return is < tupdatum, and symmetric handling
273 : * for the case where our best match is > tupdatum.
274 : *
275 : * Also sets *set_elem_result to the result _bt_compare_array_skey returned
276 : * when we used it to compare the matching array element to tupdatum/tupnull.
277 : *
278 : * cur_elem_trig indicates if array advancement was triggered by this array's
279 : * scan key, and that the array is for a required scan key. We can apply this
280 : * information to find the next matching array element in the current scan
281 : * direction using far fewer comparisons (fewer on average, compared to naive
282 : * binary search). This scheme takes advantage of an important property of
283 : * required arrays: required arrays always advance in lockstep with the index
284 : * scan's progress through the index's key space.
285 : */
286 : int
287 41654 : _bt_binsrch_array_skey(FmgrInfo *orderproc,
288 : bool cur_elem_trig, ScanDirection dir,
289 : Datum tupdatum, bool tupnull,
290 : BTArrayKeyInfo *array, ScanKey cur,
291 : int32 *set_elem_result)
292 : {
293 41654 : int low_elem = 0,
294 41654 : mid_elem = -1,
295 41654 : high_elem = array->num_elems - 1,
296 41654 : result = 0;
297 : Datum arrdatum;
298 :
299 : Assert(cur->sk_flags & SK_SEARCHARRAY);
300 : Assert(!(cur->sk_flags & SK_BT_SKIP));
301 : Assert(!(cur->sk_flags & SK_ISNULL)); /* SAOP arrays never have NULLs */
302 : Assert(cur->sk_strategy == BTEqualStrategyNumber);
303 :
304 41654 : if (cur_elem_trig)
305 : {
306 : Assert(!ScanDirectionIsNoMovement(dir));
307 : Assert(cur->sk_flags & SK_BT_REQFWD);
308 :
309 : /*
310 : * When the scan key that triggered array advancement is a required
311 : * array scan key, it is now certain that the current array element
312 : * (plus all prior elements relative to the current scan direction)
313 : * cannot possibly be at or ahead of the corresponding tuple value.
314 : * (_bt_checkkeys must have called _bt_tuple_before_array_skeys, which
315 : * makes sure this is true as a condition of advancing the arrays.)
316 : *
317 : * This makes it safe to exclude array elements up to and including
318 : * the former-current array element from our search.
319 : *
320 : * Separately, when array advancement was triggered by a required scan
321 : * key, the array element immediately after the former-current element
322 : * is often either an exact tupdatum match, or a "close by" near-match
323 : * (a near-match tupdatum is one whose key space falls _between_ the
324 : * former-current and new-current array elements). We'll detect both
325 : * cases via an optimistic comparison of the new search lower bound
326 : * (or new search upper bound in the case of backwards scans).
327 : */
328 41336 : if (ScanDirectionIsForward(dir))
329 : {
330 41276 : low_elem = array->cur_elem + 1; /* old cur_elem exhausted */
331 :
332 : /* Compare prospective new cur_elem (also the new lower bound) */
333 41276 : if (high_elem >= low_elem)
334 : {
335 33222 : arrdatum = array->elem_values[low_elem];
336 33222 : result = _bt_compare_array_skey(orderproc, tupdatum, tupnull,
337 : arrdatum, cur);
338 :
339 33222 : if (result <= 0)
340 : {
341 : /* Optimistic comparison optimization worked out */
342 33136 : *set_elem_result = result;
343 33136 : return low_elem;
344 : }
345 86 : mid_elem = low_elem;
346 86 : low_elem++; /* this cur_elem exhausted, too */
347 : }
348 :
349 8140 : if (high_elem < low_elem)
350 : {
351 : /* Caller needs to perform "beyond end" array advancement */
352 8060 : *set_elem_result = 1;
353 8060 : return high_elem;
354 : }
355 : }
356 : else
357 : {
358 60 : high_elem = array->cur_elem - 1; /* old cur_elem exhausted */
359 :
360 : /* Compare prospective new cur_elem (also the new upper bound) */
361 60 : if (high_elem >= low_elem)
362 : {
363 42 : arrdatum = array->elem_values[high_elem];
364 42 : result = _bt_compare_array_skey(orderproc, tupdatum, tupnull,
365 : arrdatum, cur);
366 :
367 42 : if (result >= 0)
368 : {
369 : /* Optimistic comparison optimization worked out */
370 30 : *set_elem_result = result;
371 30 : return high_elem;
372 : }
373 12 : mid_elem = high_elem;
374 12 : high_elem--; /* this cur_elem exhausted, too */
375 : }
376 :
377 30 : if (high_elem < low_elem)
378 : {
379 : /* Caller needs to perform "beyond end" array advancement */
380 30 : *set_elem_result = -1;
381 30 : return low_elem;
382 : }
383 : }
384 : }
385 :
386 698 : while (high_elem > low_elem)
387 : {
388 438 : mid_elem = low_elem + ((high_elem - low_elem) / 2);
389 438 : arrdatum = array->elem_values[mid_elem];
390 :
391 438 : result = _bt_compare_array_skey(orderproc, tupdatum, tupnull,
392 : arrdatum, cur);
393 :
394 438 : if (result == 0)
395 : {
396 : /*
397 : * It's safe to quit as soon as we see an equal array element.
398 : * This often saves an extra comparison or two...
399 : */
400 138 : low_elem = mid_elem;
401 138 : break;
402 : }
403 :
404 300 : if (result > 0)
405 270 : low_elem = mid_elem + 1;
406 : else
407 30 : high_elem = mid_elem;
408 : }
409 :
410 : /*
411 : * ...but our caller also cares about how its searched-for tuple datum
412 : * compares to the low_elem datum. Must always set *set_elem_result with
413 : * the result of that comparison specifically.
414 : */
415 398 : if (low_elem != mid_elem)
416 242 : result = _bt_compare_array_skey(orderproc, tupdatum, tupnull,
417 242 : array->elem_values[low_elem], cur);
418 :
419 398 : *set_elem_result = result;
420 :
421 398 : return low_elem;
422 : }
423 :
424 : /*
425 : * _bt_binsrch_skiparray_skey() -- "Binary search" within a skip array
426 : *
427 : * Does not return an index into the array, since skip arrays don't really
428 : * contain elements (they generate their array elements procedurally instead).
429 : * Our interface matches that of _bt_binsrch_array_skey in every other way.
430 : *
431 : * Sets *set_elem_result just like _bt_binsrch_array_skey would with a true
432 : * array. The value 0 indicates that tupdatum/tupnull is within the range of
433 : * the skip array. We return -1 when tupdatum/tupnull is lower that any value
434 : * within the range of the array, and 1 when it is higher than every value.
435 : * Caller should pass *set_elem_result to _bt_skiparray_set_element to advance
436 : * the array.
437 : *
438 : * cur_elem_trig indicates if array advancement was triggered by this array's
439 : * scan key. We use this to optimize-away comparisons that are known by our
440 : * caller to be unnecessary from context, just like _bt_binsrch_array_skey.
441 : */
442 : static void
443 165498 : _bt_binsrch_skiparray_skey(bool cur_elem_trig, ScanDirection dir,
444 : Datum tupdatum, bool tupnull,
445 : BTArrayKeyInfo *array, ScanKey cur,
446 : int32 *set_elem_result)
447 : {
448 : Assert(cur->sk_flags & SK_BT_SKIP);
449 : Assert(cur->sk_flags & SK_SEARCHARRAY);
450 : Assert(cur->sk_flags & SK_BT_REQFWD);
451 : Assert(array->num_elems == -1);
452 : Assert(!ScanDirectionIsNoMovement(dir));
453 :
454 165498 : if (array->null_elem)
455 : {
456 : Assert(!array->low_compare && !array->high_compare);
457 :
458 141410 : *set_elem_result = 0;
459 141410 : return;
460 : }
461 :
462 24088 : if (tupnull) /* NULL tupdatum */
463 : {
464 24 : if (cur->sk_flags & SK_BT_NULLS_FIRST)
465 0 : *set_elem_result = -1; /* NULL "<" NOT_NULL */
466 : else
467 24 : *set_elem_result = 1; /* NULL ">" NOT_NULL */
468 24 : return;
469 : }
470 :
471 : /*
472 : * Array inequalities determine whether tupdatum is within the range of
473 : * caller's skip array
474 : */
475 24064 : *set_elem_result = 0;
476 24064 : if (ScanDirectionIsForward(dir))
477 : {
478 : /*
479 : * Evaluate low_compare first (unless cur_elem_trig tells us that it
480 : * cannot possibly fail to be satisfied), then evaluate high_compare
481 : */
482 24016 : if (!cur_elem_trig && array->low_compare &&
483 458 : !DatumGetBool(FunctionCall2Coll(&array->low_compare->sk_func,
484 458 : array->low_compare->sk_collation,
485 : tupdatum,
486 458 : array->low_compare->sk_argument)))
487 0 : *set_elem_result = -1;
488 24016 : else if (array->high_compare &&
489 8104 : !DatumGetBool(FunctionCall2Coll(&array->high_compare->sk_func,
490 8104 : array->high_compare->sk_collation,
491 : tupdatum,
492 8104 : array->high_compare->sk_argument)))
493 6400 : *set_elem_result = 1;
494 : }
495 : else
496 : {
497 : /*
498 : * Evaluate high_compare first (unless cur_elem_trig tells us that it
499 : * cannot possibly fail to be satisfied), then evaluate low_compare
500 : */
501 48 : if (!cur_elem_trig && array->high_compare &&
502 6 : !DatumGetBool(FunctionCall2Coll(&array->high_compare->sk_func,
503 6 : array->high_compare->sk_collation,
504 : tupdatum,
505 6 : array->high_compare->sk_argument)))
506 0 : *set_elem_result = 1;
507 48 : else if (array->low_compare &&
508 24 : !DatumGetBool(FunctionCall2Coll(&array->low_compare->sk_func,
509 24 : array->low_compare->sk_collation,
510 : tupdatum,
511 24 : array->low_compare->sk_argument)))
512 0 : *set_elem_result = -1;
513 : }
514 :
515 : /*
516 : * Assert that any keys that were assumed to be satisfied already (due to
517 : * caller passing cur_elem_trig=true) really are satisfied as expected
518 : */
519 : #ifdef USE_ASSERT_CHECKING
520 : if (cur_elem_trig)
521 : {
522 : if (ScanDirectionIsForward(dir) && array->low_compare)
523 : Assert(DatumGetBool(FunctionCall2Coll(&array->low_compare->sk_func,
524 : array->low_compare->sk_collation,
525 : tupdatum,
526 : array->low_compare->sk_argument)));
527 :
528 : if (ScanDirectionIsBackward(dir) && array->high_compare)
529 : Assert(DatumGetBool(FunctionCall2Coll(&array->high_compare->sk_func,
530 : array->high_compare->sk_collation,
531 : tupdatum,
532 : array->high_compare->sk_argument)));
533 : }
534 : #endif
535 : }
536 :
537 : /*
538 : * _bt_skiparray_set_element() -- Set skip array scan key's sk_argument
539 : *
540 : * Caller passes set_elem_result returned by _bt_binsrch_skiparray_skey for
541 : * caller's tupdatum/tupnull.
542 : *
543 : * We copy tupdatum/tupnull into skey's sk_argument iff set_elem_result == 0.
544 : * Otherwise, we set skey to either the lowest or highest value that's within
545 : * the range of caller's skip array (whichever is the best available match to
546 : * tupdatum/tupnull that is still within the range of the skip array according
547 : * to _bt_binsrch_skiparray_skey/set_elem_result).
548 : */
549 : static void
550 155944 : _bt_skiparray_set_element(Relation rel, ScanKey skey, BTArrayKeyInfo *array,
551 : int32 set_elem_result, Datum tupdatum, bool tupnull)
552 : {
553 : Assert(skey->sk_flags & SK_BT_SKIP);
554 : Assert(skey->sk_flags & SK_SEARCHARRAY);
555 :
556 155944 : if (set_elem_result)
557 : {
558 : /* tupdatum/tupnull is out of the range of the skip array */
559 : Assert(!array->null_elem);
560 :
561 640 : _bt_array_set_low_or_high(rel, skey, array, set_elem_result < 0);
562 640 : return;
563 : }
564 :
565 : /* Advance skip array to tupdatum (or tupnull) value */
566 155304 : if (unlikely(tupnull))
567 : {
568 36 : _bt_skiparray_set_isnull(rel, skey, array);
569 36 : return;
570 : }
571 :
572 : /* Free memory previously allocated for sk_argument if needed */
573 155268 : if (!array->attbyval && skey->sk_argument)
574 80496 : pfree(DatumGetPointer(skey->sk_argument));
575 :
576 : /* tupdatum becomes new sk_argument/new current element */
577 155268 : skey->sk_flags &= ~(SK_SEARCHNULL | SK_ISNULL |
578 : SK_BT_MINVAL | SK_BT_MAXVAL |
579 : SK_BT_NEXT | SK_BT_PRIOR);
580 155268 : skey->sk_argument = datumCopy(tupdatum, array->attbyval, array->attlen);
581 : }
582 :
583 : /*
584 : * _bt_skiparray_set_isnull() -- set skip array scan key to NULL
585 : */
586 : static void
587 48 : _bt_skiparray_set_isnull(Relation rel, ScanKey skey, BTArrayKeyInfo *array)
588 : {
589 : Assert(skey->sk_flags & SK_BT_SKIP);
590 : Assert(skey->sk_flags & SK_SEARCHARRAY);
591 : Assert(array->null_elem && !array->low_compare && !array->high_compare);
592 :
593 : /* Free memory previously allocated for sk_argument if needed */
594 48 : if (!array->attbyval && skey->sk_argument)
595 6 : pfree(DatumGetPointer(skey->sk_argument));
596 :
597 : /* NULL becomes new sk_argument/new current element */
598 48 : skey->sk_argument = (Datum) 0;
599 48 : skey->sk_flags &= ~(SK_BT_MINVAL | SK_BT_MAXVAL |
600 : SK_BT_NEXT | SK_BT_PRIOR);
601 48 : skey->sk_flags |= (SK_SEARCHNULL | SK_ISNULL);
602 48 : }
603 :
604 : /*
605 : * _bt_start_array_keys() -- Initialize array keys at start of a scan
606 : *
607 : * Set up the cur_elem counters and fill in the first sk_argument value for
608 : * each array scankey.
609 : */
610 : void
611 79394 : _bt_start_array_keys(IndexScanDesc scan, ScanDirection dir)
612 : {
613 79394 : Relation rel = scan->indexRelation;
614 79394 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
615 :
616 : Assert(so->numArrayKeys);
617 : Assert(so->qual_ok);
618 :
619 159368 : for (int i = 0; i < so->numArrayKeys; i++)
620 : {
621 79974 : BTArrayKeyInfo *array = &so->arrayKeys[i];
622 79974 : ScanKey skey = &so->keyData[array->scan_key];
623 :
624 : Assert(skey->sk_flags & SK_SEARCHARRAY);
625 :
626 79974 : _bt_array_set_low_or_high(rel, skey, array,
627 : ScanDirectionIsForward(dir));
628 : }
629 79394 : so->scanBehind = so->oppositeDirCheck = false; /* reset */
630 79394 : }
631 :
632 : /*
633 : * _bt_array_set_low_or_high() -- Set array scan key to lowest/highest element
634 : *
635 : * Caller also passes associated scan key, which will have its argument set to
636 : * the lowest/highest array value in passing.
637 : */
638 : static void
639 91420 : _bt_array_set_low_or_high(Relation rel, ScanKey skey, BTArrayKeyInfo *array,
640 : bool low_not_high)
641 : {
642 : Assert(skey->sk_flags & SK_SEARCHARRAY);
643 :
644 91420 : if (array->num_elems != -1)
645 : {
646 : /* set low or high element for SAOP array */
647 84636 : int set_elem = 0;
648 :
649 : Assert(!(skey->sk_flags & SK_BT_SKIP));
650 :
651 84636 : if (!low_not_high)
652 8292 : set_elem = array->num_elems - 1;
653 :
654 : /*
655 : * Just copy over array datum (only skip arrays require freeing and
656 : * allocating memory for sk_argument)
657 : */
658 84636 : array->cur_elem = set_elem;
659 84636 : skey->sk_argument = array->elem_values[set_elem];
660 :
661 84636 : return;
662 : }
663 :
664 : /* set low or high element for skip array */
665 : Assert(skey->sk_flags & SK_BT_SKIP);
666 : Assert(array->num_elems == -1);
667 :
668 : /* Free memory previously allocated for sk_argument if needed */
669 6784 : if (!array->attbyval && skey->sk_argument)
670 530 : pfree(DatumGetPointer(skey->sk_argument));
671 :
672 : /* Reset flags */
673 6784 : skey->sk_argument = (Datum) 0;
674 6784 : skey->sk_flags &= ~(SK_SEARCHNULL | SK_ISNULL |
675 : SK_BT_MINVAL | SK_BT_MAXVAL |
676 : SK_BT_NEXT | SK_BT_PRIOR);
677 :
678 6784 : if (array->null_elem &&
679 5298 : (low_not_high == ((skey->sk_flags & SK_BT_NULLS_FIRST) != 0)))
680 : {
681 : /* Requested element (either lowest or highest) has the value NULL */
682 918 : skey->sk_flags |= (SK_SEARCHNULL | SK_ISNULL);
683 : }
684 5866 : else if (low_not_high)
685 : {
686 : /* Setting array to lowest element (according to low_compare) */
687 5150 : skey->sk_flags |= SK_BT_MINVAL;
688 : }
689 : else
690 : {
691 : /* Setting array to highest element (according to high_compare) */
692 716 : skey->sk_flags |= SK_BT_MAXVAL;
693 : }
694 : }
695 :
696 : /*
697 : * _bt_array_decrement() -- decrement array scan key's sk_argument
698 : *
699 : * Return value indicates whether caller's array was successfully decremented.
700 : * Cannot decrement an array whose current element is already the first one.
701 : */
702 : static bool
703 912 : _bt_array_decrement(Relation rel, ScanKey skey, BTArrayKeyInfo *array)
704 : {
705 912 : bool uflow = false;
706 : Datum dec_sk_argument;
707 :
708 : Assert(skey->sk_flags & SK_SEARCHARRAY);
709 : Assert(!(skey->sk_flags & (SK_BT_MAXVAL | SK_BT_NEXT | SK_BT_PRIOR)));
710 :
711 : /* SAOP array? */
712 912 : if (array->num_elems != -1)
713 : {
714 : Assert(!(skey->sk_flags & (SK_BT_SKIP | SK_BT_MINVAL | SK_BT_MAXVAL)));
715 36 : if (array->cur_elem > 0)
716 : {
717 : /*
718 : * Just decrement current element, and assign its datum to skey
719 : * (only skip arrays need us to free existing sk_argument memory)
720 : */
721 6 : array->cur_elem--;
722 6 : skey->sk_argument = array->elem_values[array->cur_elem];
723 :
724 : /* Successfully decremented array */
725 6 : return true;
726 : }
727 :
728 : /* Cannot decrement to before first array element */
729 30 : return false;
730 : }
731 :
732 : /* Nope, this is a skip array */
733 : Assert(skey->sk_flags & SK_BT_SKIP);
734 :
735 : /*
736 : * The sentinel value that represents the minimum value within the range
737 : * of a skip array (often just -inf) is never decrementable
738 : */
739 876 : if (skey->sk_flags & SK_BT_MINVAL)
740 0 : return false;
741 :
742 : /*
743 : * When the current array element is NULL, and the lowest sorting value in
744 : * the index is also NULL, we cannot decrement before first array element
745 : */
746 876 : if ((skey->sk_flags & SK_ISNULL) && (skey->sk_flags & SK_BT_NULLS_FIRST))
747 0 : return false;
748 :
749 : /*
750 : * Opclasses without skip support "decrement" the scan key's current
751 : * element by setting the PRIOR flag. The true prior value is determined
752 : * by repositioning to the last index tuple < existing sk_argument/current
753 : * array element. Note that this works in the usual way when the scan key
754 : * is already marked ISNULL (i.e. when the current element is NULL).
755 : */
756 876 : if (!array->sksup)
757 : {
758 : /* Successfully "decremented" array */
759 12 : skey->sk_flags |= SK_BT_PRIOR;
760 12 : return true;
761 : }
762 :
763 : /*
764 : * Opclasses with skip support directly decrement sk_argument
765 : */
766 864 : if (skey->sk_flags & SK_ISNULL)
767 : {
768 : Assert(!(skey->sk_flags & SK_BT_NULLS_FIRST));
769 :
770 : /*
771 : * Existing sk_argument/array element is NULL (for an IS NULL qual).
772 : *
773 : * "Decrement" from NULL to the high_elem value provided by opclass
774 : * skip support routine.
775 : */
776 6 : skey->sk_flags &= ~(SK_SEARCHNULL | SK_ISNULL);
777 12 : skey->sk_argument = datumCopy(array->sksup->high_elem,
778 6 : array->attbyval, array->attlen);
779 6 : return true;
780 : }
781 :
782 : /*
783 : * Ask opclass support routine to provide decremented copy of existing
784 : * non-NULL sk_argument
785 : */
786 858 : dec_sk_argument = array->sksup->decrement(rel, skey->sk_argument, &uflow);
787 858 : if (unlikely(uflow))
788 : {
789 : /* dec_sk_argument has undefined value (so no pfree) */
790 0 : if (array->null_elem && (skey->sk_flags & SK_BT_NULLS_FIRST))
791 : {
792 0 : _bt_skiparray_set_isnull(rel, skey, array);
793 :
794 : /* Successfully "decremented" array to NULL */
795 0 : return true;
796 : }
797 :
798 : /* Cannot decrement to before first array element */
799 0 : return false;
800 : }
801 :
802 : /*
803 : * Successfully decremented sk_argument to a non-NULL value. Make sure
804 : * that the decremented value is still within the range of the array.
805 : */
806 858 : if (array->low_compare &&
807 12 : !DatumGetBool(FunctionCall2Coll(&array->low_compare->sk_func,
808 12 : array->low_compare->sk_collation,
809 : dec_sk_argument,
810 12 : array->low_compare->sk_argument)))
811 : {
812 : /* Keep existing sk_argument after all */
813 6 : if (!array->attbyval)
814 0 : pfree(DatumGetPointer(dec_sk_argument));
815 :
816 : /* Cannot decrement to before first array element */
817 6 : return false;
818 : }
819 :
820 : /* Accept value returned by opclass decrement callback */
821 852 : if (!array->attbyval && skey->sk_argument)
822 0 : pfree(DatumGetPointer(skey->sk_argument));
823 852 : skey->sk_argument = dec_sk_argument;
824 :
825 : /* Successfully decremented array */
826 852 : return true;
827 : }
828 :
829 : /*
830 : * _bt_array_increment() -- increment array scan key's sk_argument
831 : *
832 : * Return value indicates whether caller's array was successfully incremented.
833 : * Cannot increment an array whose current element is already the final one.
834 : */
835 : static bool
836 31402 : _bt_array_increment(Relation rel, ScanKey skey, BTArrayKeyInfo *array)
837 : {
838 31402 : bool oflow = false;
839 : Datum inc_sk_argument;
840 :
841 : Assert(skey->sk_flags & SK_SEARCHARRAY);
842 : Assert(!(skey->sk_flags & (SK_BT_MINVAL | SK_BT_NEXT | SK_BT_PRIOR)));
843 :
844 : /* SAOP array? */
845 31402 : if (array->num_elems != -1)
846 : {
847 : Assert(!(skey->sk_flags & (SK_BT_SKIP | SK_BT_MINVAL | SK_BT_MAXVAL)));
848 8316 : if (array->cur_elem < array->num_elems - 1)
849 : {
850 : /*
851 : * Just increment current element, and assign its datum to skey
852 : * (only skip arrays need us to free existing sk_argument memory)
853 : */
854 38 : array->cur_elem++;
855 38 : skey->sk_argument = array->elem_values[array->cur_elem];
856 :
857 : /* Successfully incremented array */
858 38 : return true;
859 : }
860 :
861 : /* Cannot increment past final array element */
862 8278 : return false;
863 : }
864 :
865 : /* Nope, this is a skip array */
866 : Assert(skey->sk_flags & SK_BT_SKIP);
867 :
868 : /*
869 : * The sentinel value that represents the maximum value within the range
870 : * of a skip array (often just +inf) is never incrementable
871 : */
872 23086 : if (skey->sk_flags & SK_BT_MAXVAL)
873 640 : return false;
874 :
875 : /*
876 : * When the current array element is NULL, and the highest sorting value
877 : * in the index is also NULL, we cannot increment past the final element
878 : */
879 22446 : if ((skey->sk_flags & SK_ISNULL) && !(skey->sk_flags & SK_BT_NULLS_FIRST))
880 438 : return false;
881 :
882 : /*
883 : * Opclasses without skip support "increment" the scan key's current
884 : * element by setting the NEXT flag. The true next value is determined by
885 : * repositioning to the first index tuple > existing sk_argument/current
886 : * array element. Note that this works in the usual way when the scan key
887 : * is already marked ISNULL (i.e. when the current element is NULL).
888 : */
889 22008 : if (!array->sksup)
890 : {
891 : /* Successfully "incremented" array */
892 14724 : skey->sk_flags |= SK_BT_NEXT;
893 14724 : return true;
894 : }
895 :
896 : /*
897 : * Opclasses with skip support directly increment sk_argument
898 : */
899 7284 : if (skey->sk_flags & SK_ISNULL)
900 : {
901 : Assert(skey->sk_flags & SK_BT_NULLS_FIRST);
902 :
903 : /*
904 : * Existing sk_argument/array element is NULL (for an IS NULL qual).
905 : *
906 : * "Increment" from NULL to the low_elem value provided by opclass
907 : * skip support routine.
908 : */
909 36 : skey->sk_flags &= ~(SK_SEARCHNULL | SK_ISNULL);
910 72 : skey->sk_argument = datumCopy(array->sksup->low_elem,
911 36 : array->attbyval, array->attlen);
912 36 : return true;
913 : }
914 :
915 : /*
916 : * Ask opclass support routine to provide incremented copy of existing
917 : * non-NULL sk_argument
918 : */
919 7248 : inc_sk_argument = array->sksup->increment(rel, skey->sk_argument, &oflow);
920 7248 : if (unlikely(oflow))
921 : {
922 : /* inc_sk_argument has undefined value (so no pfree) */
923 30 : if (array->null_elem && !(skey->sk_flags & SK_BT_NULLS_FIRST))
924 : {
925 12 : _bt_skiparray_set_isnull(rel, skey, array);
926 :
927 : /* Successfully "incremented" array to NULL */
928 12 : return true;
929 : }
930 :
931 : /* Cannot increment past final array element */
932 18 : return false;
933 : }
934 :
935 : /*
936 : * Successfully incremented sk_argument to a non-NULL value. Make sure
937 : * that the incremented value is still within the range of the array.
938 : */
939 7218 : if (array->high_compare &&
940 42 : !DatumGetBool(FunctionCall2Coll(&array->high_compare->sk_func,
941 42 : array->high_compare->sk_collation,
942 : inc_sk_argument,
943 42 : array->high_compare->sk_argument)))
944 : {
945 : /* Keep existing sk_argument after all */
946 12 : if (!array->attbyval)
947 0 : pfree(DatumGetPointer(inc_sk_argument));
948 :
949 : /* Cannot increment past final array element */
950 12 : return false;
951 : }
952 :
953 : /* Accept value returned by opclass increment callback */
954 7206 : if (!array->attbyval && skey->sk_argument)
955 0 : pfree(DatumGetPointer(skey->sk_argument));
956 7206 : skey->sk_argument = inc_sk_argument;
957 :
958 : /* Successfully incremented array */
959 7206 : return true;
960 : }
961 :
962 : /*
963 : * _bt_advance_array_keys_increment() -- Advance to next set of array elements
964 : *
965 : * Advances the array keys by a single increment in the current scan
966 : * direction. When there are multiple array keys this can roll over from the
967 : * lowest order array to higher order arrays.
968 : *
969 : * Returns true if there is another set of values to consider, false if not.
970 : * On true result, the scankeys are initialized with the next set of values.
971 : * On false result, the scankeys stay the same, and the array keys are not
972 : * advanced (every array remains at its final element for scan direction).
973 : */
974 : static bool
975 31172 : _bt_advance_array_keys_increment(IndexScanDesc scan, ScanDirection dir,
976 : bool *skip_array_set)
977 : {
978 31172 : Relation rel = scan->indexRelation;
979 31172 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
980 :
981 : /*
982 : * We must advance the last array key most quickly, since it will
983 : * correspond to the lowest-order index column among the available
984 : * qualifications
985 : */
986 40594 : for (int i = so->numArrayKeys - 1; i >= 0; i--)
987 : {
988 32314 : BTArrayKeyInfo *array = &so->arrayKeys[i];
989 32314 : ScanKey skey = &so->keyData[array->scan_key];
990 :
991 32314 : if (array->num_elems == -1)
992 23962 : *skip_array_set = true;
993 :
994 32314 : if (ScanDirectionIsForward(dir))
995 : {
996 31402 : if (_bt_array_increment(rel, skey, array))
997 22016 : return true;
998 : }
999 : else
1000 : {
1001 912 : if (_bt_array_decrement(rel, skey, array))
1002 876 : return true;
1003 : }
1004 :
1005 : /*
1006 : * Couldn't increment (or decrement) array. Handle array roll over.
1007 : *
1008 : * Start over at the array's lowest sorting value (or its highest
1009 : * value, for backward scans)...
1010 : */
1011 9422 : _bt_array_set_low_or_high(rel, skey, array,
1012 : ScanDirectionIsForward(dir));
1013 :
1014 : /* ...then increment (or decrement) next most significant array */
1015 : }
1016 :
1017 : /*
1018 : * The array keys are now exhausted.
1019 : *
1020 : * Restore the array keys to the state they were in immediately before we
1021 : * were called. This ensures that the arrays only ever ratchet in the
1022 : * current scan direction.
1023 : *
1024 : * Without this, scans could overlook matching tuples when the scan
1025 : * direction gets reversed just before btgettuple runs out of items to
1026 : * return, but just after _bt_readpage prepares all the items from the
1027 : * scan's final page in so->currPos. When we're on the final page it is
1028 : * typical for so->currPos to get invalidated once btgettuple finally
1029 : * returns false, which'll effectively invalidate the scan's array keys.
1030 : * That hasn't happened yet, though -- and in general it may never happen.
1031 : */
1032 8280 : _bt_start_array_keys(scan, -dir);
1033 :
1034 8280 : return false;
1035 : }
1036 :
1037 : /*
1038 : * _bt_rewind_nonrequired_arrays() -- Rewind SAOP arrays not marked required
1039 : *
1040 : * Called when _bt_advance_array_keys decides to start a new primitive index
1041 : * scan on the basis of the current scan position being before the position
1042 : * that _bt_first is capable of repositioning the scan to by applying an
1043 : * inequality operator required in the opposite-to-scan direction only.
1044 : *
1045 : * Although equality strategy scan keys (for both arrays and non-arrays alike)
1046 : * are either marked required in both directions or in neither direction,
1047 : * there is a sense in which non-required arrays behave like required arrays.
1048 : * With a qual such as "WHERE a IN (100, 200) AND b >= 3 AND c IN (5, 6, 7)",
1049 : * the scan key on "c" is non-required, but nevertheless enables positioning
1050 : * the scan at the first tuple >= "(100, 3, 5)" on the leaf level during the
1051 : * first descent of the tree by _bt_first. Later on, there could also be a
1052 : * second descent, that places the scan right before tuples >= "(200, 3, 5)".
1053 : * _bt_first must never be allowed to build an insertion scan key whose "c"
1054 : * entry is set to a value other than 5, the "c" array's first element/value.
1055 : * (Actually, it's the first in the current scan direction. This example uses
1056 : * a forward scan.)
1057 : *
1058 : * Calling here resets the array scan key elements for the scan's non-required
1059 : * arrays. This is strictly necessary for correctness in a subset of cases
1060 : * involving "required in opposite direction"-triggered primitive index scans.
1061 : * Not all callers are at risk of _bt_first using a non-required array like
1062 : * this, but advancement always resets the arrays when another primitive scan
1063 : * is scheduled, just to keep things simple. Array advancement even makes
1064 : * sure to reset non-required arrays during scans that have no inequalities.
1065 : * (Advancement still won't call here when there are no inequalities, though
1066 : * that's just because it's all handled indirectly instead.)
1067 : *
1068 : * Note: _bt_verify_arrays_bt_first is called by an assertion to enforce that
1069 : * everybody got this right.
1070 : *
1071 : * Note: In practice almost all SAOP arrays are marked required during
1072 : * preprocessing (if necessary by generating skip arrays). It is hardly ever
1073 : * truly necessary to call here, but consistently doing so is simpler.
1074 : */
1075 : static void
1076 3326 : _bt_rewind_nonrequired_arrays(IndexScanDesc scan, ScanDirection dir)
1077 : {
1078 3326 : Relation rel = scan->indexRelation;
1079 3326 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
1080 3326 : int arrayidx = 0;
1081 :
1082 9662 : for (int ikey = 0; ikey < so->numberOfKeys; ikey++)
1083 : {
1084 6336 : ScanKey cur = so->keyData + ikey;
1085 6336 : BTArrayKeyInfo *array = NULL;
1086 :
1087 6336 : if (!(cur->sk_flags & SK_SEARCHARRAY) ||
1088 3380 : cur->sk_strategy != BTEqualStrategyNumber)
1089 2956 : continue;
1090 :
1091 3380 : array = &so->arrayKeys[arrayidx++];
1092 : Assert(array->scan_key == ikey);
1093 :
1094 3380 : if ((cur->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)))
1095 3380 : continue;
1096 :
1097 : Assert(array->num_elems != -1); /* No non-required skip arrays */
1098 :
1099 0 : _bt_array_set_low_or_high(rel, cur, array,
1100 : ScanDirectionIsForward(dir));
1101 : }
1102 3326 : }
1103 :
1104 : /*
1105 : * _bt_tuple_before_array_skeys() -- too early to advance required arrays?
1106 : *
1107 : * We always compare the tuple using the current array keys (which we assume
1108 : * are already set in so->keyData[]). readpagetup indicates if tuple is the
1109 : * scan's current _bt_readpage-wise tuple.
1110 : *
1111 : * readpagetup callers must only call here when _bt_check_compare already set
1112 : * continuescan=false. We help these callers deal with _bt_check_compare's
1113 : * inability to distinguishing between the < and > cases (it uses equality
1114 : * operator scan keys, whereas we use 3-way ORDER procs). These callers pass
1115 : * a _bt_check_compare-set sktrig value that indicates which scan key
1116 : * triggered the call (!readpagetup callers just pass us sktrig=0 instead).
1117 : * This information allows us to avoid wastefully checking earlier scan keys
1118 : * that were already deemed to have been satisfied inside _bt_check_compare.
1119 : *
1120 : * Returns false when caller's tuple is >= the current required equality scan
1121 : * keys (or <=, in the case of backwards scans). This happens to readpagetup
1122 : * callers when the scan has reached the point of needing its array keys
1123 : * advanced; caller will need to advance required and non-required arrays at
1124 : * scan key offsets >= sktrig, plus scan keys < sktrig iff sktrig rolls over.
1125 : * (When we return false to readpagetup callers, tuple can only be == current
1126 : * required equality scan keys when caller's sktrig indicates that the arrays
1127 : * need to be advanced due to an unsatisfied required inequality key trigger.)
1128 : *
1129 : * Returns true when caller passes a tuple that is < the current set of
1130 : * equality keys for the most significant non-equal required scan key/column
1131 : * (or > the keys, during backwards scans). This happens to readpagetup
1132 : * callers when tuple is still before the start of matches for the scan's
1133 : * required equality strategy scan keys. (sktrig can't have indicated that an
1134 : * inequality strategy scan key wasn't satisfied in _bt_check_compare when we
1135 : * return true. In fact, we automatically return false when passed such an
1136 : * inequality sktrig by readpagetup callers -- _bt_check_compare's initial
1137 : * continuescan=false doesn't really need to be confirmed here by us.)
1138 : *
1139 : * !readpagetup callers optionally pass us *scanBehind, which tracks whether
1140 : * any missing truncated attributes might have affected array advancement
1141 : * (compared to what would happen if it was shown the first non-pivot tuple on
1142 : * the page to the right of caller's finaltup/high key tuple instead). It's
1143 : * only possible that we'll set *scanBehind to true when caller passes us a
1144 : * pivot tuple (with truncated -inf attributes) that we return false for.
1145 : */
1146 : static bool
1147 358532 : _bt_tuple_before_array_skeys(IndexScanDesc scan, ScanDirection dir,
1148 : IndexTuple tuple, TupleDesc tupdesc, int tupnatts,
1149 : bool readpagetup, int sktrig, bool *scanBehind)
1150 : {
1151 358532 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
1152 :
1153 : Assert(so->numArrayKeys);
1154 : Assert(so->numberOfKeys);
1155 : Assert(sktrig == 0 || readpagetup);
1156 : Assert(!readpagetup || scanBehind == NULL);
1157 :
1158 358532 : if (scanBehind)
1159 87102 : *scanBehind = false;
1160 :
1161 361994 : for (int ikey = sktrig; ikey < so->numberOfKeys; ikey++)
1162 : {
1163 360934 : ScanKey cur = so->keyData + ikey;
1164 : Datum tupdatum;
1165 : bool tupnull;
1166 : int32 result;
1167 :
1168 : /* readpagetup calls require one ORDER proc comparison (at most) */
1169 : Assert(!readpagetup || ikey == sktrig);
1170 :
1171 : /*
1172 : * Once we reach a non-required scan key, we're completely done.
1173 : *
1174 : * Note: we deliberately don't consider the scan direction here.
1175 : * _bt_advance_array_keys caller requires that we track *scanBehind
1176 : * without concern for scan direction.
1177 : */
1178 360934 : if ((cur->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)) == 0)
1179 : {
1180 : Assert(!readpagetup);
1181 : Assert(ikey > sktrig || ikey == 0);
1182 357472 : return false;
1183 : }
1184 :
1185 360934 : if (cur->sk_attno > tupnatts)
1186 : {
1187 : Assert(!readpagetup);
1188 :
1189 : /*
1190 : * When we reach a high key's truncated attribute, assume that the
1191 : * tuple attribute's value is >= the scan's equality constraint
1192 : * scan keys (but set *scanBehind to let interested callers know
1193 : * that a truncated attribute might have affected our answer).
1194 : */
1195 34 : if (scanBehind)
1196 34 : *scanBehind = true;
1197 :
1198 34 : return false;
1199 : }
1200 :
1201 : /*
1202 : * Deal with inequality strategy scan keys that _bt_check_compare set
1203 : * continuescan=false for
1204 : */
1205 360900 : if (cur->sk_strategy != BTEqualStrategyNumber)
1206 : {
1207 : /*
1208 : * When _bt_check_compare indicated that a required inequality
1209 : * scan key wasn't satisfied, there's no need to verify anything;
1210 : * caller always calls _bt_advance_array_keys with this sktrig.
1211 : */
1212 618 : if (readpagetup)
1213 348 : return false;
1214 :
1215 : /*
1216 : * Otherwise we can't give up, since we must check all required
1217 : * scan keys (required in either direction) in order to correctly
1218 : * track *scanBehind for caller
1219 : */
1220 270 : continue;
1221 : }
1222 :
1223 360282 : tupdatum = index_getattr(tuple, cur->sk_attno, tupdesc, &tupnull);
1224 :
1225 360282 : if (likely(!(cur->sk_flags & (SK_BT_MINVAL | SK_BT_MAXVAL))))
1226 : {
1227 : /* Scankey has a valid/comparable sk_argument value */
1228 357014 : result = _bt_compare_array_skey(&so->orderProcs[ikey],
1229 : tupdatum, tupnull,
1230 : cur->sk_argument, cur);
1231 :
1232 357014 : if (result == 0)
1233 : {
1234 : /*
1235 : * Interpret result in a way that takes NEXT/PRIOR into
1236 : * account
1237 : */
1238 17084 : if (cur->sk_flags & SK_BT_NEXT)
1239 13862 : result = -1;
1240 3222 : else if (cur->sk_flags & SK_BT_PRIOR)
1241 30 : result = 1;
1242 :
1243 : Assert(result == 0 || (cur->sk_flags & SK_BT_SKIP));
1244 : }
1245 : }
1246 : else
1247 : {
1248 3268 : BTArrayKeyInfo *array = NULL;
1249 :
1250 : /*
1251 : * Current array element/array = scan key value is a sentinel
1252 : * value that represents the lowest (or highest) possible value
1253 : * that's still within the range of the array.
1254 : *
1255 : * Like _bt_first, we only see MINVAL keys during forwards scans
1256 : * (and similarly only see MAXVAL keys during backwards scans).
1257 : * Even if the scan's direction changes, we'll stop at some higher
1258 : * order key before we can ever reach any MAXVAL (or MINVAL) keys.
1259 : * (However, unlike _bt_first we _can_ get to keys marked either
1260 : * NEXT or PRIOR, regardless of the scan's current direction.)
1261 : */
1262 : Assert(ScanDirectionIsForward(dir) ?
1263 : !(cur->sk_flags & SK_BT_MAXVAL) :
1264 : !(cur->sk_flags & SK_BT_MINVAL));
1265 :
1266 : /*
1267 : * There are no valid sk_argument values in MINVAL/MAXVAL keys.
1268 : * Check if tupdatum is within the range of skip array instead.
1269 : */
1270 3794 : for (int arrayidx = 0; arrayidx < so->numArrayKeys; arrayidx++)
1271 : {
1272 3794 : array = &so->arrayKeys[arrayidx];
1273 3794 : if (array->scan_key == ikey)
1274 3268 : break;
1275 : }
1276 :
1277 3268 : _bt_binsrch_skiparray_skey(false, dir, tupdatum, tupnull,
1278 : array, cur, &result);
1279 :
1280 3268 : if (result == 0)
1281 : {
1282 : /*
1283 : * tupdatum satisfies both low_compare and high_compare, so
1284 : * it's time to advance the array keys.
1285 : *
1286 : * Note: It's possible that the skip array will "advance" from
1287 : * its MINVAL (or MAXVAL) representation to an alternative,
1288 : * logically equivalent representation of the same value: a
1289 : * representation where the = key gets a valid datum in its
1290 : * sk_argument. This is only possible when low_compare uses
1291 : * the >= strategy (or high_compare uses the <= strategy).
1292 : */
1293 3262 : return false;
1294 : }
1295 : }
1296 :
1297 : /*
1298 : * Does this comparison indicate that caller must _not_ advance the
1299 : * scan's arrays just yet?
1300 : */
1301 357020 : if ((ScanDirectionIsForward(dir) && result < 0) ||
1302 3252 : (ScanDirectionIsBackward(dir) && result > 0))
1303 75078 : return true;
1304 :
1305 : /*
1306 : * Does this comparison indicate that caller should now advance the
1307 : * scan's arrays? (Must be if we get here during a readpagetup call.)
1308 : */
1309 281942 : if (readpagetup || result != 0)
1310 : {
1311 : Assert(result != 0);
1312 278750 : return false;
1313 : }
1314 :
1315 : /*
1316 : * Inconclusive -- need to check later scan keys, too.
1317 : *
1318 : * This must be a finaltup precheck, or a call made from an assertion.
1319 : */
1320 : Assert(result == 0);
1321 : }
1322 :
1323 : Assert(!readpagetup);
1324 :
1325 1060 : return false;
1326 : }
1327 :
1328 : /*
1329 : * _bt_start_prim_scan() -- start scheduled primitive index scan?
1330 : *
1331 : * Returns true if _bt_checkkeys scheduled another primitive index scan, just
1332 : * as the last one ended. Otherwise returns false, indicating that the array
1333 : * keys are now fully exhausted.
1334 : *
1335 : * Only call here during scans with one or more equality type array scan keys,
1336 : * after _bt_first or _bt_next return false.
1337 : */
1338 : bool
1339 89054 : _bt_start_prim_scan(IndexScanDesc scan, ScanDirection dir)
1340 : {
1341 89054 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
1342 :
1343 : Assert(so->numArrayKeys);
1344 :
1345 89054 : so->scanBehind = so->oppositeDirCheck = false; /* reset */
1346 :
1347 : /*
1348 : * Array keys are advanced within _bt_checkkeys when the scan reaches the
1349 : * leaf level (more precisely, they're advanced when the scan reaches the
1350 : * end of each distinct set of array elements). This process avoids
1351 : * repeat access to leaf pages (across multiple primitive index scans) by
1352 : * advancing the scan's array keys when it allows the primitive index scan
1353 : * to find nearby matching tuples (or when it eliminates ranges of array
1354 : * key space that can't possibly be satisfied by any index tuple).
1355 : *
1356 : * _bt_checkkeys sets a simple flag variable to schedule another primitive
1357 : * index scan. The flag tells us what to do.
1358 : *
1359 : * We cannot rely on _bt_first always reaching _bt_checkkeys. There are
1360 : * various cases where that won't happen. For example, if the index is
1361 : * completely empty, then _bt_first won't call _bt_readpage/_bt_checkkeys.
1362 : * We also don't expect a call to _bt_checkkeys during searches for a
1363 : * non-existent value that happens to be lower/higher than any existing
1364 : * value in the index.
1365 : *
1366 : * We don't require special handling for these cases -- we don't need to
1367 : * be explicitly instructed to _not_ perform another primitive index scan.
1368 : * It's up to code under the control of _bt_first to always set the flag
1369 : * when another primitive index scan will be required.
1370 : *
1371 : * This works correctly, even with the tricky cases listed above, which
1372 : * all involve access to leaf pages "near the boundaries of the key space"
1373 : * (whether it's from a leftmost/rightmost page, or an imaginary empty
1374 : * leaf root page). If _bt_checkkeys cannot be reached by a primitive
1375 : * index scan for one set of array keys, then it also won't be reached for
1376 : * any later set ("later" in terms of the direction that we scan the index
1377 : * and advance the arrays). The array keys won't have advanced in these
1378 : * cases, but that's the correct behavior (even _bt_advance_array_keys
1379 : * won't always advance the arrays at the point they become "exhausted").
1380 : */
1381 89054 : if (so->needPrimScan)
1382 : {
1383 : Assert(_bt_verify_arrays_bt_first(scan, dir));
1384 :
1385 : /*
1386 : * Flag was set -- must call _bt_first again, which will reset the
1387 : * scan's needPrimScan flag
1388 : */
1389 17960 : return true;
1390 : }
1391 :
1392 : /* The top-level index scan ran out of tuples in this scan direction */
1393 71094 : if (scan->parallel_scan != NULL)
1394 30 : _bt_parallel_done(scan);
1395 :
1396 71094 : return false;
1397 : }
1398 :
1399 : /*
1400 : * _bt_advance_array_keys() -- Advance array elements using a tuple
1401 : *
1402 : * The scan always gets a new qual as a consequence of calling here (except
1403 : * when we determine that the top-level scan has run out of matching tuples).
1404 : * All later _bt_check_compare calls also use the same new qual that was first
1405 : * used here (at least until the next call here advances the keys once again).
1406 : * It's convenient to structure _bt_check_compare rechecks of caller's tuple
1407 : * (using the new qual) as one the steps of advancing the scan's array keys,
1408 : * so this function works as a wrapper around _bt_check_compare.
1409 : *
1410 : * Like _bt_check_compare, we'll set pstate.continuescan on behalf of the
1411 : * caller, and return a boolean indicating if caller's tuple satisfies the
1412 : * scan's new qual. But unlike _bt_check_compare, we set so->needPrimScan
1413 : * when we set continuescan=false, indicating if a new primitive index scan
1414 : * has been scheduled (otherwise, the top-level scan has run out of tuples in
1415 : * the current scan direction).
1416 : *
1417 : * Caller must use _bt_tuple_before_array_skeys to determine if the current
1418 : * place in the scan is >= the current array keys _before_ calling here.
1419 : * We're responsible for ensuring that caller's tuple is <= the newly advanced
1420 : * required array keys once we return. We try to find an exact match, but
1421 : * failing that we'll advance the array keys to whatever set of array elements
1422 : * comes next in the key space for the current scan direction. Required array
1423 : * keys "ratchet forwards" (or backwards). They can only advance as the scan
1424 : * itself advances through the index/key space.
1425 : *
1426 : * (The rules are the same for backwards scans, except that the operators are
1427 : * flipped: just replace the precondition's >= operator with a <=, and the
1428 : * postcondition's <= operator with a >=. In other words, just swap the
1429 : * precondition with the postcondition.)
1430 : *
1431 : * We also deal with "advancing" non-required arrays here (or arrays that are
1432 : * treated as non-required for the duration of a _bt_readpage call). Callers
1433 : * whose sktrig scan key is non-required specify sktrig_required=false. These
1434 : * calls are the only exception to the general rule about always advancing the
1435 : * required array keys (the scan may not even have a required array). These
1436 : * callers should just pass a NULL pstate (since there is never any question
1437 : * of stopping the scan). No call to _bt_tuple_before_array_skeys is required
1438 : * ahead of these calls (it's already clear that any required scan keys must
1439 : * be satisfied by caller's tuple).
1440 : *
1441 : * Note that we deal with non-array required equality strategy scan keys as
1442 : * degenerate single element arrays here. Obviously, they can never really
1443 : * advance in the way that real arrays can, but they must still affect how we
1444 : * advance real array scan keys (exactly like true array equality scan keys).
1445 : * We have to keep around a 3-way ORDER proc for these (using the "=" operator
1446 : * won't do), since in general whether the tuple is < or > _any_ unsatisfied
1447 : * required equality key influences how the scan's real arrays must advance.
1448 : *
1449 : * Note also that we may sometimes need to advance the array keys when the
1450 : * existing required array keys (and other required equality keys) are already
1451 : * an exact match for every corresponding value from caller's tuple. We must
1452 : * do this for inequalities that _bt_check_compare set continuescan=false for.
1453 : * They'll advance the array keys here, just like any other scan key that
1454 : * _bt_check_compare stops on. (This can even happen _after_ we advance the
1455 : * array keys, in which case we'll advance the array keys a second time. That
1456 : * way _bt_checkkeys caller always has its required arrays advance to the
1457 : * maximum possible extent that its tuple will allow.)
1458 : */
1459 : static bool
1460 211190 : _bt_advance_array_keys(IndexScanDesc scan, BTReadPageState *pstate,
1461 : IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
1462 : int sktrig, bool sktrig_required)
1463 : {
1464 211190 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
1465 211190 : Relation rel = scan->indexRelation;
1466 211190 : ScanDirection dir = so->currPos.dir;
1467 211190 : int arrayidx = 0;
1468 211190 : bool beyond_end_advance = false,
1469 211190 : skip_array_advanced = false,
1470 211190 : has_required_opposite_direction_only = false,
1471 211190 : all_required_satisfied = true,
1472 211190 : all_satisfied = true;
1473 :
1474 : Assert(!so->needPrimScan && !so->scanBehind && !so->oppositeDirCheck);
1475 : Assert(_bt_verify_keys_with_arraykeys(scan));
1476 :
1477 211190 : if (sktrig_required)
1478 : {
1479 : /*
1480 : * Precondition array state assertion
1481 : */
1482 : Assert(!_bt_tuple_before_array_skeys(scan, dir, tuple, tupdesc,
1483 : tupnatts, false, 0, NULL));
1484 :
1485 : /*
1486 : * Once we return we'll have a new set of required array keys, so
1487 : * reset state used by "look ahead" optimization
1488 : */
1489 202664 : pstate->rechecks = 0;
1490 202664 : pstate->targetdistance = 0;
1491 : }
1492 8526 : else if (sktrig < so->numberOfKeys - 1 &&
1493 8526 : !(so->keyData[so->numberOfKeys - 1].sk_flags & SK_SEARCHARRAY))
1494 : {
1495 8526 : int least_sign_ikey = so->numberOfKeys - 1;
1496 : bool continuescan;
1497 :
1498 : /*
1499 : * Optimization: perform a precheck of the least significant key
1500 : * during !sktrig_required calls when it isn't already our sktrig
1501 : * (provided the precheck key is not itself an array).
1502 : *
1503 : * When the precheck works out we'll avoid an expensive binary search
1504 : * of sktrig's array (plus any other arrays before least_sign_ikey).
1505 : */
1506 : Assert(so->keyData[sktrig].sk_flags & SK_SEARCHARRAY);
1507 8526 : if (!_bt_check_compare(scan, dir, tuple, tupnatts, tupdesc, false,
1508 : false, &continuescan,
1509 : &least_sign_ikey))
1510 2238 : return false;
1511 : }
1512 :
1513 604248 : for (int ikey = 0; ikey < so->numberOfKeys; ikey++)
1514 : {
1515 401062 : ScanKey cur = so->keyData + ikey;
1516 401062 : BTArrayKeyInfo *array = NULL;
1517 : Datum tupdatum;
1518 401062 : bool required = false,
1519 401062 : required_opposite_direction_only = false,
1520 : tupnull;
1521 : int32 result;
1522 401062 : int set_elem = 0;
1523 :
1524 401062 : if (cur->sk_strategy == BTEqualStrategyNumber)
1525 : {
1526 : /* Manage array state */
1527 352680 : if (cur->sk_flags & SK_SEARCHARRAY)
1528 : {
1529 219290 : array = &so->arrayKeys[arrayidx++];
1530 : Assert(array->scan_key == ikey);
1531 : }
1532 : }
1533 : else
1534 : {
1535 : /*
1536 : * Are any inequalities required in the opposite direction only
1537 : * present here?
1538 : */
1539 48382 : if (((ScanDirectionIsForward(dir) &&
1540 48382 : (cur->sk_flags & (SK_BT_REQBKWD))) ||
1541 0 : (ScanDirectionIsBackward(dir) &&
1542 0 : (cur->sk_flags & (SK_BT_REQFWD)))))
1543 15838 : has_required_opposite_direction_only =
1544 15838 : required_opposite_direction_only = true;
1545 : }
1546 :
1547 : /* Optimization: skip over known-satisfied scan keys */
1548 401062 : if (ikey < sktrig)
1549 76388 : continue;
1550 :
1551 384714 : if (cur->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD))
1552 : {
1553 384714 : required = true;
1554 :
1555 384714 : if (cur->sk_attno > tupnatts)
1556 : {
1557 : /* Set this just like _bt_tuple_before_array_skeys */
1558 : Assert(sktrig < ikey);
1559 2470 : so->scanBehind = true;
1560 : }
1561 : }
1562 :
1563 : /*
1564 : * Handle a required non-array scan key that the initial call to
1565 : * _bt_check_compare indicated triggered array advancement, if any.
1566 : *
1567 : * The non-array scan key's strategy will be <, <=, or = during a
1568 : * forwards scan (or any one of =, >=, or > during a backwards scan).
1569 : * It follows that the corresponding tuple attribute's value must now
1570 : * be either > or >= the scan key value (for backwards scans it must
1571 : * be either < or <= that value).
1572 : *
1573 : * If this is a required equality strategy scan key, this is just an
1574 : * optimization; _bt_tuple_before_array_skeys already confirmed that
1575 : * this scan key places us ahead of caller's tuple. There's no need
1576 : * to repeat that work now. (The same underlying principle also gets
1577 : * applied by the cur_elem_trig optimization used to speed up searches
1578 : * for the next array element.)
1579 : *
1580 : * If this is a required inequality strategy scan key, we _must_ rely
1581 : * on _bt_check_compare like this; we aren't capable of directly
1582 : * evaluating required inequality strategy scan keys here, on our own.
1583 : */
1584 384714 : if (ikey == sktrig && !array)
1585 : {
1586 : Assert(sktrig_required && required && all_required_satisfied);
1587 :
1588 : /* Use "beyond end" advancement. See below for an explanation. */
1589 7422 : beyond_end_advance = true;
1590 7422 : all_satisfied = all_required_satisfied = false;
1591 :
1592 7422 : continue;
1593 : }
1594 :
1595 : /*
1596 : * Nothing more for us to do with an inequality strategy scan key that
1597 : * wasn't the one that _bt_check_compare stopped on, though.
1598 : *
1599 : * Note: if our later call to _bt_check_compare (to recheck caller's
1600 : * tuple) sets continuescan=false due to finding this same inequality
1601 : * unsatisfied (possible when it's required in the scan direction),
1602 : * we'll deal with it via a recursive "second pass" call.
1603 : */
1604 377292 : else if (cur->sk_strategy != BTEqualStrategyNumber)
1605 47812 : continue;
1606 :
1607 : /*
1608 : * Nothing for us to do with an equality strategy scan key that isn't
1609 : * marked required, either -- unless it's a non-required array
1610 : */
1611 329480 : else if (!required && !array)
1612 0 : continue;
1613 :
1614 : /*
1615 : * Here we perform steps for all array scan keys after a required
1616 : * array scan key whose binary search triggered "beyond end of array
1617 : * element" array advancement due to encountering a tuple attribute
1618 : * value > the closest matching array key (or < for backwards scans).
1619 : */
1620 329480 : if (beyond_end_advance)
1621 : {
1622 1414 : if (array)
1623 596 : _bt_array_set_low_or_high(rel, cur, array,
1624 : ScanDirectionIsBackward(dir));
1625 :
1626 1414 : continue;
1627 : }
1628 :
1629 : /*
1630 : * Here we perform steps for all array scan keys after a required
1631 : * array scan key whose tuple attribute was < the closest matching
1632 : * array key when we dealt with it (or > for backwards scans).
1633 : *
1634 : * This earlier required array key already puts us ahead of caller's
1635 : * tuple in the key space (for the current scan direction). We must
1636 : * make sure that subsequent lower-order array keys do not put us too
1637 : * far ahead (ahead of tuples that have yet to be seen by our caller).
1638 : * For example, when a tuple "(a, b) = (42, 5)" advances the array
1639 : * keys on "a" from 40 to 45, we must also set "b" to whatever the
1640 : * first array element for "b" is. It would be wrong to allow "b" to
1641 : * be set based on the tuple value.
1642 : *
1643 : * Perform the same steps with truncated high key attributes. You can
1644 : * think of this as a "binary search" for the element closest to the
1645 : * value -inf. Again, the arrays must never get ahead of the scan.
1646 : */
1647 328066 : if (!all_required_satisfied || cur->sk_attno > tupnatts)
1648 : {
1649 3392 : if (array)
1650 788 : _bt_array_set_low_or_high(rel, cur, array,
1651 : ScanDirectionIsForward(dir));
1652 :
1653 3392 : continue;
1654 : }
1655 :
1656 : /*
1657 : * Search in scankey's array for the corresponding tuple attribute
1658 : * value from caller's tuple
1659 : */
1660 324674 : tupdatum = index_getattr(tuple, cur->sk_attno, tupdesc, &tupnull);
1661 :
1662 324674 : if (array)
1663 : {
1664 203334 : bool cur_elem_trig = (sktrig_required && ikey == sktrig);
1665 :
1666 : /*
1667 : * "Binary search" by checking if tupdatum/tupnull are within the
1668 : * range of the skip array
1669 : */
1670 203334 : if (array->num_elems == -1)
1671 161710 : _bt_binsrch_skiparray_skey(cur_elem_trig, dir,
1672 : tupdatum, tupnull, array, cur,
1673 : &result);
1674 :
1675 : /*
1676 : * Binary search for the closest match from the SAOP array
1677 : */
1678 : else
1679 41624 : set_elem = _bt_binsrch_array_skey(&so->orderProcs[ikey],
1680 : cur_elem_trig, dir,
1681 : tupdatum, tupnull, array, cur,
1682 : &result);
1683 : }
1684 : else
1685 : {
1686 : Assert(required);
1687 :
1688 : /*
1689 : * This is a required non-array equality strategy scan key, which
1690 : * we'll treat as a degenerate single element array.
1691 : *
1692 : * This scan key's imaginary "array" can't really advance, but it
1693 : * can still roll over like any other array. (Actually, this is
1694 : * no different to real single value arrays, which never advance
1695 : * without rolling over -- they can never truly advance, either.)
1696 : */
1697 121340 : result = _bt_compare_array_skey(&so->orderProcs[ikey],
1698 : tupdatum, tupnull,
1699 : cur->sk_argument, cur);
1700 : }
1701 :
1702 : /*
1703 : * Consider "beyond end of array element" array advancement.
1704 : *
1705 : * When the tuple attribute value is > the closest matching array key
1706 : * (or < in the backwards scan case), we need to ratchet this array
1707 : * forward (backward) by one increment, so that caller's tuple ends up
1708 : * being < final array value instead (or > final array value instead).
1709 : * This process has to work for all of the arrays, not just this one:
1710 : * it must "carry" to higher-order arrays when the set_elem that we
1711 : * just found happens to be the final one for the scan's direction.
1712 : * Incrementing (decrementing) set_elem itself isn't good enough.
1713 : *
1714 : * Our approach is to provisionally use set_elem as if it was an exact
1715 : * match now, then set each later/less significant array to whatever
1716 : * its final element is. Once outside the loop we'll then "increment
1717 : * this array's set_elem" by calling _bt_advance_array_keys_increment.
1718 : * That way the process rolls over to higher order arrays as needed.
1719 : *
1720 : * Under this scheme any required arrays only ever ratchet forwards
1721 : * (or backwards), and always do so to the maximum possible extent
1722 : * that we can know will be safe without seeing the scan's next tuple.
1723 : * We don't need any special handling for required scan keys that lack
1724 : * a real array to advance, nor for redundant scan keys that couldn't
1725 : * be eliminated by _bt_preprocess_keys. It won't matter if some of
1726 : * our "true" array scan keys (or even all of them) are non-required.
1727 : */
1728 324674 : if (sktrig_required && required &&
1729 318386 : ((ScanDirectionIsForward(dir) && result > 0) ||
1730 1704 : (ScanDirectionIsBackward(dir) && result < 0)))
1731 23750 : beyond_end_advance = true;
1732 :
1733 : Assert(all_required_satisfied && all_satisfied);
1734 324674 : if (result != 0)
1735 : {
1736 : /*
1737 : * Track whether caller's tuple satisfies our new post-advancement
1738 : * qual, for required scan keys, as well as for the entire set of
1739 : * interesting scan keys (all required scan keys plus non-required
1740 : * array scan keys are considered interesting.)
1741 : */
1742 148466 : all_satisfied = false;
1743 148466 : if (sktrig_required && required)
1744 142700 : all_required_satisfied = false;
1745 : else
1746 : {
1747 : /*
1748 : * There's no need to advance the arrays using the best
1749 : * available match for a non-required array. Give up now.
1750 : * (Though note that sktrig_required calls still have to do
1751 : * all the usual post-advancement steps, including the recheck
1752 : * call to _bt_check_compare.)
1753 : */
1754 : break;
1755 : }
1756 : }
1757 :
1758 : /* Advance array keys, even when we don't have an exact match */
1759 318908 : if (array)
1760 : {
1761 197568 : if (array->num_elems == -1)
1762 : {
1763 : /* Skip array's new element is tupdatum (or MINVAL/MAXVAL) */
1764 155944 : _bt_skiparray_set_element(rel, cur, array, result,
1765 : tupdatum, tupnull);
1766 155944 : skip_array_advanced = true;
1767 : }
1768 41624 : else if (array->cur_elem != set_elem)
1769 : {
1770 : /* SAOP array's new element is set_elem datum */
1771 33492 : array->cur_elem = set_elem;
1772 33492 : cur->sk_argument = array->elem_values[set_elem];
1773 : }
1774 : }
1775 : }
1776 :
1777 : /*
1778 : * Advance the array keys incrementally whenever "beyond end of array
1779 : * element" array advancement happens, so that advancement will carry to
1780 : * higher-order arrays (might exhaust all the scan's arrays instead, which
1781 : * ends the top-level scan).
1782 : */
1783 208952 : if (beyond_end_advance &&
1784 31172 : !_bt_advance_array_keys_increment(scan, dir, &skip_array_advanced))
1785 8280 : goto end_toplevel_scan;
1786 :
1787 : Assert(_bt_verify_keys_with_arraykeys(scan));
1788 :
1789 : /*
1790 : * Maintain a page-level count of the number of times the scan's array
1791 : * keys advanced in a way that affected at least one skip array
1792 : */
1793 200672 : if (sktrig_required && skip_array_advanced)
1794 161956 : pstate->nskipadvances++;
1795 :
1796 : /*
1797 : * Does tuple now satisfy our new qual? Recheck with _bt_check_compare.
1798 : *
1799 : * Calls triggered by an unsatisfied required scan key, whose tuple now
1800 : * satisfies all required scan keys, but not all nonrequired array keys,
1801 : * will still require a recheck call to _bt_check_compare. They'll still
1802 : * need its "second pass" handling of required inequality scan keys.
1803 : * (Might have missed a still-unsatisfied required inequality scan key
1804 : * that caller didn't detect as the sktrig scan key during its initial
1805 : * _bt_check_compare call that used the old/original qual.)
1806 : *
1807 : * Calls triggered by an unsatisfied nonrequired array scan key never need
1808 : * "second pass" handling of required inequalities (nor any other handling
1809 : * of any required scan key). All that matters is whether caller's tuple
1810 : * satisfies the new qual, so it's safe to just skip the _bt_check_compare
1811 : * recheck when we've already determined that it can only return 'false'.
1812 : *
1813 : * Note: In practice most scan keys are marked required by preprocessing,
1814 : * if necessary by generating a preceding skip array. We nevertheless
1815 : * often handle array keys marked required as if they were nonrequired.
1816 : * This behavior is requested by our _bt_check_compare caller, though only
1817 : * when it is passed "forcenonrequired=true" by _bt_checkkeys.
1818 : */
1819 200672 : if ((sktrig_required && all_required_satisfied) ||
1820 148130 : (!sktrig_required && all_satisfied))
1821 : {
1822 53064 : int nsktrig = sktrig + 1;
1823 : bool continuescan;
1824 :
1825 : Assert(all_required_satisfied);
1826 :
1827 : /* Recheck _bt_check_compare on behalf of caller */
1828 53064 : if (_bt_check_compare(scan, dir, tuple, tupnatts, tupdesc, false,
1829 : false, &continuescan,
1830 45298 : &nsktrig) &&
1831 45298 : !so->scanBehind)
1832 : {
1833 : /* This tuple satisfies the new qual */
1834 : Assert(all_satisfied && continuescan);
1835 :
1836 42924 : if (pstate)
1837 42402 : pstate->continuescan = true;
1838 :
1839 43146 : return true;
1840 : }
1841 :
1842 : /*
1843 : * Consider "second pass" handling of required inequalities.
1844 : *
1845 : * It's possible that our _bt_check_compare call indicated that the
1846 : * scan should end due to some unsatisfied inequality that wasn't
1847 : * initially recognized as such by us. Handle this by calling
1848 : * ourselves recursively, this time indicating that the trigger is the
1849 : * inequality that we missed first time around (and using a set of
1850 : * required array/equality keys that are now exact matches for tuple).
1851 : *
1852 : * We make a strong, general guarantee that every _bt_checkkeys call
1853 : * here will advance the array keys to the maximum possible extent
1854 : * that we can know to be safe based on caller's tuple alone. If we
1855 : * didn't perform this step, then that guarantee wouldn't quite hold.
1856 : */
1857 10140 : if (unlikely(!continuescan))
1858 : {
1859 : bool satisfied PG_USED_FOR_ASSERTS_ONLY;
1860 :
1861 : Assert(sktrig_required);
1862 : Assert(so->keyData[nsktrig].sk_strategy != BTEqualStrategyNumber);
1863 :
1864 : /*
1865 : * The tuple must use "beyond end" advancement during the
1866 : * recursive call, so we cannot possibly end up back here when
1867 : * recursing. We'll consume a small, fixed amount of stack space.
1868 : */
1869 : Assert(!beyond_end_advance);
1870 :
1871 : /* Advance the array keys a second time using same tuple */
1872 222 : satisfied = _bt_advance_array_keys(scan, pstate, tuple, tupnatts,
1873 : tupdesc, nsktrig, true);
1874 :
1875 : /* This tuple doesn't satisfy the inequality */
1876 : Assert(!satisfied);
1877 222 : return false;
1878 : }
1879 :
1880 : /*
1881 : * Some non-required scan key (from new qual) still not satisfied.
1882 : *
1883 : * All scan keys required in the current scan direction must still be
1884 : * satisfied, though, so we can trust all_required_satisfied below.
1885 : */
1886 : }
1887 :
1888 : /*
1889 : * When we were called just to deal with "advancing" non-required arrays,
1890 : * this is as far as we can go (cannot stop the scan for these callers)
1891 : */
1892 157526 : if (!sktrig_required)
1893 : {
1894 : /* Caller's tuple doesn't match any qual */
1895 5766 : return false;
1896 : }
1897 :
1898 : /*
1899 : * Postcondition array state assertion (for still-unsatisfied tuples).
1900 : *
1901 : * By here we have established that the scan's required arrays (scan must
1902 : * have at least one required array) advanced, without becoming exhausted.
1903 : *
1904 : * Caller's tuple is now < the newly advanced array keys (or > when this
1905 : * is a backwards scan), except in the case where we only got this far due
1906 : * to an unsatisfied non-required scan key. Verify that with an assert.
1907 : *
1908 : * Note: we don't just quit at this point when all required scan keys were
1909 : * found to be satisfied because we need to consider edge-cases involving
1910 : * scan keys required in the opposite direction only; those aren't tracked
1911 : * by all_required_satisfied.
1912 : */
1913 : Assert(_bt_tuple_before_array_skeys(scan, dir, tuple, tupdesc, tupnatts,
1914 : false, 0, NULL) ==
1915 : !all_required_satisfied);
1916 :
1917 : /*
1918 : * We generally permit primitive index scans to continue onto the next
1919 : * sibling page when the page's finaltup satisfies all required scan keys
1920 : * at the point where we're between pages.
1921 : *
1922 : * If caller's tuple is also the page's finaltup, and we see that required
1923 : * scan keys still aren't satisfied, start a new primitive index scan.
1924 : */
1925 151760 : if (!all_required_satisfied && pstate->finaltup == tuple)
1926 584 : goto new_prim_scan;
1927 :
1928 : /*
1929 : * Proactively check finaltup (don't wait until finaltup is reached by the
1930 : * scan) when it might well turn out to not be satisfied later on.
1931 : *
1932 : * Note: if so->scanBehind hasn't already been set for finaltup by us,
1933 : * it'll be set during this call to _bt_tuple_before_array_skeys. Either
1934 : * way, it'll be set correctly (for the whole page) after this point.
1935 : */
1936 238278 : if (!all_required_satisfied && pstate->finaltup &&
1937 174204 : _bt_tuple_before_array_skeys(scan, dir, pstate->finaltup, tupdesc,
1938 174204 : BTreeTupleGetNAtts(pstate->finaltup, rel),
1939 : false, 0, &so->scanBehind))
1940 17852 : goto new_prim_scan;
1941 :
1942 : /*
1943 : * When we encounter a truncated finaltup high key attribute, we're
1944 : * optimistic about the chances of its corresponding required scan key
1945 : * being satisfied when we go on to recheck it against tuples from this
1946 : * page's right sibling leaf page. We consider truncated attributes to be
1947 : * satisfied by required scan keys, which allows the primitive index scan
1948 : * to continue to the next leaf page. We must set so->scanBehind to true
1949 : * to remember that the last page's finaltup had "satisfied" required scan
1950 : * keys for one or more truncated attribute values (scan keys required in
1951 : * _either_ scan direction).
1952 : *
1953 : * There is a chance that _bt_checkkeys (which checks so->scanBehind) will
1954 : * find that even the sibling leaf page's finaltup is < the new array
1955 : * keys. When that happens, our optimistic policy will have incurred a
1956 : * single extra leaf page access that could have been avoided.
1957 : *
1958 : * A pessimistic policy would give backward scans a gratuitous advantage
1959 : * over forward scans. We'd punish forward scans for applying more
1960 : * accurate information from the high key, rather than just using the
1961 : * final non-pivot tuple as finaltup, in the style of backward scans.
1962 : * Being pessimistic would also give some scans with non-required arrays a
1963 : * perverse advantage over similar scans that use required arrays instead.
1964 : *
1965 : * This is similar to our scan-level heuristics, below. They also set
1966 : * scanBehind to speculatively continue the primscan onto the next page.
1967 : */
1968 133324 : if (so->scanBehind)
1969 : {
1970 : /* Truncated high key -- _bt_scanbehind_checkkeys recheck scheduled */
1971 : }
1972 :
1973 : /*
1974 : * Handle inequalities marked required in the opposite scan direction.
1975 : * They can also signal that we should start a new primitive index scan.
1976 : *
1977 : * It's possible that the scan is now positioned where "matching" tuples
1978 : * begin, and that caller's tuple satisfies all scan keys required in the
1979 : * current scan direction. But if caller's tuple still doesn't satisfy
1980 : * other scan keys that are required in the opposite scan direction only
1981 : * (e.g., a required >= strategy scan key when scan direction is forward),
1982 : * it's still possible that there are many leaf pages before the page that
1983 : * _bt_first could skip straight to. Groveling through all those pages
1984 : * will always give correct answers, but it can be very inefficient. We
1985 : * must avoid needlessly scanning extra pages.
1986 : *
1987 : * Separately, it's possible that _bt_check_compare set continuescan=false
1988 : * for a scan key that's required in the opposite direction only. This is
1989 : * a special case, that happens only when _bt_check_compare sees that the
1990 : * inequality encountered a NULL value. This signals the end of non-NULL
1991 : * values in the current scan direction, which is reason enough to end the
1992 : * (primitive) scan. If this happens at the start of a large group of
1993 : * NULL values, then we shouldn't expect to be called again until after
1994 : * the scan has already read indefinitely-many leaf pages full of tuples
1995 : * with NULL suffix values. (_bt_first is expected to skip over the group
1996 : * of NULLs by applying a similar "deduce NOT NULL" rule of its own, which
1997 : * involves consing up an explicit SK_SEARCHNOTNULL key.)
1998 : *
1999 : * Apply a test against finaltup to detect and recover from the problem:
2000 : * if even finaltup doesn't satisfy such an inequality, we just skip by
2001 : * starting a new primitive index scan. When we skip, we know for sure
2002 : * that all of the tuples on the current page following caller's tuple are
2003 : * also before the _bt_first-wise start of tuples for our new qual. That
2004 : * at least suggests many more skippable pages beyond the current page.
2005 : * (when so->scanBehind and so->oppositeDirCheck are set, this'll happen
2006 : * when we test the next page's finaltup/high key instead.)
2007 : */
2008 130916 : else if (has_required_opposite_direction_only && pstate->finaltup &&
2009 4326 : unlikely(!_bt_oppodir_checkkeys(scan, dir, pstate->finaltup)))
2010 : {
2011 : /*
2012 : * Make sure that any SAOP arrays that were not marked required by
2013 : * preprocessing are reset to their first element for this direction
2014 : */
2015 0 : _bt_rewind_nonrequired_arrays(scan, dir);
2016 0 : goto new_prim_scan;
2017 : }
2018 :
2019 130916 : continue_scan:
2020 :
2021 : /*
2022 : * Stick with the ongoing primitive index scan for now.
2023 : *
2024 : * It's possible that later tuples will also turn out to have values that
2025 : * are still < the now-current array keys (or > the current array keys).
2026 : * Our caller will handle this by performing what amounts to a linear
2027 : * search of the page, implemented by calling _bt_check_compare and then
2028 : * _bt_tuple_before_array_skeys for each tuple.
2029 : *
2030 : * This approach has various advantages over a binary search of the page.
2031 : * Repeated binary searches of the page (one binary search for every array
2032 : * advancement) won't outperform a continuous linear search. While there
2033 : * are workloads that a naive linear search won't handle well, our caller
2034 : * has a "look ahead" fallback mechanism to deal with that problem.
2035 : */
2036 134242 : pstate->continuescan = true; /* Override _bt_check_compare */
2037 134242 : so->needPrimScan = false; /* _bt_readpage has more tuples to check */
2038 :
2039 134242 : if (so->scanBehind)
2040 : {
2041 : /*
2042 : * Remember if recheck needs to call _bt_oppodir_checkkeys for next
2043 : * page's finaltup (see above comments about "Handle inequalities
2044 : * marked required in the opposite scan direction" for why).
2045 : */
2046 3326 : so->oppositeDirCheck = has_required_opposite_direction_only;
2047 :
2048 3326 : _bt_rewind_nonrequired_arrays(scan, dir);
2049 :
2050 : /*
2051 : * skip by setting "look ahead" mechanism's offnum for forwards scans
2052 : * (backwards scans check scanBehind flag directly instead)
2053 : */
2054 3326 : if (ScanDirectionIsForward(dir))
2055 3308 : pstate->skip = pstate->maxoff + 1;
2056 : }
2057 :
2058 : /* Caller's tuple doesn't match the new qual */
2059 134242 : return false;
2060 :
2061 18436 : new_prim_scan:
2062 :
2063 : Assert(pstate->finaltup); /* not on rightmost/leftmost page */
2064 :
2065 : /*
2066 : * Looks like another primitive index scan is required. But consider
2067 : * continuing the current primscan based on scan-level heuristics.
2068 : *
2069 : * Continue the ongoing primitive scan (and schedule a recheck for when
2070 : * the scan arrives on the next sibling leaf page) when it has already
2071 : * read at least one leaf page before the one we're reading now. This
2072 : * makes primscan scheduling more efficient when scanning subsets of an
2073 : * index with many distinct attribute values matching many array elements.
2074 : * It encourages fewer, larger primitive scans where that makes sense.
2075 : * This will in turn encourage _bt_readpage to apply the pstate.startikey
2076 : * optimization more often.
2077 : *
2078 : * Also continue the ongoing primitive index scan when it is still on the
2079 : * first page if there have been more than NSKIPADVANCES_THRESHOLD calls
2080 : * here that each advanced at least one of the scan's skip arrays
2081 : * (deliberately ignore advancements that only affected SAOP arrays here).
2082 : * A page that cycles through this many skip array elements is quite
2083 : * likely to neighbor similar pages, that we'll also need to read.
2084 : *
2085 : * Note: These heuristics aren't as aggressive as you might think. We're
2086 : * conservative about allowing a primitive scan to step from the first
2087 : * leaf page it reads to the page's sibling page (we only allow it on
2088 : * first pages whose finaltup strongly suggests that it'll work out, as
2089 : * well as first pages that have a large number of skip array advances).
2090 : * Clearing this first page finaltup hurdle is a strong signal in itself.
2091 : *
2092 : * Note: The NSKIPADVANCES_THRESHOLD heuristic exists only to avoid
2093 : * pathological cases. Specifically, cases where a skip scan should just
2094 : * behave like a traditional full index scan, but ends up "skipping" again
2095 : * and again, descending to the prior leaf page's direct sibling leaf page
2096 : * each time. This misbehavior would otherwise be possible during scans
2097 : * that never quite manage to "clear the first page finaltup hurdle".
2098 : */
2099 18436 : if (!pstate->firstpage || pstate->nskipadvances > NSKIPADVANCES_THRESHOLD)
2100 : {
2101 : /* Schedule a recheck once on the next (or previous) page */
2102 918 : so->scanBehind = true;
2103 :
2104 : /* Continue the current primitive scan after all */
2105 918 : goto continue_scan;
2106 : }
2107 :
2108 : /*
2109 : * End this primitive index scan, but schedule another.
2110 : *
2111 : * Note: We make a soft assumption that the current scan direction will
2112 : * also be used within _bt_next, when it is asked to step off this page.
2113 : * It is up to _bt_next to cancel this scheduled primitive index scan
2114 : * whenever it steps to a page in the direction opposite currPos.dir.
2115 : */
2116 17518 : pstate->continuescan = false; /* Tell _bt_readpage we're done... */
2117 17518 : so->needPrimScan = true; /* ...but call _bt_first again */
2118 :
2119 17518 : if (scan->parallel_scan)
2120 36 : _bt_parallel_primscan_schedule(scan, so->currPos.currPage);
2121 :
2122 : /* Caller's tuple doesn't match the new qual */
2123 17518 : return false;
2124 :
2125 8280 : end_toplevel_scan:
2126 :
2127 : /*
2128 : * End the current primitive index scan, but don't schedule another.
2129 : *
2130 : * This ends the entire top-level scan in the current scan direction.
2131 : *
2132 : * Note: The scan's arrays (including any non-required arrays) are now in
2133 : * their final positions for the current scan direction. If the scan
2134 : * direction happens to change, then the arrays will already be in their
2135 : * first positions for what will then be the current scan direction.
2136 : */
2137 8280 : pstate->continuescan = false; /* Tell _bt_readpage we're done... */
2138 8280 : so->needPrimScan = false; /* ...and don't call _bt_first again */
2139 :
2140 : /* Caller's tuple doesn't match any qual */
2141 8280 : return false;
2142 : }
2143 :
2144 : #ifdef USE_ASSERT_CHECKING
2145 : /*
2146 : * Verify that the scan's qual state matches what we expect at the point that
2147 : * _bt_start_prim_scan is about to start a just-scheduled new primitive scan.
2148 : *
2149 : * We enforce a rule against non-required array scan keys: they must start out
2150 : * with whatever element is the first for the scan's current scan direction.
2151 : * See _bt_rewind_nonrequired_arrays comments for an explanation.
2152 : */
2153 : static bool
2154 : _bt_verify_arrays_bt_first(IndexScanDesc scan, ScanDirection dir)
2155 : {
2156 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
2157 : int arrayidx = 0;
2158 :
2159 : for (int ikey = 0; ikey < so->numberOfKeys; ikey++)
2160 : {
2161 : ScanKey cur = so->keyData + ikey;
2162 : BTArrayKeyInfo *array = NULL;
2163 : int first_elem_dir;
2164 :
2165 : if (!(cur->sk_flags & SK_SEARCHARRAY) ||
2166 : cur->sk_strategy != BTEqualStrategyNumber)
2167 : continue;
2168 :
2169 : array = &so->arrayKeys[arrayidx++];
2170 :
2171 : if (((cur->sk_flags & SK_BT_REQFWD) && ScanDirectionIsForward(dir)) ||
2172 : ((cur->sk_flags & SK_BT_REQBKWD) && ScanDirectionIsBackward(dir)))
2173 : continue;
2174 :
2175 : if (ScanDirectionIsForward(dir))
2176 : first_elem_dir = 0;
2177 : else
2178 : first_elem_dir = array->num_elems - 1;
2179 :
2180 : if (array->cur_elem != first_elem_dir)
2181 : return false;
2182 : }
2183 :
2184 : return _bt_verify_keys_with_arraykeys(scan);
2185 : }
2186 :
2187 : /*
2188 : * Verify that the scan's "so->keyData[]" scan keys are in agreement with
2189 : * its array key state
2190 : */
2191 : static bool
2192 : _bt_verify_keys_with_arraykeys(IndexScanDesc scan)
2193 : {
2194 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
2195 : int last_sk_attno = InvalidAttrNumber,
2196 : arrayidx = 0;
2197 :
2198 : if (!so->qual_ok)
2199 : return false;
2200 :
2201 : for (int ikey = 0; ikey < so->numberOfKeys; ikey++)
2202 : {
2203 : ScanKey cur = so->keyData + ikey;
2204 : BTArrayKeyInfo *array;
2205 :
2206 : if (cur->sk_strategy != BTEqualStrategyNumber ||
2207 : !(cur->sk_flags & SK_SEARCHARRAY))
2208 : continue;
2209 :
2210 : array = &so->arrayKeys[arrayidx++];
2211 : if (array->scan_key != ikey)
2212 : return false;
2213 :
2214 : if (array->num_elems == 0 || array->num_elems < -1)
2215 : return false;
2216 :
2217 : if (array->num_elems != -1 &&
2218 : cur->sk_argument != array->elem_values[array->cur_elem])
2219 : return false;
2220 : if (last_sk_attno > cur->sk_attno)
2221 : return false;
2222 : last_sk_attno = cur->sk_attno;
2223 : }
2224 :
2225 : if (arrayidx != so->numArrayKeys)
2226 : return false;
2227 :
2228 : return true;
2229 : }
2230 : #endif
2231 :
2232 : /*
2233 : * Test whether an indextuple satisfies all the scankey conditions.
2234 : *
2235 : * Return true if so, false if not. If the tuple fails to pass the qual,
2236 : * we also determine whether there's any need to continue the scan beyond
2237 : * this tuple, and set pstate.continuescan accordingly. See comments for
2238 : * _bt_preprocess_keys() about how this is done.
2239 : *
2240 : * Forward scan callers can pass a high key tuple in the hopes of having
2241 : * us set *continuescan to false, and avoiding an unnecessary visit to
2242 : * the page to the right.
2243 : *
2244 : * Advances the scan's array keys when necessary for arrayKeys=true callers.
2245 : * Scans without any array keys must always pass arrayKeys=false.
2246 : *
2247 : * Also stops and starts primitive index scans for arrayKeys=true callers.
2248 : * Scans with array keys are required to set up page state that helps us with
2249 : * this. The page's finaltup tuple (the page high key for a forward scan, or
2250 : * the page's first non-pivot tuple for a backward scan) must be set in
2251 : * pstate.finaltup ahead of the first call here for the page. Set this to
2252 : * NULL for rightmost page (or the leftmost page for backwards scans).
2253 : *
2254 : * scan: index scan descriptor (containing a search-type scankey)
2255 : * pstate: page level input and output parameters
2256 : * arrayKeys: should we advance the scan's array keys if necessary?
2257 : * tuple: index tuple to test
2258 : * tupnatts: number of attributes in tupnatts (high key may be truncated)
2259 : */
2260 : bool
2261 59747594 : _bt_checkkeys(IndexScanDesc scan, BTReadPageState *pstate, bool arrayKeys,
2262 : IndexTuple tuple, int tupnatts)
2263 : {
2264 59747594 : TupleDesc tupdesc = RelationGetDescr(scan->indexRelation);
2265 59747594 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
2266 59747594 : ScanDirection dir = so->currPos.dir;
2267 59747594 : int ikey = pstate->startikey;
2268 : bool res;
2269 :
2270 : Assert(BTreeTupleGetNAtts(tuple, scan->indexRelation) == tupnatts);
2271 : Assert(!so->needPrimScan && !so->scanBehind && !so->oppositeDirCheck);
2272 : Assert(arrayKeys || so->numArrayKeys == 0);
2273 :
2274 59747594 : res = _bt_check_compare(scan, dir, tuple, tupnatts, tupdesc, arrayKeys,
2275 59747594 : pstate->forcenonrequired, &pstate->continuescan,
2276 : &ikey);
2277 :
2278 : /*
2279 : * If _bt_check_compare relied on the pstate.startikey optimization, call
2280 : * again (in assert-enabled builds) to verify it didn't affect our answer.
2281 : *
2282 : * Note: we can't do this when !pstate.forcenonrequired, since any arrays
2283 : * before pstate.startikey won't have advanced on this page at all.
2284 : */
2285 : Assert(!pstate->forcenonrequired || arrayKeys);
2286 : #ifdef USE_ASSERT_CHECKING
2287 : if (pstate->startikey > 0 && !pstate->forcenonrequired)
2288 : {
2289 : bool dres,
2290 : dcontinuescan;
2291 : int dikey = 0;
2292 :
2293 : /* Pass arrayKeys=false to avoid array side-effects */
2294 : dres = _bt_check_compare(scan, dir, tuple, tupnatts, tupdesc, false,
2295 : pstate->forcenonrequired, &dcontinuescan,
2296 : &dikey);
2297 : Assert(res == dres);
2298 : Assert(pstate->continuescan == dcontinuescan);
2299 :
2300 : /*
2301 : * Should also get the same ikey result. We need a slightly weaker
2302 : * assertion during arrayKeys calls, since they might be using an
2303 : * array that couldn't be marked required during preprocessing.
2304 : */
2305 : Assert(arrayKeys || ikey == dikey);
2306 : Assert(ikey <= dikey);
2307 : }
2308 : #endif
2309 :
2310 : /*
2311 : * Only one _bt_check_compare call is required in the common case where
2312 : * there are no equality strategy array scan keys. Otherwise we can only
2313 : * accept _bt_check_compare's answer unreservedly when it didn't set
2314 : * pstate.continuescan=false.
2315 : */
2316 59747594 : if (!arrayKeys || pstate->continuescan)
2317 59493546 : return res;
2318 :
2319 : /*
2320 : * _bt_check_compare call set continuescan=false in the presence of
2321 : * equality type array keys. This could mean that the tuple is just past
2322 : * the end of matches for the current array keys.
2323 : *
2324 : * It's also possible that the scan is still _before_ the _start_ of
2325 : * tuples matching the current set of array keys. Check for that first.
2326 : */
2327 : Assert(!pstate->forcenonrequired);
2328 254048 : if (_bt_tuple_before_array_skeys(scan, dir, tuple, tupdesc, tupnatts, true,
2329 : ikey, NULL))
2330 : {
2331 : /* Override _bt_check_compare, continue primitive scan */
2332 51606 : pstate->continuescan = true;
2333 :
2334 : /*
2335 : * We will end up here repeatedly given a group of tuples > the
2336 : * previous array keys and < the now-current keys (for a backwards
2337 : * scan it's just the same, though the operators swap positions).
2338 : *
2339 : * We must avoid allowing this linear search process to scan very many
2340 : * tuples from well before the start of tuples matching the current
2341 : * array keys (or from well before the point where we'll once again
2342 : * have to advance the scan's array keys).
2343 : *
2344 : * We keep the overhead under control by speculatively "looking ahead"
2345 : * to later still-unscanned items from this same leaf page. We'll
2346 : * only attempt this once the number of tuples that the linear search
2347 : * process has examined starts to get out of hand.
2348 : */
2349 51606 : pstate->rechecks++;
2350 51606 : if (pstate->rechecks >= LOOK_AHEAD_REQUIRED_RECHECKS)
2351 : {
2352 : /* See if we should skip ahead within the current leaf page */
2353 15404 : _bt_checkkeys_look_ahead(scan, pstate, tupnatts, tupdesc);
2354 :
2355 : /*
2356 : * Might have set pstate.skip to a later page offset. When that
2357 : * happens then _bt_readpage caller will inexpensively skip ahead
2358 : * to a later tuple from the same page (the one just after the
2359 : * tuple we successfully "looked ahead" to).
2360 : */
2361 : }
2362 :
2363 : /* This indextuple doesn't match the current qual, in any case */
2364 51606 : return false;
2365 : }
2366 :
2367 : /*
2368 : * Caller's tuple is >= the current set of array keys and other equality
2369 : * constraint scan keys (or <= if this is a backwards scan). It's now
2370 : * clear that we _must_ advance any required array keys in lockstep with
2371 : * the scan.
2372 : */
2373 202442 : return _bt_advance_array_keys(scan, pstate, tuple, tupnatts, tupdesc,
2374 : ikey, true);
2375 : }
2376 :
2377 : /*
2378 : * Test whether caller's finaltup tuple is still before the start of matches
2379 : * for the current array keys.
2380 : *
2381 : * Called at the start of reading a page during a scan with array keys, though
2382 : * only when the so->scanBehind flag was set on the scan's prior page.
2383 : *
2384 : * Returns false if the tuple is still before the start of matches. When that
2385 : * happens, caller should cut its losses and start a new primitive index scan.
2386 : * Otherwise returns true.
2387 : */
2388 : bool
2389 2802 : _bt_scanbehind_checkkeys(IndexScanDesc scan, ScanDirection dir,
2390 : IndexTuple finaltup)
2391 : {
2392 2802 : Relation rel = scan->indexRelation;
2393 2802 : TupleDesc tupdesc = RelationGetDescr(rel);
2394 2802 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
2395 2802 : int nfinaltupatts = BTreeTupleGetNAtts(finaltup, rel);
2396 :
2397 : Assert(so->numArrayKeys);
2398 :
2399 2802 : if (_bt_tuple_before_array_skeys(scan, dir, finaltup, tupdesc,
2400 : nfinaltupatts, false, 0, NULL))
2401 442 : return false;
2402 :
2403 2360 : if (!so->oppositeDirCheck)
2404 2170 : return true;
2405 :
2406 190 : return _bt_oppodir_checkkeys(scan, dir, finaltup);
2407 : }
2408 :
2409 : /*
2410 : * Test whether an indextuple fails to satisfy an inequality required in the
2411 : * opposite direction only.
2412 : *
2413 : * Caller's finaltup tuple is the page high key (for forwards scans), or the
2414 : * first non-pivot tuple (for backwards scans). Called during scans with
2415 : * required array keys and required opposite-direction inequalities.
2416 : *
2417 : * Returns false if an inequality scan key required in the opposite direction
2418 : * only isn't satisfied (and any earlier required scan keys are satisfied).
2419 : * Otherwise returns true.
2420 : *
2421 : * An unsatisfied inequality required in the opposite direction only might
2422 : * well enable skipping over many leaf pages, provided another _bt_first call
2423 : * takes place. This type of unsatisfied inequality won't usually cause
2424 : * _bt_checkkeys to stop the scan to consider array advancement/starting a new
2425 : * primitive index scan.
2426 : */
2427 : static bool
2428 4516 : _bt_oppodir_checkkeys(IndexScanDesc scan, ScanDirection dir,
2429 : IndexTuple finaltup)
2430 : {
2431 4516 : Relation rel = scan->indexRelation;
2432 4516 : TupleDesc tupdesc = RelationGetDescr(rel);
2433 4516 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
2434 4516 : int nfinaltupatts = BTreeTupleGetNAtts(finaltup, rel);
2435 : bool continuescan;
2436 4516 : ScanDirection flipped = -dir;
2437 4516 : int ikey = 0;
2438 :
2439 : Assert(so->numArrayKeys);
2440 :
2441 4516 : _bt_check_compare(scan, flipped, finaltup, nfinaltupatts, tupdesc, false,
2442 : false, &continuescan,
2443 : &ikey);
2444 :
2445 4516 : if (!continuescan && so->keyData[ikey].sk_strategy != BTEqualStrategyNumber)
2446 0 : return false;
2447 :
2448 4516 : return true;
2449 : }
2450 :
2451 : /*
2452 : * Determines an offset to the first scan key (an so->keyData[]-wise offset)
2453 : * that is _not_ guaranteed to be satisfied by every tuple from pstate.page,
2454 : * which is set in pstate.startikey for _bt_checkkeys calls for the page.
2455 : * This allows caller to save cycles on comparisons of a prefix of keys while
2456 : * reading pstate.page.
2457 : *
2458 : * Also determines if later calls to _bt_checkkeys (for pstate.page) should be
2459 : * forced to treat all required scan keys >= pstate.startikey as nonrequired
2460 : * (that is, if they're to be treated as if any SK_BT_REQFWD/SK_BT_REQBKWD
2461 : * markings that were set by preprocessing were not set at all, for the
2462 : * duration of _bt_checkkeys calls prior to the call for pstate.finaltup).
2463 : * This is indicated to caller by setting pstate.forcenonrequired.
2464 : *
2465 : * Call here at the start of reading a leaf page beyond the first one for the
2466 : * primitive index scan. We consider all non-pivot tuples, so it doesn't make
2467 : * sense to call here when only a subset of those tuples can ever be read.
2468 : * This is also a good idea on performance grounds; not calling here when on
2469 : * the first page (first for the current primitive scan) avoids wasting cycles
2470 : * during selective point queries. They typically don't stand to gain as much
2471 : * when we can set pstate.startikey, and are likely to notice the overhead of
2472 : * calling here. (Also, allowing pstate.forcenonrequired to be set on a
2473 : * primscan's first page would mislead _bt_advance_array_keys, which expects
2474 : * pstate.nskipadvances to be representative of every first page's key space.)
2475 : *
2476 : * Caller must reset startikey and forcenonrequired ahead of the _bt_checkkeys
2477 : * call for pstate.finaltup iff we set forcenonrequired=true. This will give
2478 : * _bt_checkkeys the opportunity to call _bt_advance_array_keys once more,
2479 : * with sktrig_required=true, to advance the arrays that were ignored during
2480 : * checks of all of the page's prior tuples. Caller doesn't need to do this
2481 : * on the rightmost/leftmost page in the index (where pstate.finaltup isn't
2482 : * set), since forcenonrequired won't be set here by us in the first place.
2483 : */
2484 : void
2485 31892 : _bt_set_startikey(IndexScanDesc scan, BTReadPageState *pstate)
2486 : {
2487 31892 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
2488 31892 : Relation rel = scan->indexRelation;
2489 31892 : TupleDesc tupdesc = RelationGetDescr(rel);
2490 : ItemId iid;
2491 : IndexTuple firsttup,
2492 : lasttup;
2493 31892 : int startikey = 0,
2494 31892 : arrayidx = 0,
2495 : firstchangingattnum;
2496 31892 : bool start_past_saop_eq = false;
2497 :
2498 : Assert(!so->scanBehind);
2499 : Assert(pstate->minoff < pstate->maxoff);
2500 : Assert(!pstate->firstpage);
2501 : Assert(pstate->startikey == 0);
2502 : Assert(!so->numArrayKeys || pstate->finaltup ||
2503 : P_RIGHTMOST(BTPageGetOpaque(pstate->page)) ||
2504 : P_LEFTMOST(BTPageGetOpaque(pstate->page)));
2505 :
2506 31892 : if (so->numberOfKeys == 0)
2507 12894 : return;
2508 :
2509 : /* minoff is an offset to the lowest non-pivot tuple on the page */
2510 18998 : iid = PageGetItemId(pstate->page, pstate->minoff);
2511 18998 : firsttup = (IndexTuple) PageGetItem(pstate->page, iid);
2512 :
2513 : /* maxoff is an offset to the highest non-pivot tuple on the page */
2514 18998 : iid = PageGetItemId(pstate->page, pstate->maxoff);
2515 18998 : lasttup = (IndexTuple) PageGetItem(pstate->page, iid);
2516 :
2517 : /* Determine the first attribute whose values change on caller's page */
2518 18998 : firstchangingattnum = _bt_keep_natts_fast(rel, firsttup, lasttup);
2519 :
2520 28824 : for (; startikey < so->numberOfKeys; startikey++)
2521 : {
2522 22064 : ScanKey key = so->keyData + startikey;
2523 : BTArrayKeyInfo *array;
2524 : Datum firstdatum,
2525 : lastdatum;
2526 : bool firstnull,
2527 : lastnull;
2528 : int32 result;
2529 :
2530 : /*
2531 : * Determine if it's safe to set pstate.startikey to an offset to a
2532 : * key that comes after this key, by examining this key
2533 : */
2534 22064 : if (!(key->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)))
2535 : {
2536 : /* Scan key isn't marked required (corner case) */
2537 : Assert(!(key->sk_flags & SK_ROW_HEADER));
2538 12238 : break; /* unsafe */
2539 : }
2540 22064 : if (key->sk_flags & SK_ROW_HEADER)
2541 : {
2542 : /*
2543 : * Can't let pstate.startikey get set to an ikey beyond a
2544 : * RowCompare inequality
2545 : */
2546 0 : break; /* unsafe */
2547 : }
2548 22064 : if (key->sk_strategy != BTEqualStrategyNumber)
2549 : {
2550 : /*
2551 : * Scalar inequality key.
2552 : *
2553 : * It's definitely safe for _bt_checkkeys to avoid assessing this
2554 : * inequality when the page's first and last non-pivot tuples both
2555 : * satisfy the inequality (since the same must also be true of all
2556 : * the tuples in between these two).
2557 : *
2558 : * Unlike the "=" case, it doesn't matter if this attribute has
2559 : * more than one distinct value (though it _is_ necessary for any
2560 : * and all _prior_ attributes to contain no more than one distinct
2561 : * value amongst all of the tuples from pstate.page).
2562 : */
2563 4886 : if (key->sk_attno > firstchangingattnum) /* >, not >= */
2564 432 : break; /* unsafe, preceding attr has multiple
2565 : * distinct values */
2566 :
2567 4454 : firstdatum = index_getattr(firsttup, key->sk_attno, tupdesc, &firstnull);
2568 4454 : lastdatum = index_getattr(lasttup, key->sk_attno, tupdesc, &lastnull);
2569 :
2570 4454 : if (key->sk_flags & SK_ISNULL)
2571 : {
2572 : /* IS NOT NULL key */
2573 : Assert(key->sk_flags & SK_SEARCHNOTNULL);
2574 :
2575 196 : if (firstnull || lastnull)
2576 : break; /* unsafe */
2577 :
2578 : /* Safe, IS NOT NULL key satisfied by every tuple */
2579 9578 : continue;
2580 : }
2581 :
2582 : /* Test firsttup */
2583 4258 : if (firstnull ||
2584 4258 : !DatumGetBool(FunctionCall2Coll(&key->sk_func,
2585 : key->sk_collation, firstdatum,
2586 : key->sk_argument)))
2587 : break; /* unsafe */
2588 :
2589 : /* Test lasttup */
2590 4258 : if (lastnull ||
2591 4258 : !DatumGetBool(FunctionCall2Coll(&key->sk_func,
2592 : key->sk_collation, lastdatum,
2593 : key->sk_argument)))
2594 : break; /* unsafe */
2595 :
2596 : /* Safe, scalar inequality satisfied by every tuple */
2597 4158 : continue;
2598 : }
2599 :
2600 : /* Some = key (could be a scalar = key, could be an array = key) */
2601 : Assert(key->sk_strategy == BTEqualStrategyNumber);
2602 :
2603 17178 : if (!(key->sk_flags & SK_SEARCHARRAY))
2604 : {
2605 : /*
2606 : * Scalar = key (possibly an IS NULL key).
2607 : *
2608 : * It is unsafe to set pstate.startikey to an ikey beyond this
2609 : * key, unless the = key is satisfied by every possible tuple on
2610 : * the page (possible only when attribute has just one distinct
2611 : * value among all tuples on the page).
2612 : */
2613 13664 : if (key->sk_attno >= firstchangingattnum)
2614 11002 : break; /* unsafe, multiple distinct attr values */
2615 :
2616 2662 : firstdatum = index_getattr(firsttup, key->sk_attno, tupdesc,
2617 : &firstnull);
2618 2662 : if (key->sk_flags & SK_ISNULL)
2619 : {
2620 : /* IS NULL key */
2621 : Assert(key->sk_flags & SK_SEARCHNULL);
2622 :
2623 0 : if (!firstnull)
2624 0 : break; /* unsafe */
2625 :
2626 : /* Safe, IS NULL key satisfied by every tuple */
2627 0 : continue;
2628 : }
2629 2662 : if (firstnull ||
2630 2662 : !DatumGetBool(FunctionCall2Coll(&key->sk_func,
2631 : key->sk_collation, firstdatum,
2632 : key->sk_argument)))
2633 : break; /* unsafe */
2634 :
2635 : /* Safe, scalar = key satisfied by every tuple */
2636 2662 : continue;
2637 : }
2638 :
2639 : /* = array key (could be a SAOP array, could be a skip array) */
2640 3514 : array = &so->arrayKeys[arrayidx++];
2641 : Assert(array->scan_key == startikey);
2642 3514 : if (array->num_elems != -1)
2643 : {
2644 : /*
2645 : * SAOP array = key.
2646 : *
2647 : * Handle this like we handle scalar = keys (though binary search
2648 : * for a matching element, to avoid relying on key's sk_argument).
2649 : */
2650 644 : if (key->sk_attno >= firstchangingattnum)
2651 644 : break; /* unsafe, multiple distinct attr values */
2652 :
2653 0 : firstdatum = index_getattr(firsttup, key->sk_attno, tupdesc,
2654 : &firstnull);
2655 0 : _bt_binsrch_array_skey(&so->orderProcs[startikey],
2656 : false, NoMovementScanDirection,
2657 : firstdatum, firstnull, array, key,
2658 : &result);
2659 0 : if (result != 0)
2660 0 : break; /* unsafe */
2661 :
2662 : /* Safe, SAOP = key satisfied by every tuple */
2663 0 : start_past_saop_eq = true;
2664 0 : continue;
2665 : }
2666 :
2667 : /*
2668 : * Skip array = key
2669 : */
2670 : Assert(key->sk_flags & SK_BT_SKIP);
2671 2870 : if (array->null_elem)
2672 : {
2673 : /*
2674 : * Non-range skip array = key.
2675 : *
2676 : * Safe, non-range skip array "satisfied" by every tuple on page
2677 : * (safe even when "key->sk_attno > firstchangingattnum").
2678 : */
2679 2562 : continue;
2680 : }
2681 :
2682 : /*
2683 : * Range skip array = key.
2684 : *
2685 : * Handle this like we handle scalar inequality keys (but avoid using
2686 : * key's sk_argument directly, as in the SAOP array case).
2687 : */
2688 308 : if (key->sk_attno > firstchangingattnum) /* >, not >= */
2689 48 : break; /* unsafe, preceding attr has multiple
2690 : * distinct values */
2691 :
2692 260 : firstdatum = index_getattr(firsttup, key->sk_attno, tupdesc, &firstnull);
2693 260 : lastdatum = index_getattr(lasttup, key->sk_attno, tupdesc, &lastnull);
2694 :
2695 : /* Test firsttup */
2696 260 : _bt_binsrch_skiparray_skey(false, ForwardScanDirection,
2697 : firstdatum, firstnull, array, key,
2698 : &result);
2699 260 : if (result != 0)
2700 0 : break; /* unsafe */
2701 :
2702 : /* Test lasttup */
2703 260 : _bt_binsrch_skiparray_skey(false, ForwardScanDirection,
2704 : lastdatum, lastnull, array, key,
2705 : &result);
2706 260 : if (result != 0)
2707 12 : break; /* unsafe */
2708 :
2709 : /* Safe, range skip array satisfied by every tuple on page */
2710 : }
2711 :
2712 : /*
2713 : * Use of forcenonrequired is typically undesirable, since it'll force
2714 : * _bt_readpage caller to read every tuple on the page -- even though, in
2715 : * general, it might well be possible to end the scan on an earlier tuple.
2716 : * However, caller must use forcenonrequired when start_past_saop_eq=true,
2717 : * since the usual required array behavior might fail to roll over to the
2718 : * SAOP array.
2719 : *
2720 : * We always prefer forcenonrequired=true during scans with skip arrays
2721 : * (except on the first page of each primitive index scan), though -- even
2722 : * when "startikey == 0". That way, _bt_advance_array_keys's low-order
2723 : * key precheck optimization can always be used (unless on the first page
2724 : * of the scan). It seems slightly preferable to check more tuples when
2725 : * that allows us to do significantly less skip array maintenance.
2726 : */
2727 18998 : pstate->forcenonrequired = (start_past_saop_eq || so->skipScan);
2728 18998 : pstate->startikey = startikey;
2729 :
2730 : /*
2731 : * _bt_readpage caller is required to call _bt_checkkeys against page's
2732 : * finaltup with forcenonrequired=false whenever we initially set
2733 : * forcenonrequired=true. That way the scan's arrays will reliably track
2734 : * its progress through the index's key space.
2735 : *
2736 : * We don't expect this when _bt_readpage caller has no finaltup due to
2737 : * its page being the rightmost (or the leftmost, during backwards scans).
2738 : * When we see that _bt_readpage has no finaltup, back out of everything.
2739 : */
2740 : Assert(!pstate->forcenonrequired || so->numArrayKeys);
2741 18998 : if (pstate->forcenonrequired && !pstate->finaltup)
2742 : {
2743 480 : pstate->forcenonrequired = false;
2744 480 : pstate->startikey = 0;
2745 : }
2746 : }
2747 :
2748 : /*
2749 : * Test whether an indextuple satisfies current scan condition.
2750 : *
2751 : * Return true if so, false if not. If not, also sets *continuescan to false
2752 : * when it's also not possible for any later tuples to pass the current qual
2753 : * (with the scan's current set of array keys, in the current scan direction),
2754 : * in addition to setting *ikey to the so->keyData[] subscript/offset for the
2755 : * unsatisfied scan key (needed when caller must consider advancing the scan's
2756 : * array keys).
2757 : *
2758 : * This is a subroutine for _bt_checkkeys. We provisionally assume that
2759 : * reaching the end of the current set of required keys (in particular the
2760 : * current required array keys) ends the ongoing (primitive) index scan.
2761 : * Callers without array keys should just end the scan right away when they
2762 : * find that continuescan has been set to false here by us. Things are more
2763 : * complicated for callers with array keys.
2764 : *
2765 : * Callers with array keys must first consider advancing the arrays when
2766 : * continuescan has been set to false here by us. They must then consider if
2767 : * it really does make sense to end the current (primitive) index scan, in
2768 : * light of everything that is known at that point. (In general when we set
2769 : * continuescan=false for these callers it must be treated as provisional.)
2770 : *
2771 : * We deal with advancing unsatisfied non-required arrays directly, though.
2772 : * This is safe, since by definition non-required keys can't end the scan.
2773 : * This is just how we determine if non-required arrays are just unsatisfied
2774 : * by the current array key, or if they're truly unsatisfied (that is, if
2775 : * they're unsatisfied by every possible array key).
2776 : *
2777 : * Pass advancenonrequired=false to avoid all array related side effects.
2778 : * This allows _bt_advance_array_keys caller to avoid infinite recursion.
2779 : *
2780 : * Pass forcenonrequired=true to instruct us to treat all keys as nonrequired.
2781 : * This is used to make it safe to temporarily stop properly maintaining the
2782 : * scan's required arrays. _bt_checkkeys caller (_bt_readpage, actually)
2783 : * determines a prefix of keys that must satisfy every possible corresponding
2784 : * index attribute value from its page, which is passed to us via *ikey arg
2785 : * (this is the first key that might be unsatisfied by tuples on the page).
2786 : * Obviously, we won't maintain any array keys from before *ikey, so it's
2787 : * quite possible for such arrays to "fall behind" the index's keyspace.
2788 : * Caller will need to "catch up" by passing forcenonrequired=true (alongside
2789 : * an *ikey=0) once the page's finaltup is reached.
2790 : *
2791 : * Note: it's safe to pass an *ikey > 0 with forcenonrequired=false, but only
2792 : * when caller determines that it won't affect array maintenance.
2793 : */
2794 : static bool
2795 59813700 : _bt_check_compare(IndexScanDesc scan, ScanDirection dir,
2796 : IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
2797 : bool advancenonrequired, bool forcenonrequired,
2798 : bool *continuescan, int *ikey)
2799 : {
2800 59813700 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
2801 :
2802 : Assert(!forcenonrequired || advancenonrequired);
2803 :
2804 59813700 : *continuescan = true; /* default assumption */
2805 :
2806 114593122 : for (; *ikey < so->numberOfKeys; (*ikey)++)
2807 : {
2808 67029882 : ScanKey key = so->keyData + *ikey;
2809 : Datum datum;
2810 : bool isNull;
2811 67029882 : bool requiredSameDir = false,
2812 67029882 : requiredOppositeDirOnly = false;
2813 :
2814 : /*
2815 : * Check if the key is required in the current scan direction, in the
2816 : * opposite scan direction _only_, or in neither direction (except
2817 : * when we're forced to treat all scan keys as nonrequired)
2818 : */
2819 67029882 : if (forcenonrequired)
2820 : {
2821 : /* treating scan's keys as non-required */
2822 : }
2823 66579358 : else if (((key->sk_flags & SK_BT_REQFWD) && ScanDirectionIsForward(dir)) ||
2824 14099562 : ((key->sk_flags & SK_BT_REQBKWD) && ScanDirectionIsBackward(dir)))
2825 52504596 : requiredSameDir = true;
2826 14074762 : else if (((key->sk_flags & SK_BT_REQFWD) && ScanDirectionIsBackward(dir)) ||
2827 5445962 : ((key->sk_flags & SK_BT_REQBKWD) && ScanDirectionIsForward(dir)))
2828 14074762 : requiredOppositeDirOnly = true;
2829 :
2830 67029882 : if (key->sk_attno > tupnatts)
2831 : {
2832 : /*
2833 : * This attribute is truncated (must be high key). The value for
2834 : * this attribute in the first non-pivot tuple on the page to the
2835 : * right could be any possible value. Assume that truncated
2836 : * attribute passes the qual.
2837 : */
2838 : Assert(BTreeTupleIsPivot(tuple));
2839 18896634 : continue;
2840 : }
2841 :
2842 : /*
2843 : * A skip array scan key uses one of several sentinel values. We just
2844 : * fall back on _bt_tuple_before_array_skeys when we see such a value.
2845 : */
2846 67027412 : if (key->sk_flags & (SK_BT_MINVAL | SK_BT_MAXVAL |
2847 : SK_BT_NEXT | SK_BT_PRIOR))
2848 : {
2849 : Assert(key->sk_flags & SK_SEARCHARRAY);
2850 : Assert(key->sk_flags & SK_BT_SKIP);
2851 : Assert(requiredSameDir || forcenonrequired);
2852 :
2853 : /*
2854 : * Cannot fall back on _bt_tuple_before_array_skeys when we're
2855 : * treating the scan's keys as nonrequired, though. Just handle
2856 : * this like any other non-required equality-type array key.
2857 : */
2858 33574 : if (forcenonrequired)
2859 12250460 : return _bt_advance_array_keys(scan, NULL, tuple, tupnatts,
2860 : tupdesc, *ikey, false);
2861 :
2862 31546 : *continuescan = false;
2863 31546 : return false;
2864 : }
2865 :
2866 : /* row-comparison keys need special processing */
2867 66993838 : if (key->sk_flags & SK_ROW_HEADER)
2868 : {
2869 2448 : if (_bt_check_rowcompare(key, tuple, tupnatts, tupdesc, dir,
2870 : forcenonrequired, continuescan))
2871 2382 : continue;
2872 66 : return false;
2873 : }
2874 :
2875 66991390 : datum = index_getattr(tuple,
2876 66991390 : key->sk_attno,
2877 : tupdesc,
2878 : &isNull);
2879 :
2880 66991390 : if (key->sk_flags & SK_ISNULL)
2881 : {
2882 : /* Handle IS NULL/NOT NULL tests */
2883 18909506 : if (key->sk_flags & SK_SEARCHNULL)
2884 : {
2885 18080 : if (isNull)
2886 428 : continue; /* tuple satisfies this qual */
2887 : }
2888 : else
2889 : {
2890 : Assert(key->sk_flags & SK_SEARCHNOTNULL);
2891 : Assert(!(key->sk_flags & SK_BT_SKIP));
2892 18891426 : if (!isNull)
2893 18891354 : continue; /* tuple satisfies this qual */
2894 : }
2895 :
2896 : /*
2897 : * Tuple fails this qual. If it's a required qual for the current
2898 : * scan direction, then we can conclude no further tuples will
2899 : * pass, either.
2900 : */
2901 17724 : if (requiredSameDir)
2902 156 : *continuescan = false;
2903 17568 : else if (unlikely(key->sk_flags & SK_BT_SKIP))
2904 : {
2905 : /*
2906 : * If we're treating scan keys as nonrequired, and encounter a
2907 : * skip array scan key whose current element is NULL, then it
2908 : * must be a non-range skip array. It must be satisfied, so
2909 : * there's no need to call _bt_advance_array_keys to check.
2910 : */
2911 : Assert(forcenonrequired && *ikey > 0);
2912 0 : continue;
2913 : }
2914 :
2915 : /*
2916 : * This indextuple doesn't match the qual.
2917 : */
2918 17724 : return false;
2919 : }
2920 :
2921 48081884 : if (isNull)
2922 : {
2923 228 : if (key->sk_flags & SK_BT_NULLS_FIRST)
2924 : {
2925 : /*
2926 : * Since NULLs are sorted before non-NULLs, we know we have
2927 : * reached the lower limit of the range of values for this
2928 : * index attr. On a backward scan, we can stop if this qual
2929 : * is one of the "must match" subset. We can stop regardless
2930 : * of whether the qual is > or <, so long as it's required,
2931 : * because it's not possible for any future tuples to pass. On
2932 : * a forward scan, however, we must keep going, because we may
2933 : * have initially positioned to the start of the index.
2934 : * (_bt_advance_array_keys also relies on this behavior during
2935 : * forward scans.)
2936 : */
2937 0 : if ((requiredSameDir || requiredOppositeDirOnly) &&
2938 : ScanDirectionIsBackward(dir))
2939 0 : *continuescan = false;
2940 : }
2941 : else
2942 : {
2943 : /*
2944 : * Since NULLs are sorted after non-NULLs, we know we have
2945 : * reached the upper limit of the range of values for this
2946 : * index attr. On a forward scan, we can stop if this qual is
2947 : * one of the "must match" subset. We can stop regardless of
2948 : * whether the qual is > or <, so long as it's required,
2949 : * because it's not possible for any future tuples to pass. On
2950 : * a backward scan, however, we must keep going, because we
2951 : * may have initially positioned to the end of the index.
2952 : * (_bt_advance_array_keys also relies on this behavior during
2953 : * backward scans.)
2954 : */
2955 228 : if ((requiredSameDir || requiredOppositeDirOnly) &&
2956 : ScanDirectionIsForward(dir))
2957 222 : *continuescan = false;
2958 : }
2959 :
2960 : /*
2961 : * In any case, this indextuple doesn't match the qual.
2962 : */
2963 228 : return false;
2964 : }
2965 :
2966 48081656 : if (!DatumGetBool(FunctionCall2Coll(&key->sk_func, key->sk_collation,
2967 : datum, key->sk_argument)))
2968 : {
2969 : /*
2970 : * Tuple fails this qual. If it's a required qual for the current
2971 : * scan direction, then we can conclude no further tuples will
2972 : * pass, either.
2973 : *
2974 : * Note: because we stop the scan as soon as any required equality
2975 : * qual fails, it is critical that equality quals be used for the
2976 : * initial positioning in _bt_first() when they are available. See
2977 : * comments in _bt_first().
2978 : */
2979 12198868 : if (requiredSameDir)
2980 11810196 : *continuescan = false;
2981 :
2982 : /*
2983 : * If this is a non-required equality-type array key, the tuple
2984 : * needs to be checked against every possible array key. Handle
2985 : * this by "advancing" the scan key's array to a matching value
2986 : * (if we're successful then the tuple might match the qual).
2987 : */
2988 388672 : else if (advancenonrequired &&
2989 381128 : key->sk_strategy == BTEqualStrategyNumber &&
2990 291916 : (key->sk_flags & SK_SEARCHARRAY))
2991 6498 : return _bt_advance_array_keys(scan, NULL, tuple, tupnatts,
2992 : tupdesc, *ikey, false);
2993 :
2994 : /*
2995 : * This indextuple doesn't match the qual.
2996 : */
2997 12192370 : return false;
2998 : }
2999 : }
3000 :
3001 : /* If we get here, the tuple passes all index quals. */
3002 47563240 : return true;
3003 : }
3004 :
3005 : /*
3006 : * Test whether an indextuple satisfies a row-comparison scan condition.
3007 : *
3008 : * Return true if so, false if not. If not, also clear *continuescan if
3009 : * it's not possible for any future tuples in the current scan direction
3010 : * to pass the qual.
3011 : *
3012 : * This is a subroutine for _bt_checkkeys/_bt_check_compare.
3013 : */
3014 : static bool
3015 2448 : _bt_check_rowcompare(ScanKey skey, IndexTuple tuple, int tupnatts,
3016 : TupleDesc tupdesc, ScanDirection dir,
3017 : bool forcenonrequired, bool *continuescan)
3018 : {
3019 2448 : ScanKey subkey = (ScanKey) DatumGetPointer(skey->sk_argument);
3020 2448 : int32 cmpresult = 0;
3021 : bool result;
3022 :
3023 : /* First subkey should be same as the header says */
3024 : Assert(subkey->sk_attno == skey->sk_attno);
3025 :
3026 : /* Loop over columns of the row condition */
3027 : for (;;)
3028 240 : {
3029 : Datum datum;
3030 : bool isNull;
3031 :
3032 : Assert(subkey->sk_flags & SK_ROW_MEMBER);
3033 :
3034 2688 : if (subkey->sk_attno > tupnatts)
3035 : {
3036 : /*
3037 : * This attribute is truncated (must be high key). The value for
3038 : * this attribute in the first non-pivot tuple on the page to the
3039 : * right could be any possible value. Assume that truncated
3040 : * attribute passes the qual.
3041 : */
3042 : Assert(BTreeTupleIsPivot(tuple));
3043 6 : cmpresult = 0;
3044 6 : if (subkey->sk_flags & SK_ROW_END)
3045 6 : break;
3046 0 : subkey++;
3047 0 : continue;
3048 : }
3049 :
3050 2682 : datum = index_getattr(tuple,
3051 2682 : subkey->sk_attno,
3052 : tupdesc,
3053 : &isNull);
3054 :
3055 2682 : if (isNull)
3056 : {
3057 48 : if (forcenonrequired)
3058 : {
3059 : /* treating scan's keys as non-required */
3060 : }
3061 48 : else if (subkey->sk_flags & SK_BT_NULLS_FIRST)
3062 : {
3063 : /*
3064 : * Since NULLs are sorted before non-NULLs, we know we have
3065 : * reached the lower limit of the range of values for this
3066 : * index attr. On a backward scan, we can stop if this qual
3067 : * is one of the "must match" subset. We can stop regardless
3068 : * of whether the qual is > or <, so long as it's required,
3069 : * because it's not possible for any future tuples to pass. On
3070 : * a forward scan, however, we must keep going, because we may
3071 : * have initially positioned to the start of the index.
3072 : * (_bt_advance_array_keys also relies on this behavior during
3073 : * forward scans.)
3074 : */
3075 0 : if ((subkey->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)) &&
3076 : ScanDirectionIsBackward(dir))
3077 0 : *continuescan = false;
3078 : }
3079 : else
3080 : {
3081 : /*
3082 : * Since NULLs are sorted after non-NULLs, we know we have
3083 : * reached the upper limit of the range of values for this
3084 : * index attr. On a forward scan, we can stop if this qual is
3085 : * one of the "must match" subset. We can stop regardless of
3086 : * whether the qual is > or <, so long as it's required,
3087 : * because it's not possible for any future tuples to pass. On
3088 : * a backward scan, however, we must keep going, because we
3089 : * may have initially positioned to the end of the index.
3090 : * (_bt_advance_array_keys also relies on this behavior during
3091 : * backward scans.)
3092 : */
3093 48 : if ((subkey->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)) &&
3094 : ScanDirectionIsForward(dir))
3095 0 : *continuescan = false;
3096 : }
3097 :
3098 : /*
3099 : * In any case, this indextuple doesn't match the qual.
3100 : */
3101 60 : return false;
3102 : }
3103 :
3104 2634 : if (subkey->sk_flags & SK_ISNULL)
3105 : {
3106 : /*
3107 : * Unlike the simple-scankey case, this isn't a disallowed case
3108 : * (except when it's the first row element that has the NULL arg).
3109 : * But it can never match. If all the earlier row comparison
3110 : * columns are required for the scan direction, we can stop the
3111 : * scan, because there can't be another tuple that will succeed.
3112 : */
3113 : Assert(subkey != (ScanKey) DatumGetPointer(skey->sk_argument));
3114 12 : subkey--;
3115 12 : if (forcenonrequired)
3116 : {
3117 : /* treating scan's keys as non-required */
3118 : }
3119 12 : else if ((subkey->sk_flags & SK_BT_REQFWD) &&
3120 : ScanDirectionIsForward(dir))
3121 6 : *continuescan = false;
3122 6 : else if ((subkey->sk_flags & SK_BT_REQBKWD) &&
3123 : ScanDirectionIsBackward(dir))
3124 6 : *continuescan = false;
3125 12 : return false;
3126 : }
3127 :
3128 : /* Perform the test --- three-way comparison not bool operator */
3129 2622 : cmpresult = DatumGetInt32(FunctionCall2Coll(&subkey->sk_func,
3130 : subkey->sk_collation,
3131 : datum,
3132 : subkey->sk_argument));
3133 :
3134 2622 : if (subkey->sk_flags & SK_BT_DESC)
3135 0 : INVERT_COMPARE_RESULT(cmpresult);
3136 :
3137 : /* Done comparing if unequal, else advance to next column */
3138 2622 : if (cmpresult != 0)
3139 2382 : break;
3140 :
3141 240 : if (subkey->sk_flags & SK_ROW_END)
3142 0 : break;
3143 240 : subkey++;
3144 : }
3145 :
3146 : /*
3147 : * At this point cmpresult indicates the overall result of the row
3148 : * comparison, and subkey points to the deciding column (or the last
3149 : * column if the result is "=").
3150 : */
3151 2388 : switch (subkey->sk_strategy)
3152 : {
3153 : /* EQ and NE cases aren't allowed here */
3154 186 : case BTLessStrategyNumber:
3155 186 : result = (cmpresult < 0);
3156 186 : break;
3157 1590 : case BTLessEqualStrategyNumber:
3158 1590 : result = (cmpresult <= 0);
3159 1590 : break;
3160 240 : case BTGreaterEqualStrategyNumber:
3161 240 : result = (cmpresult >= 0);
3162 240 : break;
3163 372 : case BTGreaterStrategyNumber:
3164 372 : result = (cmpresult > 0);
3165 372 : break;
3166 0 : default:
3167 0 : elog(ERROR, "unexpected strategy number %d", subkey->sk_strategy);
3168 : result = 0; /* keep compiler quiet */
3169 : break;
3170 : }
3171 :
3172 2388 : if (!result && !forcenonrequired)
3173 : {
3174 : /*
3175 : * Tuple fails this qual. If it's a required qual for the current
3176 : * scan direction, then we can conclude no further tuples will pass,
3177 : * either. Note we have to look at the deciding column, not
3178 : * necessarily the first or last column of the row condition.
3179 : */
3180 6 : if ((subkey->sk_flags & SK_BT_REQFWD) &&
3181 : ScanDirectionIsForward(dir))
3182 6 : *continuescan = false;
3183 0 : else if ((subkey->sk_flags & SK_BT_REQBKWD) &&
3184 : ScanDirectionIsBackward(dir))
3185 0 : *continuescan = false;
3186 : }
3187 :
3188 2388 : return result;
3189 : }
3190 :
3191 : /*
3192 : * Determine if a scan with array keys should skip over uninteresting tuples.
3193 : *
3194 : * This is a subroutine for _bt_checkkeys. Called when _bt_readpage's linear
3195 : * search process (started after it finishes reading an initial group of
3196 : * matching tuples, used to locate the start of the next group of tuples
3197 : * matching the next set of required array keys) has already scanned an
3198 : * excessive number of tuples whose key space is "between arrays".
3199 : *
3200 : * When we perform look ahead successfully, we'll sets pstate.skip, which
3201 : * instructs _bt_readpage to skip ahead to that tuple next (could be past the
3202 : * end of the scan's leaf page). Pages where the optimization is effective
3203 : * will generally still need to skip several times. Each call here performs
3204 : * only a single "look ahead" comparison of a later tuple, whose distance from
3205 : * the current tuple's offset number is determined by applying heuristics.
3206 : */
3207 : static void
3208 15404 : _bt_checkkeys_look_ahead(IndexScanDesc scan, BTReadPageState *pstate,
3209 : int tupnatts, TupleDesc tupdesc)
3210 : {
3211 15404 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
3212 15404 : ScanDirection dir = so->currPos.dir;
3213 : OffsetNumber aheadoffnum;
3214 : IndexTuple ahead;
3215 :
3216 : Assert(!pstate->forcenonrequired);
3217 :
3218 : /* Avoid looking ahead when comparing the page high key */
3219 15404 : if (pstate->offnum < pstate->minoff)
3220 0 : return;
3221 :
3222 : /*
3223 : * Don't look ahead when there aren't enough tuples remaining on the page
3224 : * (in the current scan direction) for it to be worth our while
3225 : */
3226 15404 : if (ScanDirectionIsForward(dir) &&
3227 15326 : pstate->offnum >= pstate->maxoff - LOOK_AHEAD_DEFAULT_DISTANCE)
3228 800 : return;
3229 14604 : else if (ScanDirectionIsBackward(dir) &&
3230 78 : pstate->offnum <= pstate->minoff + LOOK_AHEAD_DEFAULT_DISTANCE)
3231 24 : return;
3232 :
3233 : /*
3234 : * The look ahead distance starts small, and ramps up as each call here
3235 : * allows _bt_readpage to skip over more tuples
3236 : */
3237 14580 : if (!pstate->targetdistance)
3238 7978 : pstate->targetdistance = LOOK_AHEAD_DEFAULT_DISTANCE;
3239 6602 : else if (pstate->targetdistance < MaxIndexTuplesPerPage / 2)
3240 6602 : pstate->targetdistance *= 2;
3241 :
3242 : /* Don't read past the end (or before the start) of the page, though */
3243 14580 : if (ScanDirectionIsForward(dir))
3244 14526 : aheadoffnum = Min((int) pstate->maxoff,
3245 : (int) pstate->offnum + pstate->targetdistance);
3246 : else
3247 54 : aheadoffnum = Max((int) pstate->minoff,
3248 : (int) pstate->offnum - pstate->targetdistance);
3249 :
3250 14580 : ahead = (IndexTuple) PageGetItem(pstate->page,
3251 14580 : PageGetItemId(pstate->page, aheadoffnum));
3252 14580 : if (_bt_tuple_before_array_skeys(scan, dir, ahead, tupdesc, tupnatts,
3253 : false, 0, NULL))
3254 : {
3255 : /*
3256 : * Success -- instruct _bt_readpage to skip ahead to very next tuple
3257 : * after the one we determined was still before the current array keys
3258 : */
3259 5178 : if (ScanDirectionIsForward(dir))
3260 5142 : pstate->skip = aheadoffnum + 1;
3261 : else
3262 36 : pstate->skip = aheadoffnum - 1;
3263 : }
3264 : else
3265 : {
3266 : /*
3267 : * Failure -- "ahead" tuple is too far ahead (we were too aggressive).
3268 : *
3269 : * Reset the number of rechecks, and aggressively reduce the target
3270 : * distance (we're much more aggressive here than we were when the
3271 : * distance was initially ramped up).
3272 : */
3273 9402 : pstate->rechecks = 0;
3274 9402 : pstate->targetdistance = Max(pstate->targetdistance / 8, 1);
3275 : }
3276 : }
3277 :
3278 : /*
3279 : * _bt_killitems - set LP_DEAD state for items an indexscan caller has
3280 : * told us were killed
3281 : *
3282 : * scan->opaque, referenced locally through so, contains information about the
3283 : * current page and killed tuples thereon (generally, this should only be
3284 : * called if so->numKilled > 0).
3285 : *
3286 : * The caller does not have a lock on the page and may or may not have the
3287 : * page pinned in a buffer. Note that read-lock is sufficient for setting
3288 : * LP_DEAD status (which is only a hint).
3289 : *
3290 : * We match items by heap TID before assuming they are the right ones to
3291 : * delete. We cope with cases where items have moved right due to insertions.
3292 : * If an item has moved off the current page due to a split, we'll fail to
3293 : * find it and do nothing (this is not an error case --- we assume the item
3294 : * will eventually get marked in a future indexscan).
3295 : *
3296 : * Note that if we hold a pin on the target page continuously from initially
3297 : * reading the items until applying this function, VACUUM cannot have deleted
3298 : * any items from the page, and so there is no need to search left from the
3299 : * recorded offset. (This observation also guarantees that the item is still
3300 : * the right one to delete, which might otherwise be questionable since heap
3301 : * TIDs can get recycled.) This holds true even if the page has been modified
3302 : * by inserts and page splits, so there is no need to consult the LSN.
3303 : *
3304 : * If the pin was released after reading the page, then we re-read it. If it
3305 : * has been modified since we read it (as determined by the LSN), we dare not
3306 : * flag any entries because it is possible that the old entry was vacuumed
3307 : * away and the TID was re-used by a completely different heap tuple.
3308 : */
3309 : void
3310 173194 : _bt_killitems(IndexScanDesc scan)
3311 : {
3312 173194 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
3313 : Page page;
3314 : BTPageOpaque opaque;
3315 : OffsetNumber minoff;
3316 : OffsetNumber maxoff;
3317 : int i;
3318 173194 : int numKilled = so->numKilled;
3319 173194 : bool killedsomething = false;
3320 : bool droppedpin PG_USED_FOR_ASSERTS_ONLY;
3321 :
3322 : Assert(BTScanPosIsValid(so->currPos));
3323 :
3324 : /*
3325 : * Always reset the scan state, so we don't look for same items on other
3326 : * pages.
3327 : */
3328 173194 : so->numKilled = 0;
3329 :
3330 173194 : if (BTScanPosIsPinned(so->currPos))
3331 : {
3332 : /*
3333 : * We have held the pin on this page since we read the index tuples,
3334 : * so all we need to do is lock it. The pin will have prevented
3335 : * re-use of any TID on the page, so there is no need to check the
3336 : * LSN.
3337 : */
3338 34688 : droppedpin = false;
3339 34688 : _bt_lockbuf(scan->indexRelation, so->currPos.buf, BT_READ);
3340 :
3341 34688 : page = BufferGetPage(so->currPos.buf);
3342 : }
3343 : else
3344 : {
3345 : Buffer buf;
3346 :
3347 138506 : droppedpin = true;
3348 : /* Attempt to re-read the buffer, getting pin and lock. */
3349 138506 : buf = _bt_getbuf(scan->indexRelation, so->currPos.currPage, BT_READ);
3350 :
3351 138506 : page = BufferGetPage(buf);
3352 138506 : if (BufferGetLSNAtomic(buf) == so->currPos.lsn)
3353 138390 : so->currPos.buf = buf;
3354 : else
3355 : {
3356 : /* Modified while not pinned means hinting is not safe. */
3357 116 : _bt_relbuf(scan->indexRelation, buf);
3358 116 : return;
3359 : }
3360 : }
3361 :
3362 173078 : opaque = BTPageGetOpaque(page);
3363 173078 : minoff = P_FIRSTDATAKEY(opaque);
3364 173078 : maxoff = PageGetMaxOffsetNumber(page);
3365 :
3366 661288 : for (i = 0; i < numKilled; i++)
3367 : {
3368 488210 : int itemIndex = so->killedItems[i];
3369 488210 : BTScanPosItem *kitem = &so->currPos.items[itemIndex];
3370 488210 : OffsetNumber offnum = kitem->indexOffset;
3371 :
3372 : Assert(itemIndex >= so->currPos.firstItem &&
3373 : itemIndex <= so->currPos.lastItem);
3374 488210 : if (offnum < minoff)
3375 0 : continue; /* pure paranoia */
3376 9411070 : while (offnum <= maxoff)
3377 : {
3378 9339502 : ItemId iid = PageGetItemId(page, offnum);
3379 9339502 : IndexTuple ituple = (IndexTuple) PageGetItem(page, iid);
3380 9339502 : bool killtuple = false;
3381 :
3382 9339502 : if (BTreeTupleIsPosting(ituple))
3383 : {
3384 2921756 : int pi = i + 1;
3385 2921756 : int nposting = BTreeTupleGetNPosting(ituple);
3386 : int j;
3387 :
3388 : /*
3389 : * We rely on the convention that heap TIDs in the scanpos
3390 : * items array are stored in ascending heap TID order for a
3391 : * group of TIDs that originally came from a posting list
3392 : * tuple. This convention even applies during backwards
3393 : * scans, where returning the TIDs in descending order might
3394 : * seem more natural. This is about effectiveness, not
3395 : * correctness.
3396 : *
3397 : * Note that the page may have been modified in almost any way
3398 : * since we first read it (in the !droppedpin case), so it's
3399 : * possible that this posting list tuple wasn't a posting list
3400 : * tuple when we first encountered its heap TIDs.
3401 : */
3402 2995170 : for (j = 0; j < nposting; j++)
3403 : {
3404 2992478 : ItemPointer item = BTreeTupleGetPostingN(ituple, j);
3405 :
3406 2992478 : if (!ItemPointerEquals(item, &kitem->heapTid))
3407 2919064 : break; /* out of posting list loop */
3408 :
3409 : /*
3410 : * kitem must have matching offnum when heap TIDs match,
3411 : * though only in the common case where the page can't
3412 : * have been concurrently modified
3413 : */
3414 : Assert(kitem->indexOffset == offnum || !droppedpin);
3415 :
3416 : /*
3417 : * Read-ahead to later kitems here.
3418 : *
3419 : * We rely on the assumption that not advancing kitem here
3420 : * will prevent us from considering the posting list tuple
3421 : * fully dead by not matching its next heap TID in next
3422 : * loop iteration.
3423 : *
3424 : * If, on the other hand, this is the final heap TID in
3425 : * the posting list tuple, then tuple gets killed
3426 : * regardless (i.e. we handle the case where the last
3427 : * kitem is also the last heap TID in the last index tuple
3428 : * correctly -- posting tuple still gets killed).
3429 : */
3430 73414 : if (pi < numKilled)
3431 37234 : kitem = &so->currPos.items[so->killedItems[pi++]];
3432 : }
3433 :
3434 : /*
3435 : * Don't bother advancing the outermost loop's int iterator to
3436 : * avoid processing killed items that relate to the same
3437 : * offnum/posting list tuple. This micro-optimization hardly
3438 : * seems worth it. (Further iterations of the outermost loop
3439 : * will fail to match on this same posting list's first heap
3440 : * TID instead, so we'll advance to the next offnum/index
3441 : * tuple pretty quickly.)
3442 : */
3443 2921756 : if (j == nposting)
3444 2692 : killtuple = true;
3445 : }
3446 6417746 : else if (ItemPointerEquals(&ituple->t_tid, &kitem->heapTid))
3447 415270 : killtuple = true;
3448 :
3449 : /*
3450 : * Mark index item as dead, if it isn't already. Since this
3451 : * happens while holding a buffer lock possibly in shared mode,
3452 : * it's possible that multiple processes attempt to do this
3453 : * simultaneously, leading to multiple full-page images being sent
3454 : * to WAL (if wal_log_hints or data checksums are enabled), which
3455 : * is undesirable.
3456 : */
3457 9339502 : if (killtuple && !ItemIdIsDead(iid))
3458 : {
3459 : /* found the item/all posting list items */
3460 416642 : ItemIdMarkDead(iid);
3461 416642 : killedsomething = true;
3462 416642 : break; /* out of inner search loop */
3463 : }
3464 8922860 : offnum = OffsetNumberNext(offnum);
3465 : }
3466 : }
3467 :
3468 : /*
3469 : * Since this can be redone later if needed, mark as dirty hint.
3470 : *
3471 : * Whenever we mark anything LP_DEAD, we also set the page's
3472 : * BTP_HAS_GARBAGE flag, which is likewise just a hint. (Note that we
3473 : * only rely on the page-level flag in !heapkeyspace indexes.)
3474 : */
3475 173078 : if (killedsomething)
3476 : {
3477 135592 : opaque->btpo_flags |= BTP_HAS_GARBAGE;
3478 135592 : MarkBufferDirtyHint(so->currPos.buf, true);
3479 : }
3480 :
3481 173078 : _bt_unlockbuf(scan->indexRelation, so->currPos.buf);
3482 : }
3483 :
3484 :
3485 : /*
3486 : * The following routines manage a shared-memory area in which we track
3487 : * assignment of "vacuum cycle IDs" to currently-active btree vacuuming
3488 : * operations. There is a single counter which increments each time we
3489 : * start a vacuum to assign it a cycle ID. Since multiple vacuums could
3490 : * be active concurrently, we have to track the cycle ID for each active
3491 : * vacuum; this requires at most MaxBackends entries (usually far fewer).
3492 : * We assume at most one vacuum can be active for a given index.
3493 : *
3494 : * Access to the shared memory area is controlled by BtreeVacuumLock.
3495 : * In principle we could use a separate lmgr locktag for each index,
3496 : * but a single LWLock is much cheaper, and given the short time that
3497 : * the lock is ever held, the concurrency hit should be minimal.
3498 : */
3499 :
3500 : typedef struct BTOneVacInfo
3501 : {
3502 : LockRelId relid; /* global identifier of an index */
3503 : BTCycleId cycleid; /* cycle ID for its active VACUUM */
3504 : } BTOneVacInfo;
3505 :
3506 : typedef struct BTVacInfo
3507 : {
3508 : BTCycleId cycle_ctr; /* cycle ID most recently assigned */
3509 : int num_vacuums; /* number of currently active VACUUMs */
3510 : int max_vacuums; /* allocated length of vacuums[] array */
3511 : BTOneVacInfo vacuums[FLEXIBLE_ARRAY_MEMBER];
3512 : } BTVacInfo;
3513 :
3514 : static BTVacInfo *btvacinfo;
3515 :
3516 :
3517 : /*
3518 : * _bt_vacuum_cycleid --- get the active vacuum cycle ID for an index,
3519 : * or zero if there is no active VACUUM
3520 : *
3521 : * Note: for correct interlocking, the caller must already hold pin and
3522 : * exclusive lock on each buffer it will store the cycle ID into. This
3523 : * ensures that even if a VACUUM starts immediately afterwards, it cannot
3524 : * process those pages until the page split is complete.
3525 : */
3526 : BTCycleId
3527 22522 : _bt_vacuum_cycleid(Relation rel)
3528 : {
3529 22522 : BTCycleId result = 0;
3530 : int i;
3531 :
3532 : /* Share lock is enough since this is a read-only operation */
3533 22522 : LWLockAcquire(BtreeVacuumLock, LW_SHARED);
3534 :
3535 22522 : for (i = 0; i < btvacinfo->num_vacuums; i++)
3536 : {
3537 0 : BTOneVacInfo *vac = &btvacinfo->vacuums[i];
3538 :
3539 0 : if (vac->relid.relId == rel->rd_lockInfo.lockRelId.relId &&
3540 0 : vac->relid.dbId == rel->rd_lockInfo.lockRelId.dbId)
3541 : {
3542 0 : result = vac->cycleid;
3543 0 : break;
3544 : }
3545 : }
3546 :
3547 22522 : LWLockRelease(BtreeVacuumLock);
3548 22522 : return result;
3549 : }
3550 :
3551 : /*
3552 : * _bt_start_vacuum --- assign a cycle ID to a just-starting VACUUM operation
3553 : *
3554 : * Note: the caller must guarantee that it will eventually call
3555 : * _bt_end_vacuum, else we'll permanently leak an array slot. To ensure
3556 : * that this happens even in elog(FATAL) scenarios, the appropriate coding
3557 : * is not just a PG_TRY, but
3558 : * PG_ENSURE_ERROR_CLEANUP(_bt_end_vacuum_callback, PointerGetDatum(rel))
3559 : */
3560 : BTCycleId
3561 2774 : _bt_start_vacuum(Relation rel)
3562 : {
3563 : BTCycleId result;
3564 : int i;
3565 : BTOneVacInfo *vac;
3566 :
3567 2774 : LWLockAcquire(BtreeVacuumLock, LW_EXCLUSIVE);
3568 :
3569 : /*
3570 : * Assign the next cycle ID, being careful to avoid zero as well as the
3571 : * reserved high values.
3572 : */
3573 2774 : result = ++(btvacinfo->cycle_ctr);
3574 2774 : if (result == 0 || result > MAX_BT_CYCLE_ID)
3575 0 : result = btvacinfo->cycle_ctr = 1;
3576 :
3577 : /* Let's just make sure there's no entry already for this index */
3578 2774 : for (i = 0; i < btvacinfo->num_vacuums; i++)
3579 : {
3580 0 : vac = &btvacinfo->vacuums[i];
3581 0 : if (vac->relid.relId == rel->rd_lockInfo.lockRelId.relId &&
3582 0 : vac->relid.dbId == rel->rd_lockInfo.lockRelId.dbId)
3583 : {
3584 : /*
3585 : * Unlike most places in the backend, we have to explicitly
3586 : * release our LWLock before throwing an error. This is because
3587 : * we expect _bt_end_vacuum() to be called before transaction
3588 : * abort cleanup can run to release LWLocks.
3589 : */
3590 0 : LWLockRelease(BtreeVacuumLock);
3591 0 : elog(ERROR, "multiple active vacuums for index \"%s\"",
3592 : RelationGetRelationName(rel));
3593 : }
3594 : }
3595 :
3596 : /* OK, add an entry */
3597 2774 : if (btvacinfo->num_vacuums >= btvacinfo->max_vacuums)
3598 : {
3599 0 : LWLockRelease(BtreeVacuumLock);
3600 0 : elog(ERROR, "out of btvacinfo slots");
3601 : }
3602 2774 : vac = &btvacinfo->vacuums[btvacinfo->num_vacuums];
3603 2774 : vac->relid = rel->rd_lockInfo.lockRelId;
3604 2774 : vac->cycleid = result;
3605 2774 : btvacinfo->num_vacuums++;
3606 :
3607 2774 : LWLockRelease(BtreeVacuumLock);
3608 2774 : return result;
3609 : }
3610 :
3611 : /*
3612 : * _bt_end_vacuum --- mark a btree VACUUM operation as done
3613 : *
3614 : * Note: this is deliberately coded not to complain if no entry is found;
3615 : * this allows the caller to put PG_TRY around the start_vacuum operation.
3616 : */
3617 : void
3618 2774 : _bt_end_vacuum(Relation rel)
3619 : {
3620 : int i;
3621 :
3622 2774 : LWLockAcquire(BtreeVacuumLock, LW_EXCLUSIVE);
3623 :
3624 : /* Find the array entry */
3625 2774 : for (i = 0; i < btvacinfo->num_vacuums; i++)
3626 : {
3627 2774 : BTOneVacInfo *vac = &btvacinfo->vacuums[i];
3628 :
3629 2774 : if (vac->relid.relId == rel->rd_lockInfo.lockRelId.relId &&
3630 2774 : vac->relid.dbId == rel->rd_lockInfo.lockRelId.dbId)
3631 : {
3632 : /* Remove it by shifting down the last entry */
3633 2774 : *vac = btvacinfo->vacuums[btvacinfo->num_vacuums - 1];
3634 2774 : btvacinfo->num_vacuums--;
3635 2774 : break;
3636 : }
3637 : }
3638 :
3639 2774 : LWLockRelease(BtreeVacuumLock);
3640 2774 : }
3641 :
3642 : /*
3643 : * _bt_end_vacuum wrapped as an on_shmem_exit callback function
3644 : */
3645 : void
3646 0 : _bt_end_vacuum_callback(int code, Datum arg)
3647 : {
3648 0 : _bt_end_vacuum((Relation) DatumGetPointer(arg));
3649 0 : }
3650 :
3651 : /*
3652 : * BTreeShmemSize --- report amount of shared memory space needed
3653 : */
3654 : Size
3655 5994 : BTreeShmemSize(void)
3656 : {
3657 : Size size;
3658 :
3659 5994 : size = offsetof(BTVacInfo, vacuums);
3660 5994 : size = add_size(size, mul_size(MaxBackends, sizeof(BTOneVacInfo)));
3661 5994 : return size;
3662 : }
3663 :
3664 : /*
3665 : * BTreeShmemInit --- initialize this module's shared memory
3666 : */
3667 : void
3668 2096 : BTreeShmemInit(void)
3669 : {
3670 : bool found;
3671 :
3672 2096 : btvacinfo = (BTVacInfo *) ShmemInitStruct("BTree Vacuum State",
3673 : BTreeShmemSize(),
3674 : &found);
3675 :
3676 2096 : if (!IsUnderPostmaster)
3677 : {
3678 : /* Initialize shared memory area */
3679 : Assert(!found);
3680 :
3681 : /*
3682 : * It doesn't really matter what the cycle counter starts at, but
3683 : * having it always start the same doesn't seem good. Seed with
3684 : * low-order bits of time() instead.
3685 : */
3686 2096 : btvacinfo->cycle_ctr = (BTCycleId) time(NULL);
3687 :
3688 2096 : btvacinfo->num_vacuums = 0;
3689 2096 : btvacinfo->max_vacuums = MaxBackends;
3690 : }
3691 : else
3692 : Assert(found);
3693 2096 : }
3694 :
3695 : bytea *
3696 362 : btoptions(Datum reloptions, bool validate)
3697 : {
3698 : static const relopt_parse_elt tab[] = {
3699 : {"fillfactor", RELOPT_TYPE_INT, offsetof(BTOptions, fillfactor)},
3700 : {"vacuum_cleanup_index_scale_factor", RELOPT_TYPE_REAL,
3701 : offsetof(BTOptions, vacuum_cleanup_index_scale_factor)},
3702 : {"deduplicate_items", RELOPT_TYPE_BOOL,
3703 : offsetof(BTOptions, deduplicate_items)}
3704 : };
3705 :
3706 362 : return (bytea *) build_reloptions(reloptions, validate,
3707 : RELOPT_KIND_BTREE,
3708 : sizeof(BTOptions),
3709 : tab, lengthof(tab));
3710 : }
3711 :
3712 : /*
3713 : * btproperty() -- Check boolean properties of indexes.
3714 : *
3715 : * This is optional, but handling AMPROP_RETURNABLE here saves opening the rel
3716 : * to call btcanreturn.
3717 : */
3718 : bool
3719 756 : btproperty(Oid index_oid, int attno,
3720 : IndexAMProperty prop, const char *propname,
3721 : bool *res, bool *isnull)
3722 : {
3723 756 : switch (prop)
3724 : {
3725 42 : case AMPROP_RETURNABLE:
3726 : /* answer only for columns, not AM or whole index */
3727 42 : if (attno == 0)
3728 12 : return false;
3729 : /* otherwise, btree can always return data */
3730 30 : *res = true;
3731 30 : return true;
3732 :
3733 714 : default:
3734 714 : return false; /* punt to generic code */
3735 : }
3736 : }
3737 :
3738 : /*
3739 : * btbuildphasename() -- Return name of index build phase.
3740 : */
3741 : char *
3742 0 : btbuildphasename(int64 phasenum)
3743 : {
3744 0 : switch (phasenum)
3745 : {
3746 0 : case PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE:
3747 0 : return "initializing";
3748 0 : case PROGRESS_BTREE_PHASE_INDEXBUILD_TABLESCAN:
3749 0 : return "scanning table";
3750 0 : case PROGRESS_BTREE_PHASE_PERFORMSORT_1:
3751 0 : return "sorting live tuples";
3752 0 : case PROGRESS_BTREE_PHASE_PERFORMSORT_2:
3753 0 : return "sorting dead tuples";
3754 0 : case PROGRESS_BTREE_PHASE_LEAF_LOAD:
3755 0 : return "loading tuples in tree";
3756 0 : default:
3757 0 : return NULL;
3758 : }
3759 : }
3760 :
3761 : /*
3762 : * _bt_truncate() -- create tuple without unneeded suffix attributes.
3763 : *
3764 : * Returns truncated pivot index tuple allocated in caller's memory context,
3765 : * with key attributes copied from caller's firstright argument. If rel is
3766 : * an INCLUDE index, non-key attributes will definitely be truncated away,
3767 : * since they're not part of the key space. More aggressive suffix
3768 : * truncation can take place when it's clear that the returned tuple does not
3769 : * need one or more suffix key attributes. We only need to keep firstright
3770 : * attributes up to and including the first non-lastleft-equal attribute.
3771 : * Caller's insertion scankey is used to compare the tuples; the scankey's
3772 : * argument values are not considered here.
3773 : *
3774 : * Note that returned tuple's t_tid offset will hold the number of attributes
3775 : * present, so the original item pointer offset is not represented. Caller
3776 : * should only change truncated tuple's downlink. Note also that truncated
3777 : * key attributes are treated as containing "minus infinity" values by
3778 : * _bt_compare().
3779 : *
3780 : * In the worst case (when a heap TID must be appended to distinguish lastleft
3781 : * from firstright), the size of the returned tuple is the size of firstright
3782 : * plus the size of an additional MAXALIGN()'d item pointer. This guarantee
3783 : * is important, since callers need to stay under the 1/3 of a page
3784 : * restriction on tuple size. If this routine is ever taught to truncate
3785 : * within an attribute/datum, it will need to avoid returning an enlarged
3786 : * tuple to caller when truncation + TOAST compression ends up enlarging the
3787 : * final datum.
3788 : */
3789 : IndexTuple
3790 63274 : _bt_truncate(Relation rel, IndexTuple lastleft, IndexTuple firstright,
3791 : BTScanInsert itup_key)
3792 : {
3793 63274 : TupleDesc itupdesc = RelationGetDescr(rel);
3794 63274 : int16 nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
3795 : int keepnatts;
3796 : IndexTuple pivot;
3797 : IndexTuple tidpivot;
3798 : ItemPointer pivotheaptid;
3799 : Size newsize;
3800 :
3801 : /*
3802 : * We should only ever truncate non-pivot tuples from leaf pages. It's
3803 : * never okay to truncate when splitting an internal page.
3804 : */
3805 : Assert(!BTreeTupleIsPivot(lastleft) && !BTreeTupleIsPivot(firstright));
3806 :
3807 : /* Determine how many attributes must be kept in truncated tuple */
3808 63274 : keepnatts = _bt_keep_natts(rel, lastleft, firstright, itup_key);
3809 :
3810 : #ifdef DEBUG_NO_TRUNCATE
3811 : /* Force truncation to be ineffective for testing purposes */
3812 : keepnatts = nkeyatts + 1;
3813 : #endif
3814 :
3815 63274 : pivot = index_truncate_tuple(itupdesc, firstright,
3816 : Min(keepnatts, nkeyatts));
3817 :
3818 63274 : if (BTreeTupleIsPosting(pivot))
3819 : {
3820 : /*
3821 : * index_truncate_tuple() just returns a straight copy of firstright
3822 : * when it has no attributes to truncate. When that happens, we may
3823 : * need to truncate away a posting list here instead.
3824 : */
3825 : Assert(keepnatts == nkeyatts || keepnatts == nkeyatts + 1);
3826 : Assert(IndexRelationGetNumberOfAttributes(rel) == nkeyatts);
3827 1290 : pivot->t_info &= ~INDEX_SIZE_MASK;
3828 1290 : pivot->t_info |= MAXALIGN(BTreeTupleGetPostingOffset(firstright));
3829 : }
3830 :
3831 : /*
3832 : * If there is a distinguishing key attribute within pivot tuple, we're
3833 : * done
3834 : */
3835 63274 : if (keepnatts <= nkeyatts)
3836 : {
3837 62118 : BTreeTupleSetNAtts(pivot, keepnatts, false);
3838 62118 : return pivot;
3839 : }
3840 :
3841 : /*
3842 : * We have to store a heap TID in the new pivot tuple, since no non-TID
3843 : * key attribute value in firstright distinguishes the right side of the
3844 : * split from the left side. nbtree conceptualizes this case as an
3845 : * inability to truncate away any key attributes, since heap TID is
3846 : * treated as just another key attribute (despite lacking a pg_attribute
3847 : * entry).
3848 : *
3849 : * Use enlarged space that holds a copy of pivot. We need the extra space
3850 : * to store a heap TID at the end (using the special pivot tuple
3851 : * representation). Note that the original pivot already has firstright's
3852 : * possible posting list/non-key attribute values removed at this point.
3853 : */
3854 1156 : newsize = MAXALIGN(IndexTupleSize(pivot)) + MAXALIGN(sizeof(ItemPointerData));
3855 1156 : tidpivot = palloc0(newsize);
3856 1156 : memcpy(tidpivot, pivot, MAXALIGN(IndexTupleSize(pivot)));
3857 : /* Cannot leak memory here */
3858 1156 : pfree(pivot);
3859 :
3860 : /*
3861 : * Store all of firstright's key attribute values plus a tiebreaker heap
3862 : * TID value in enlarged pivot tuple
3863 : */
3864 1156 : tidpivot->t_info &= ~INDEX_SIZE_MASK;
3865 1156 : tidpivot->t_info |= newsize;
3866 1156 : BTreeTupleSetNAtts(tidpivot, nkeyatts, true);
3867 1156 : pivotheaptid = BTreeTupleGetHeapTID(tidpivot);
3868 :
3869 : /*
3870 : * Lehman & Yao use lastleft as the leaf high key in all cases, but don't
3871 : * consider suffix truncation. It seems like a good idea to follow that
3872 : * example in cases where no truncation takes place -- use lastleft's heap
3873 : * TID. (This is also the closest value to negative infinity that's
3874 : * legally usable.)
3875 : */
3876 1156 : ItemPointerCopy(BTreeTupleGetMaxHeapTID(lastleft), pivotheaptid);
3877 :
3878 : /*
3879 : * We're done. Assert() that heap TID invariants hold before returning.
3880 : *
3881 : * Lehman and Yao require that the downlink to the right page, which is to
3882 : * be inserted into the parent page in the second phase of a page split be
3883 : * a strict lower bound on items on the right page, and a non-strict upper
3884 : * bound for items on the left page. Assert that heap TIDs follow these
3885 : * invariants, since a heap TID value is apparently needed as a
3886 : * tiebreaker.
3887 : */
3888 : #ifndef DEBUG_NO_TRUNCATE
3889 : Assert(ItemPointerCompare(BTreeTupleGetMaxHeapTID(lastleft),
3890 : BTreeTupleGetHeapTID(firstright)) < 0);
3891 : Assert(ItemPointerCompare(pivotheaptid,
3892 : BTreeTupleGetHeapTID(lastleft)) >= 0);
3893 : Assert(ItemPointerCompare(pivotheaptid,
3894 : BTreeTupleGetHeapTID(firstright)) < 0);
3895 : #else
3896 :
3897 : /*
3898 : * Those invariants aren't guaranteed to hold for lastleft + firstright
3899 : * heap TID attribute values when they're considered here only because
3900 : * DEBUG_NO_TRUNCATE is defined (a heap TID is probably not actually
3901 : * needed as a tiebreaker). DEBUG_NO_TRUNCATE must therefore use a heap
3902 : * TID value that always works as a strict lower bound for items to the
3903 : * right. In particular, it must avoid using firstright's leading key
3904 : * attribute values along with lastleft's heap TID value when lastleft's
3905 : * TID happens to be greater than firstright's TID.
3906 : */
3907 : ItemPointerCopy(BTreeTupleGetHeapTID(firstright), pivotheaptid);
3908 :
3909 : /*
3910 : * Pivot heap TID should never be fully equal to firstright. Note that
3911 : * the pivot heap TID will still end up equal to lastleft's heap TID when
3912 : * that's the only usable value.
3913 : */
3914 : ItemPointerSetOffsetNumber(pivotheaptid,
3915 : OffsetNumberPrev(ItemPointerGetOffsetNumber(pivotheaptid)));
3916 : Assert(ItemPointerCompare(pivotheaptid,
3917 : BTreeTupleGetHeapTID(firstright)) < 0);
3918 : #endif
3919 :
3920 1156 : return tidpivot;
3921 : }
3922 :
3923 : /*
3924 : * _bt_keep_natts - how many key attributes to keep when truncating.
3925 : *
3926 : * Caller provides two tuples that enclose a split point. Caller's insertion
3927 : * scankey is used to compare the tuples; the scankey's argument values are
3928 : * not considered here.
3929 : *
3930 : * This can return a number of attributes that is one greater than the
3931 : * number of key attributes for the index relation. This indicates that the
3932 : * caller must use a heap TID as a unique-ifier in new pivot tuple.
3933 : */
3934 : static int
3935 63274 : _bt_keep_natts(Relation rel, IndexTuple lastleft, IndexTuple firstright,
3936 : BTScanInsert itup_key)
3937 : {
3938 63274 : int nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
3939 63274 : TupleDesc itupdesc = RelationGetDescr(rel);
3940 : int keepnatts;
3941 : ScanKey scankey;
3942 :
3943 : /*
3944 : * _bt_compare() treats truncated key attributes as having the value minus
3945 : * infinity, which would break searches within !heapkeyspace indexes. We
3946 : * must still truncate away non-key attribute values, though.
3947 : */
3948 63274 : if (!itup_key->heapkeyspace)
3949 0 : return nkeyatts;
3950 :
3951 63274 : scankey = itup_key->scankeys;
3952 63274 : keepnatts = 1;
3953 76860 : for (int attnum = 1; attnum <= nkeyatts; attnum++, scankey++)
3954 : {
3955 : Datum datum1,
3956 : datum2;
3957 : bool isNull1,
3958 : isNull2;
3959 :
3960 75704 : datum1 = index_getattr(lastleft, attnum, itupdesc, &isNull1);
3961 75704 : datum2 = index_getattr(firstright, attnum, itupdesc, &isNull2);
3962 :
3963 75704 : if (isNull1 != isNull2)
3964 62118 : break;
3965 :
3966 151376 : if (!isNull1 &&
3967 75672 : DatumGetInt32(FunctionCall2Coll(&scankey->sk_func,
3968 : scankey->sk_collation,
3969 : datum1,
3970 : datum2)) != 0)
3971 62118 : break;
3972 :
3973 13586 : keepnatts++;
3974 : }
3975 :
3976 : /*
3977 : * Assert that _bt_keep_natts_fast() agrees with us in passing. This is
3978 : * expected in an allequalimage index.
3979 : */
3980 : Assert(!itup_key->allequalimage ||
3981 : keepnatts == _bt_keep_natts_fast(rel, lastleft, firstright));
3982 :
3983 63274 : return keepnatts;
3984 : }
3985 :
3986 : /*
3987 : * _bt_keep_natts_fast - fast bitwise variant of _bt_keep_natts.
3988 : *
3989 : * This is exported so that a candidate split point can have its effect on
3990 : * suffix truncation inexpensively evaluated ahead of time when finding a
3991 : * split location. A naive bitwise approach to datum comparisons is used to
3992 : * save cycles.
3993 : *
3994 : * The approach taken here usually provides the same answer as _bt_keep_natts
3995 : * will (for the same pair of tuples from a heapkeyspace index), since the
3996 : * majority of btree opclasses can never indicate that two datums are equal
3997 : * unless they're bitwise equal after detoasting. When an index only has
3998 : * "equal image" columns, routine is guaranteed to give the same result as
3999 : * _bt_keep_natts would.
4000 : *
4001 : * Callers can rely on the fact that attributes considered equal here are
4002 : * definitely also equal according to _bt_keep_natts, even when the index uses
4003 : * an opclass or collation that is not "allequalimage"/deduplication-safe.
4004 : * This weaker guarantee is good enough for nbtsplitloc.c caller, since false
4005 : * negatives generally only have the effect of making leaf page splits use a
4006 : * more balanced split point.
4007 : */
4008 : int
4009 13827282 : _bt_keep_natts_fast(Relation rel, IndexTuple lastleft, IndexTuple firstright)
4010 : {
4011 13827282 : TupleDesc itupdesc = RelationGetDescr(rel);
4012 13827282 : int keysz = IndexRelationGetNumberOfKeyAttributes(rel);
4013 : int keepnatts;
4014 :
4015 13827282 : keepnatts = 1;
4016 23141368 : for (int attnum = 1; attnum <= keysz; attnum++)
4017 : {
4018 : Datum datum1,
4019 : datum2;
4020 : bool isNull1,
4021 : isNull2;
4022 : CompactAttribute *att;
4023 :
4024 20698354 : datum1 = index_getattr(lastleft, attnum, itupdesc, &isNull1);
4025 20698354 : datum2 = index_getattr(firstright, attnum, itupdesc, &isNull2);
4026 20698354 : att = TupleDescCompactAttr(itupdesc, attnum - 1);
4027 :
4028 20698354 : if (isNull1 != isNull2)
4029 11384268 : break;
4030 :
4031 20698148 : if (!isNull1 &&
4032 20651072 : !datum_image_eq(datum1, datum2, att->attbyval, att->attlen))
4033 11384062 : break;
4034 :
4035 9314086 : keepnatts++;
4036 : }
4037 :
4038 13827282 : return keepnatts;
4039 : }
4040 :
4041 : /*
4042 : * _bt_check_natts() -- Verify tuple has expected number of attributes.
4043 : *
4044 : * Returns value indicating if the expected number of attributes were found
4045 : * for a particular offset on page. This can be used as a general purpose
4046 : * sanity check.
4047 : *
4048 : * Testing a tuple directly with BTreeTupleGetNAtts() should generally be
4049 : * preferred to calling here. That's usually more convenient, and is always
4050 : * more explicit. Call here instead when offnum's tuple may be a negative
4051 : * infinity tuple that uses the pre-v11 on-disk representation, or when a low
4052 : * context check is appropriate. This routine is as strict as possible about
4053 : * what is expected on each version of btree.
4054 : */
4055 : bool
4056 4053222 : _bt_check_natts(Relation rel, bool heapkeyspace, Page page, OffsetNumber offnum)
4057 : {
4058 4053222 : int16 natts = IndexRelationGetNumberOfAttributes(rel);
4059 4053222 : int16 nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
4060 4053222 : BTPageOpaque opaque = BTPageGetOpaque(page);
4061 : IndexTuple itup;
4062 : int tupnatts;
4063 :
4064 : /*
4065 : * We cannot reliably test a deleted or half-dead page, since they have
4066 : * dummy high keys
4067 : */
4068 4053222 : if (P_IGNORE(opaque))
4069 0 : return true;
4070 :
4071 : Assert(offnum >= FirstOffsetNumber &&
4072 : offnum <= PageGetMaxOffsetNumber(page));
4073 :
4074 4053222 : itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offnum));
4075 4053222 : tupnatts = BTreeTupleGetNAtts(itup, rel);
4076 :
4077 : /* !heapkeyspace indexes do not support deduplication */
4078 4053222 : if (!heapkeyspace && BTreeTupleIsPosting(itup))
4079 0 : return false;
4080 :
4081 : /* Posting list tuples should never have "pivot heap TID" bit set */
4082 4053222 : if (BTreeTupleIsPosting(itup) &&
4083 21906 : (ItemPointerGetOffsetNumberNoCheck(&itup->t_tid) &
4084 : BT_PIVOT_HEAP_TID_ATTR) != 0)
4085 0 : return false;
4086 :
4087 : /* INCLUDE indexes do not support deduplication */
4088 4053222 : if (natts != nkeyatts && BTreeTupleIsPosting(itup))
4089 0 : return false;
4090 :
4091 4053222 : if (P_ISLEAF(opaque))
4092 : {
4093 4038870 : if (offnum >= P_FIRSTDATAKEY(opaque))
4094 : {
4095 : /*
4096 : * Non-pivot tuple should never be explicitly marked as a pivot
4097 : * tuple
4098 : */
4099 4025638 : if (BTreeTupleIsPivot(itup))
4100 0 : return false;
4101 :
4102 : /*
4103 : * Leaf tuples that are not the page high key (non-pivot tuples)
4104 : * should never be truncated. (Note that tupnatts must have been
4105 : * inferred, even with a posting list tuple, because only pivot
4106 : * tuples store tupnatts directly.)
4107 : */
4108 4025638 : return tupnatts == natts;
4109 : }
4110 : else
4111 : {
4112 : /*
4113 : * Rightmost page doesn't contain a page high key, so tuple was
4114 : * checked above as ordinary leaf tuple
4115 : */
4116 : Assert(!P_RIGHTMOST(opaque));
4117 :
4118 : /*
4119 : * !heapkeyspace high key tuple contains only key attributes. Note
4120 : * that tupnatts will only have been explicitly represented in
4121 : * !heapkeyspace indexes that happen to have non-key attributes.
4122 : */
4123 13232 : if (!heapkeyspace)
4124 0 : return tupnatts == nkeyatts;
4125 :
4126 : /* Use generic heapkeyspace pivot tuple handling */
4127 : }
4128 : }
4129 : else /* !P_ISLEAF(opaque) */
4130 : {
4131 14352 : if (offnum == P_FIRSTDATAKEY(opaque))
4132 : {
4133 : /*
4134 : * The first tuple on any internal page (possibly the first after
4135 : * its high key) is its negative infinity tuple. Negative
4136 : * infinity tuples are always truncated to zero attributes. They
4137 : * are a particular kind of pivot tuple.
4138 : */
4139 1114 : if (heapkeyspace)
4140 1114 : return tupnatts == 0;
4141 :
4142 : /*
4143 : * The number of attributes won't be explicitly represented if the
4144 : * negative infinity tuple was generated during a page split that
4145 : * occurred with a version of Postgres before v11. There must be
4146 : * a problem when there is an explicit representation that is
4147 : * non-zero, or when there is no explicit representation and the
4148 : * tuple is evidently not a pre-pg_upgrade tuple.
4149 : *
4150 : * Prior to v11, downlinks always had P_HIKEY as their offset.
4151 : * Accept that as an alternative indication of a valid
4152 : * !heapkeyspace negative infinity tuple.
4153 : */
4154 0 : return tupnatts == 0 ||
4155 0 : ItemPointerGetOffsetNumber(&(itup->t_tid)) == P_HIKEY;
4156 : }
4157 : else
4158 : {
4159 : /*
4160 : * !heapkeyspace downlink tuple with separator key contains only
4161 : * key attributes. Note that tupnatts will only have been
4162 : * explicitly represented in !heapkeyspace indexes that happen to
4163 : * have non-key attributes.
4164 : */
4165 13238 : if (!heapkeyspace)
4166 0 : return tupnatts == nkeyatts;
4167 :
4168 : /* Use generic heapkeyspace pivot tuple handling */
4169 : }
4170 : }
4171 :
4172 : /* Handle heapkeyspace pivot tuples (excluding minus infinity items) */
4173 : Assert(heapkeyspace);
4174 :
4175 : /*
4176 : * Explicit representation of the number of attributes is mandatory with
4177 : * heapkeyspace index pivot tuples, regardless of whether or not there are
4178 : * non-key attributes.
4179 : */
4180 26470 : if (!BTreeTupleIsPivot(itup))
4181 0 : return false;
4182 :
4183 : /* Pivot tuple should not use posting list representation (redundant) */
4184 26470 : if (BTreeTupleIsPosting(itup))
4185 0 : return false;
4186 :
4187 : /*
4188 : * Heap TID is a tiebreaker key attribute, so it cannot be untruncated
4189 : * when any other key attribute is truncated
4190 : */
4191 26470 : if (BTreeTupleGetHeapTID(itup) != NULL && tupnatts != nkeyatts)
4192 0 : return false;
4193 :
4194 : /*
4195 : * Pivot tuple must have at least one untruncated key attribute (minus
4196 : * infinity pivot tuples are the only exception). Pivot tuples can never
4197 : * represent that there is a value present for a key attribute that
4198 : * exceeds pg_index.indnkeyatts for the index.
4199 : */
4200 26470 : return tupnatts > 0 && tupnatts <= nkeyatts;
4201 : }
4202 :
4203 : /*
4204 : *
4205 : * _bt_check_third_page() -- check whether tuple fits on a btree page at all.
4206 : *
4207 : * We actually need to be able to fit three items on every page, so restrict
4208 : * any one item to 1/3 the per-page available space. Note that itemsz should
4209 : * not include the ItemId overhead.
4210 : *
4211 : * It might be useful to apply TOAST methods rather than throw an error here.
4212 : * Using out of line storage would break assumptions made by suffix truncation
4213 : * and by contrib/amcheck, though.
4214 : */
4215 : void
4216 264 : _bt_check_third_page(Relation rel, Relation heap, bool needheaptidspace,
4217 : Page page, IndexTuple newtup)
4218 : {
4219 : Size itemsz;
4220 : BTPageOpaque opaque;
4221 :
4222 264 : itemsz = MAXALIGN(IndexTupleSize(newtup));
4223 :
4224 : /* Double check item size against limit */
4225 264 : if (itemsz <= BTMaxItemSize)
4226 0 : return;
4227 :
4228 : /*
4229 : * Tuple is probably too large to fit on page, but it's possible that the
4230 : * index uses version 2 or version 3, or that page is an internal page, in
4231 : * which case a slightly higher limit applies.
4232 : */
4233 264 : if (!needheaptidspace && itemsz <= BTMaxItemSizeNoHeapTid)
4234 264 : return;
4235 :
4236 : /*
4237 : * Internal page insertions cannot fail here, because that would mean that
4238 : * an earlier leaf level insertion that should have failed didn't
4239 : */
4240 0 : opaque = BTPageGetOpaque(page);
4241 0 : if (!P_ISLEAF(opaque))
4242 0 : elog(ERROR, "cannot insert oversized tuple of size %zu on internal page of index \"%s\"",
4243 : itemsz, RelationGetRelationName(rel));
4244 :
4245 0 : ereport(ERROR,
4246 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
4247 : errmsg("index row size %zu exceeds btree version %u maximum %zu for index \"%s\"",
4248 : itemsz,
4249 : needheaptidspace ? BTREE_VERSION : BTREE_NOVAC_VERSION,
4250 : needheaptidspace ? BTMaxItemSize : BTMaxItemSizeNoHeapTid,
4251 : RelationGetRelationName(rel)),
4252 : errdetail("Index row references tuple (%u,%u) in relation \"%s\".",
4253 : ItemPointerGetBlockNumber(BTreeTupleGetHeapTID(newtup)),
4254 : ItemPointerGetOffsetNumber(BTreeTupleGetHeapTID(newtup)),
4255 : RelationGetRelationName(heap)),
4256 : errhint("Values larger than 1/3 of a buffer page cannot be indexed.\n"
4257 : "Consider a function index of an MD5 hash of the value, "
4258 : "or use full text indexing."),
4259 : errtableconstraint(heap, RelationGetRelationName(rel))));
4260 : }
4261 :
4262 : /*
4263 : * Are all attributes in rel "equality is image equality" attributes?
4264 : *
4265 : * We use each attribute's BTEQUALIMAGE_PROC opclass procedure. If any
4266 : * opclass either lacks a BTEQUALIMAGE_PROC procedure or returns false, we
4267 : * return false; otherwise we return true.
4268 : *
4269 : * Returned boolean value is stored in index metapage during index builds.
4270 : * Deduplication can only be used when we return true.
4271 : */
4272 : bool
4273 59380 : _bt_allequalimage(Relation rel, bool debugmessage)
4274 : {
4275 59380 : bool allequalimage = true;
4276 :
4277 : /* INCLUDE indexes can never support deduplication */
4278 59380 : if (IndexRelationGetNumberOfAttributes(rel) !=
4279 59380 : IndexRelationGetNumberOfKeyAttributes(rel))
4280 292 : return false;
4281 :
4282 156112 : for (int i = 0; i < IndexRelationGetNumberOfKeyAttributes(rel); i++)
4283 : {
4284 97560 : Oid opfamily = rel->rd_opfamily[i];
4285 97560 : Oid opcintype = rel->rd_opcintype[i];
4286 97560 : Oid collation = rel->rd_indcollation[i];
4287 : Oid equalimageproc;
4288 :
4289 97560 : equalimageproc = get_opfamily_proc(opfamily, opcintype, opcintype,
4290 : BTEQUALIMAGE_PROC);
4291 :
4292 : /*
4293 : * If there is no BTEQUALIMAGE_PROC then deduplication is assumed to
4294 : * be unsafe. Otherwise, actually call proc and see what it says.
4295 : */
4296 97560 : if (!OidIsValid(equalimageproc) ||
4297 97068 : !DatumGetBool(OidFunctionCall1Coll(equalimageproc, collation,
4298 : ObjectIdGetDatum(opcintype))))
4299 : {
4300 536 : allequalimage = false;
4301 536 : break;
4302 : }
4303 : }
4304 :
4305 59088 : if (debugmessage)
4306 : {
4307 50992 : if (allequalimage)
4308 50456 : elog(DEBUG1, "index \"%s\" can safely use deduplication",
4309 : RelationGetRelationName(rel));
4310 : else
4311 536 : elog(DEBUG1, "index \"%s\" cannot use deduplication",
4312 : RelationGetRelationName(rel));
4313 : }
4314 :
4315 59088 : return allequalimage;
4316 : }
|
