Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * nbtutils.c
4 : : * Utility code for Postgres btree implementation.
5 : : *
6 : : * Portions Copyright (c) 1996-2024, 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 "access/relscan.h"
23 : : #include "commands/progress.h"
24 : : #include "lib/qunique.h"
25 : : #include "miscadmin.h"
26 : : #include "utils/array.h"
27 : : #include "utils/datum.h"
28 : : #include "utils/lsyscache.h"
29 : : #include "utils/memutils.h"
30 : : #include "utils/rel.h"
31 : :
32 : : #define LOOK_AHEAD_REQUIRED_RECHECKS 3
33 : : #define LOOK_AHEAD_DEFAULT_DISTANCE 5
34 : :
35 : : typedef struct BTSortArrayContext
36 : : {
37 : : FmgrInfo *sortproc;
38 : : Oid collation;
39 : : bool reverse;
40 : : } BTSortArrayContext;
41 : :
42 : : typedef struct BTScanKeyPreproc
43 : : {
44 : : ScanKey skey;
45 : : int ikey;
46 : : int arrayidx;
47 : : } BTScanKeyPreproc;
48 : :
49 : : static void _bt_setup_array_cmp(IndexScanDesc scan, ScanKey skey, Oid elemtype,
50 : : FmgrInfo *orderproc, FmgrInfo **sortprocp);
51 : : static Datum _bt_find_extreme_element(IndexScanDesc scan, ScanKey skey,
52 : : Oid elemtype, StrategyNumber strat,
53 : : Datum *elems, int nelems);
54 : : static int _bt_sort_array_elements(ScanKey skey, FmgrInfo *sortproc,
55 : : bool reverse, Datum *elems, int nelems);
56 : : static bool _bt_merge_arrays(IndexScanDesc scan, ScanKey skey,
57 : : FmgrInfo *sortproc, bool reverse,
58 : : Oid origelemtype, Oid nextelemtype,
59 : : Datum *elems_orig, int *nelems_orig,
60 : : Datum *elems_next, int nelems_next);
61 : : static bool _bt_compare_array_scankey_args(IndexScanDesc scan,
62 : : ScanKey arraysk, ScanKey skey,
63 : : FmgrInfo *orderproc, BTArrayKeyInfo *array,
64 : : bool *qual_ok);
65 : : static ScanKey _bt_preprocess_array_keys(IndexScanDesc scan);
66 : : static void _bt_preprocess_array_keys_final(IndexScanDesc scan, int *keyDataMap);
67 : : static int _bt_compare_array_elements(const void *a, const void *b, void *arg);
68 : : static inline int32 _bt_compare_array_skey(FmgrInfo *orderproc,
69 : : Datum tupdatum, bool tupnull,
70 : : Datum arrdatum, ScanKey cur);
71 : : static int _bt_binsrch_array_skey(FmgrInfo *orderproc,
72 : : bool cur_elem_trig, ScanDirection dir,
73 : : Datum tupdatum, bool tupnull,
74 : : BTArrayKeyInfo *array, ScanKey cur,
75 : : int32 *set_elem_result);
76 : : static bool _bt_advance_array_keys_increment(IndexScanDesc scan, ScanDirection dir);
77 : : static void _bt_rewind_nonrequired_arrays(IndexScanDesc scan, ScanDirection dir);
78 : : static bool _bt_tuple_before_array_skeys(IndexScanDesc scan, ScanDirection dir,
79 : : IndexTuple tuple, TupleDesc tupdesc, int tupnatts,
80 : : bool readpagetup, int sktrig, bool *scanBehind);
81 : : static bool _bt_advance_array_keys(IndexScanDesc scan, BTReadPageState *pstate,
82 : : IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
83 : : int sktrig, bool sktrig_required);
84 : : #ifdef USE_ASSERT_CHECKING
85 : : static bool _bt_verify_arrays_bt_first(IndexScanDesc scan, ScanDirection dir);
86 : : static bool _bt_verify_keys_with_arraykeys(IndexScanDesc scan);
87 : : #endif
88 : : static bool _bt_compare_scankey_args(IndexScanDesc scan, ScanKey op,
89 : : ScanKey leftarg, ScanKey rightarg,
90 : : BTArrayKeyInfo *array, FmgrInfo *orderproc,
91 : : bool *result);
92 : : static bool _bt_fix_scankey_strategy(ScanKey skey, int16 *indoption);
93 : : static void _bt_mark_scankey_required(ScanKey skey);
94 : : static bool _bt_check_compare(IndexScanDesc scan, ScanDirection dir,
95 : : IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
96 : : bool advancenonrequired, bool prechecked, bool firstmatch,
97 : : bool *continuescan, int *ikey);
98 : : static bool _bt_check_rowcompare(ScanKey skey,
99 : : IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
100 : : ScanDirection dir, bool *continuescan);
101 : : static void _bt_checkkeys_look_ahead(IndexScanDesc scan, BTReadPageState *pstate,
102 : : int tupnatts, TupleDesc tupdesc);
103 : : static int _bt_keep_natts(Relation rel, IndexTuple lastleft,
104 : : IndexTuple firstright, BTScanInsert itup_key);
105 : :
106 : :
107 : : /*
108 : : * _bt_mkscankey
109 : : * Build an insertion scan key that contains comparison data from itup
110 : : * as well as comparator routines appropriate to the key datatypes.
111 : : *
112 : : * The result is intended for use with _bt_compare() and _bt_truncate().
113 : : * Callers that don't need to fill out the insertion scankey arguments
114 : : * (e.g. they use an ad-hoc comparison routine, or only need a scankey
115 : : * for _bt_truncate()) can pass a NULL index tuple. The scankey will
116 : : * be initialized as if an "all truncated" pivot tuple was passed
117 : : * instead.
118 : : *
119 : : * Note that we may occasionally have to share lock the metapage to
120 : : * determine whether or not the keys in the index are expected to be
121 : : * unique (i.e. if this is a "heapkeyspace" index). We assume a
122 : : * heapkeyspace index when caller passes a NULL tuple, allowing index
123 : : * build callers to avoid accessing the non-existent metapage. We
124 : : * also assume that the index is _not_ allequalimage when a NULL tuple
125 : : * is passed; CREATE INDEX callers call _bt_allequalimage() to set the
126 : : * field themselves.
127 : : */
128 : : BTScanInsert
309 pg@bowt.ie 129 :CBC 5651737 : _bt_mkscankey(Relation rel, IndexTuple itup)
130 : : {
131 : : BTScanInsert key;
132 : : ScanKey skey;
133 : : TupleDesc itupdesc;
134 : : int indnkeyatts;
135 : : int16 *indoption;
136 : : int tupnatts;
137 : : int i;
138 : :
9357 bruce@momjian.us 139 : 5651737 : itupdesc = RelationGetDescr(rel);
2199 teodor@sigaev.ru 140 : 5651737 : indnkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
6305 tgl@sss.pgh.pa.us 141 : 5651737 : indoption = rel->rd_indoption;
1852 pg@bowt.ie 142 [ + + + + ]: 5651737 : tupnatts = itup ? BTreeTupleGetNAtts(itup, rel) : 0;
143 : :
144 [ - + ]: 5651737 : Assert(tupnatts <= IndexRelationGetNumberOfAttributes(rel));
145 : :
146 : : /*
147 : : * We'll execute search using scan key constructed on key columns.
148 : : * Truncated attributes and non-key attributes are omitted from the final
149 : : * scan key.
150 : : */
151 : 5651737 : key = palloc(offsetof(BTScanInsertData, scankeys) +
152 : 5651737 : sizeof(ScanKeyData) * indnkeyatts);
1509 153 [ + + ]: 5651737 : if (itup)
309 154 : 5587934 : _bt_metaversion(rel, &key->heapkeyspace, &key->allequalimage);
155 : : else
156 : : {
157 : : /* Utility statement callers can set these fields themselves */
1509 158 : 63803 : key->heapkeyspace = true;
159 : 63803 : key->allequalimage = false;
160 : : }
1789 tgl@sss.pgh.pa.us 161 : 5651737 : key->anynullkeys = false; /* initial assumption */
128 pg@bowt.ie 162 :GNC 5651737 : key->nextkey = false; /* usual case, required by btinsert */
163 : 5651737 : key->backward = false; /* usual case, required by btinsert */
1852 pg@bowt.ie 164 :CBC 5651737 : key->keysz = Min(indnkeyatts, tupnatts);
165 [ + + ]: 5651737 : key->scantid = key->heapkeyspace && itup ?
166 [ + - ]: 11303474 : BTreeTupleGetHeapTID(itup) : NULL;
167 : 5651737 : skey = key->scankeys;
2199 teodor@sigaev.ru 168 [ + + ]: 15067511 : for (i = 0; i < indnkeyatts; i++)
169 : : {
170 : : FmgrInfo *procinfo;
171 : : Datum arg;
172 : : bool null;
173 : : int flags;
174 : :
175 : : /*
176 : : * We can use the cached (default) support procs since no cross-type
177 : : * comparison can be needed.
178 : : */
8226 tgl@sss.pgh.pa.us 179 : 9415774 : procinfo = index_getprocinfo(rel, i + 1, BTORDER_PROC);
180 : :
181 : : /*
182 : : * Key arguments built from truncated attributes (or when caller
183 : : * provides no tuple) are defensively represented as NULL values. They
184 : : * should never be used.
185 : : */
1852 pg@bowt.ie 186 [ + + ]: 9415774 : if (i < tupnatts)
187 : 9300948 : arg = index_getattr(itup, i + 1, itupdesc, &null);
188 : : else
189 : : {
190 : 114826 : arg = (Datum) 0;
191 : 114826 : null = true;
192 : : }
193 : 9415774 : flags = (null ? SK_ISNULL : 0) | (indoption[i] << SK_BT_INDOPTION_SHIFT);
8226 tgl@sss.pgh.pa.us 194 : 9415774 : ScanKeyEntryInitializeWithInfo(&skey[i],
195 : : flags,
196 : 9415774 : (AttrNumber) (i + 1),
197 : : InvalidStrategy,
198 : : InvalidOid,
4751 199 : 9415774 : rel->rd_indcollation[i],
200 : : procinfo,
201 : : arg);
202 : : /* Record if any key attribute is NULL (or truncated) */
1818 pg@bowt.ie 203 [ + + ]: 9415774 : if (null)
204 : 125114 : key->anynullkeys = true;
205 : : }
206 : :
207 : : /*
208 : : * In NULLS NOT DISTINCT mode, we pretend that there are no null keys, so
209 : : * that full uniqueness check is done.
210 : : */
801 peter@eisentraut.org 211 [ + + ]: 5651737 : if (rel->rd_index->indnullsnotdistinct)
212 : 84 : key->anynullkeys = false;
213 : :
1852 pg@bowt.ie 214 : 5651737 : return key;
215 : : }
216 : :
217 : : /*
218 : : * free a retracement stack made by _bt_search.
219 : : */
220 : : void
10141 scrappy@hub.org 221 : 9537512 : _bt_freestack(BTStack stack)
222 : : {
223 : : BTStack ostack;
224 : :
7403 neilc@samurai.com 225 [ + + ]: 17547802 : while (stack != NULL)
226 : : {
9716 bruce@momjian.us 227 : 8010290 : ostack = stack;
228 : 8010290 : stack = stack->bts_parent;
229 : 8010290 : pfree(ostack);
230 : : }
10141 scrappy@hub.org 231 : 9537512 : }
232 : :
233 : :
234 : : /*
235 : : * _bt_preprocess_array_keys() -- Preprocess SK_SEARCHARRAY scan keys
236 : : *
237 : : * If there are any SK_SEARCHARRAY scan keys, deconstruct the array(s) and
238 : : * set up BTArrayKeyInfo info for each one that is an equality-type key.
239 : : * Returns modified scan keys as input for further, standard preprocessing.
240 : : *
241 : : * Currently we perform two kinds of preprocessing to deal with redundancies.
242 : : * For inequality array keys, it's sufficient to find the extreme element
243 : : * value and replace the whole array with that scalar value. This eliminates
244 : : * all but one array element as redundant. Similarly, we are capable of
245 : : * "merging together" multiple equality array keys (from two or more input
246 : : * scan keys) into a single output scan key containing only the intersecting
247 : : * array elements. This can eliminate many redundant array elements, as well
248 : : * as eliminating whole array scan keys as redundant. It can also allow us to
249 : : * detect contradictory quals.
250 : : *
251 : : * It is convenient for _bt_preprocess_keys caller to have to deal with no
252 : : * more than one equality strategy array scan key per index attribute. We'll
253 : : * always be able to set things up that way when complete opfamilies are used.
254 : : * Eliminated array scan keys can be recognized as those that have had their
255 : : * sk_strategy field set to InvalidStrategy here by us. Caller should avoid
256 : : * including these in the scan's so->keyData[] output array.
257 : : *
258 : : * We set the scan key references from the scan's BTArrayKeyInfo info array to
259 : : * offsets into the temp modified input array returned to caller. Scans that
260 : : * have array keys should call _bt_preprocess_array_keys_final when standard
261 : : * preprocessing steps are complete. This will convert the scan key offset
262 : : * references into references to the scan's so->keyData[] output scan keys.
263 : : *
264 : : * Note: the reason we need to return a temp scan key array, rather than just
265 : : * scribbling on scan->keyData, is that callers are permitted to call btrescan
266 : : * without supplying a new set of scankey data.
267 : : */
268 : : static ScanKey
4564 tgl@sss.pgh.pa.us 269 : 6415243 : _bt_preprocess_array_keys(IndexScanDesc scan)
270 : : {
271 : 6415243 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
8 pg@bowt.ie 272 :GNC 6415243 : Relation rel = scan->indexRelation;
4564 tgl@sss.pgh.pa.us 273 :CBC 6415243 : int numberOfKeys = scan->numberOfKeys;
8 pg@bowt.ie 274 :GNC 6415243 : int16 *indoption = rel->rd_indoption;
275 : : int numArrayKeys;
276 : 6415243 : int origarrayatt = InvalidAttrNumber,
277 : 6415243 : origarraykey = -1;
278 : 6415243 : Oid origelemtype = InvalidOid;
279 : : ScanKey cur;
280 : : MemoryContext oldContext;
281 : : ScanKey arrayKeyData; /* modified copy of scan->keyData */
282 : :
283 [ - + ]: 6415243 : Assert(numberOfKeys);
284 : :
285 : : /* Quick check to see if there are any array keys */
4564 tgl@sss.pgh.pa.us 286 :CBC 6415243 : numArrayKeys = 0;
8 pg@bowt.ie 287 [ + + ]:GNC 16773511 : for (int i = 0; i < numberOfKeys; i++)
288 : : {
4564 tgl@sss.pgh.pa.us 289 :CBC 10358268 : cur = &scan->keyData[i];
290 [ + + ]: 10358268 : if (cur->sk_flags & SK_SEARCHARRAY)
291 : : {
292 : 494 : numArrayKeys++;
293 [ - + ]: 494 : Assert(!(cur->sk_flags & (SK_ROW_HEADER | SK_SEARCHNULL | SK_SEARCHNOTNULL)));
294 : : /* If any arrays are null as a whole, we can quit right now. */
295 [ - + ]: 494 : if (cur->sk_flags & SK_ISNULL)
296 : : {
8 pg@bowt.ie 297 :UNC 0 : so->qual_ok = false;
298 : 0 : return NULL;
299 : : }
300 : : }
301 : : }
302 : :
303 : : /* Quit if nothing to do. */
4564 tgl@sss.pgh.pa.us 304 [ + + ]:CBC 6415243 : if (numArrayKeys == 0)
8 pg@bowt.ie 305 :GNC 6414824 : return NULL;
306 : :
307 : : /*
308 : : * Make a scan-lifespan context to hold array-associated data, or reset it
309 : : * if we already have one from a previous rescan cycle.
310 : : */
4564 tgl@sss.pgh.pa.us 311 [ + - ]:CBC 419 : if (so->arrayContext == NULL)
312 : 419 : so->arrayContext = AllocSetContextCreate(CurrentMemoryContext,
313 : : "BTree array context",
314 : : ALLOCSET_SMALL_SIZES);
315 : : else
4564 tgl@sss.pgh.pa.us 316 :UBC 0 : MemoryContextReset(so->arrayContext);
317 : :
4564 tgl@sss.pgh.pa.us 318 :CBC 419 : oldContext = MemoryContextSwitchTo(so->arrayContext);
319 : :
320 : : /* Create modifiable copy of scan->keyData in the workspace context */
8 pg@bowt.ie 321 :GNC 419 : arrayKeyData = (ScanKey) palloc(numberOfKeys * sizeof(ScanKeyData));
322 : 419 : memcpy(arrayKeyData, scan->keyData, numberOfKeys * sizeof(ScanKeyData));
323 : :
324 : : /* Allocate space for per-array data in the workspace context */
325 : 419 : so->arrayKeys = (BTArrayKeyInfo *) palloc(numArrayKeys * sizeof(BTArrayKeyInfo));
326 : :
327 : : /* Allocate space for ORDER procs used to help _bt_checkkeys */
328 : 419 : so->orderProcs = (FmgrInfo *) palloc(numberOfKeys * sizeof(FmgrInfo));
329 : :
330 : : /* Now process each array key */
4564 tgl@sss.pgh.pa.us 331 :CBC 419 : numArrayKeys = 0;
8 pg@bowt.ie 332 [ + + ]:GNC 1061 : for (int i = 0; i < numberOfKeys; i++)
333 : : {
334 : : FmgrInfo sortproc;
335 : 645 : FmgrInfo *sortprocp = &sortproc;
336 : : Oid elemtype;
337 : : bool reverse;
338 : : ArrayType *arrayval;
339 : : int16 elmlen;
340 : : bool elmbyval;
341 : : char elmalign;
342 : : int num_elems;
343 : : Datum *elem_values;
344 : : bool *elem_nulls;
345 : : int num_nonnulls;
346 : : int j;
347 : :
348 : 645 : cur = &arrayKeyData[i];
4564 tgl@sss.pgh.pa.us 349 [ + + ]:CBC 645 : if (!(cur->sk_flags & SK_SEARCHARRAY))
350 : 157 : continue;
351 : :
352 : : /*
353 : : * First, deconstruct the array into elements. Anything allocated
354 : : * here (including a possibly detoasted array value) is in the
355 : : * workspace context.
356 : : */
357 : 494 : arrayval = DatumGetArrayTypeP(cur->sk_argument);
358 : : /* We could cache this data, but not clear it's worth it */
359 : 494 : get_typlenbyvalalign(ARR_ELEMTYPE(arrayval),
360 : : &elmlen, &elmbyval, &elmalign);
361 : 494 : deconstruct_array(arrayval,
362 : : ARR_ELEMTYPE(arrayval),
363 : : elmlen, elmbyval, elmalign,
364 : : &elem_values, &elem_nulls, &num_elems);
365 : :
366 : : /*
367 : : * Compress out any null elements. We can ignore them since we assume
368 : : * all btree operators are strict.
369 : : */
370 : 494 : num_nonnulls = 0;
371 [ + + ]: 2673 : for (j = 0; j < num_elems; j++)
372 : : {
373 [ + - ]: 2179 : if (!elem_nulls[j])
374 : 2179 : elem_values[num_nonnulls++] = elem_values[j];
375 : : }
376 : :
377 : : /* We could pfree(elem_nulls) now, but not worth the cycles */
378 : :
379 : : /* If there's no non-nulls, the scan qual is unsatisfiable */
380 [ - + ]: 494 : if (num_nonnulls == 0)
381 : : {
8 pg@bowt.ie 382 :UNC 0 : so->qual_ok = false;
4564 tgl@sss.pgh.pa.us 383 :UBC 0 : break;
384 : : }
385 : :
386 : : /*
387 : : * Determine the nominal datatype of the array elements. We have to
388 : : * support the convention that sk_subtype == InvalidOid means the
389 : : * opclass input type; this is a hack to simplify life for
390 : : * ScanKeyInit().
391 : : */
8 pg@bowt.ie 392 :GNC 494 : elemtype = cur->sk_subtype;
393 [ - + ]: 494 : if (elemtype == InvalidOid)
8 pg@bowt.ie 394 :UNC 0 : elemtype = rel->rd_opcintype[cur->sk_attno - 1];
8 pg@bowt.ie 395 [ - + ]:GNC 494 : Assert(elemtype == ARR_ELEMTYPE(arrayval));
396 : :
397 : : /*
398 : : * If the comparison operator is not equality, then the array qual
399 : : * degenerates to a simple comparison against the smallest or largest
400 : : * non-null array element, as appropriate.
401 : : */
4564 tgl@sss.pgh.pa.us 402 [ - + + - ]:CBC 494 : switch (cur->sk_strategy)
403 : : {
4564 tgl@sss.pgh.pa.us 404 :UBC 0 : case BTLessStrategyNumber:
405 : : case BTLessEqualStrategyNumber:
406 : 0 : cur->sk_argument =
8 pg@bowt.ie 407 :UNC 0 : _bt_find_extreme_element(scan, cur, elemtype,
408 : : BTGreaterStrategyNumber,
409 : : elem_values, num_nonnulls);
4564 tgl@sss.pgh.pa.us 410 :UBC 0 : continue;
4564 tgl@sss.pgh.pa.us 411 :CBC 491 : case BTEqualStrategyNumber:
412 : : /* proceed with rest of loop */
413 : 491 : break;
414 : 3 : case BTGreaterEqualStrategyNumber:
415 : : case BTGreaterStrategyNumber:
416 : 3 : cur->sk_argument =
8 pg@bowt.ie 417 :GNC 3 : _bt_find_extreme_element(scan, cur, elemtype,
418 : : BTLessStrategyNumber,
419 : : elem_values, num_nonnulls);
4564 tgl@sss.pgh.pa.us 420 :CBC 3 : continue;
4564 tgl@sss.pgh.pa.us 421 :UBC 0 : default:
422 [ # # ]: 0 : elog(ERROR, "unrecognized StrategyNumber: %d",
423 : : (int) cur->sk_strategy);
424 : : break;
425 : : }
426 : :
427 : : /*
428 : : * We'll need a 3-way ORDER proc to perform binary searches for the
429 : : * next matching array element. Set that up now.
430 : : *
431 : : * Array scan keys with cross-type equality operators will require a
432 : : * separate same-type ORDER proc for sorting their array. Otherwise,
433 : : * sortproc just points to the same proc used during binary searches.
434 : : */
8 pg@bowt.ie 435 :GNC 491 : _bt_setup_array_cmp(scan, cur, elemtype,
436 : 491 : &so->orderProcs[i], &sortprocp);
437 : :
438 : : /*
439 : : * Sort the non-null elements and eliminate any duplicates. We must
440 : : * sort in the same ordering used by the index column, so that the
441 : : * arrays can be advanced in lockstep with the scan's progress through
442 : : * the index's key space.
443 : : */
444 : 491 : reverse = (indoption[cur->sk_attno - 1] & INDOPTION_DESC) != 0;
445 : 491 : num_elems = _bt_sort_array_elements(cur, sortprocp, reverse,
4564 tgl@sss.pgh.pa.us 446 :ECB (344) : elem_values, num_nonnulls);
447 : :
8 pg@bowt.ie 448 [ + + ]:GNC 491 : if (origarrayatt == cur->sk_attno)
449 : : {
450 : 6 : BTArrayKeyInfo *orig = &so->arrayKeys[origarraykey];
451 : :
452 : : /*
453 : : * This array scan key is redundant with a previous equality
454 : : * operator array scan key. Merge the two arrays together to
455 : : * eliminate contradictory non-intersecting elements (or try to).
456 : : *
457 : : * We merge this next array back into attribute's original array.
458 : : */
459 [ - + ]: 6 : Assert(arrayKeyData[orig->scan_key].sk_attno == cur->sk_attno);
460 [ - + ]: 6 : Assert(arrayKeyData[orig->scan_key].sk_collation ==
461 : : cur->sk_collation);
462 [ + - ]: 6 : if (_bt_merge_arrays(scan, cur, sortprocp, reverse,
463 : : origelemtype, elemtype,
464 : : orig->elem_values, &orig->num_elems,
465 : : elem_values, num_elems))
466 : : {
467 : : /* Successfully eliminated this array */
468 : 6 : pfree(elem_values);
469 : :
470 : : /*
471 : : * If no intersecting elements remain in the original array,
472 : : * the scan qual is unsatisfiable
473 : : */
474 [ + + ]: 6 : if (orig->num_elems == 0)
475 : : {
476 : 3 : so->qual_ok = false;
477 : 3 : break;
478 : : }
479 : :
480 : : /*
481 : : * Indicate to _bt_preprocess_keys caller that it must ignore
482 : : * this scan key
483 : : */
484 : 3 : cur->sk_strategy = InvalidStrategy;
485 : 3 : continue;
486 : : }
487 : :
488 : : /*
489 : : * Unable to merge this array with previous array due to a lack of
490 : : * suitable cross-type opfamily support. Will need to keep both
491 : : * scan keys/arrays.
492 : : */
493 : : }
494 : : else
495 : : {
496 : : /*
497 : : * This array is the first for current index attribute.
498 : : *
499 : : * If it turns out to not be the last array (that is, if the next
500 : : * array is redundantly applied to this same index attribute),
501 : : * we'll then treat this array as the attribute's "original" array
502 : : * when merging.
503 : : */
504 : 485 : origarrayatt = cur->sk_attno;
505 : 485 : origarraykey = numArrayKeys;
506 : 485 : origelemtype = elemtype;
507 : : }
508 : :
509 : : /*
510 : : * And set up the BTArrayKeyInfo data.
511 : : *
512 : : * Note: _bt_preprocess_array_keys_final will fix-up each array's
513 : : * scan_key field later on, after so->keyData[] has been finalized.
514 : : */
4564 tgl@sss.pgh.pa.us 515 :CBC 485 : so->arrayKeys[numArrayKeys].scan_key = i;
516 : 485 : so->arrayKeys[numArrayKeys].num_elems = num_elems;
517 : 485 : so->arrayKeys[numArrayKeys].elem_values = elem_values;
518 : 485 : numArrayKeys++;
519 : : }
520 : :
521 : 419 : so->numArrayKeys = numArrayKeys;
522 : :
523 : 419 : MemoryContextSwitchTo(oldContext);
524 : :
8 pg@bowt.ie 525 :GNC 419 : return arrayKeyData;
526 : : }
527 : :
528 : : /*
529 : : * _bt_preprocess_array_keys_final() -- fix up array scan key references
530 : : *
531 : : * When _bt_preprocess_array_keys performed initial array preprocessing, it
532 : : * set each array's array->scan_key to the array's arrayKeys[] entry offset
533 : : * (that also work as references into the original scan->keyData[] array).
534 : : * This function handles translation of the scan key references from the
535 : : * BTArrayKeyInfo info array, from input scan key references (to the keys in
536 : : * scan->keyData[]), into output references (to the keys in so->keyData[]).
537 : : * Caller's keyDataMap[] array tells us how to perform this remapping.
538 : : *
539 : : * Also finalizes so->orderProcs[] for the scan. Arrays already have an ORDER
540 : : * proc, which might need to be repositioned to its so->keyData[]-wise offset
541 : : * (very much like the remapping that we apply to array->scan_key references).
542 : : * Non-array equality strategy scan keys (that survived preprocessing) don't
543 : : * yet have an so->orderProcs[] entry, so we set one for them here.
544 : : *
545 : : * Also converts single-element array scan keys into equivalent non-array
546 : : * equality scan keys, which decrements so->numArrayKeys. It's possible that
547 : : * this will leave this new btrescan without any arrays at all. This isn't
548 : : * necessary for correctness; it's just an optimization. Non-array equality
549 : : * scan keys are slightly faster than equivalent array scan keys at runtime.
550 : : */
551 : : static void
552 : 196 : _bt_preprocess_array_keys_final(IndexScanDesc scan, int *keyDataMap)
553 : : {
554 : 196 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
555 : 196 : Relation rel = scan->indexRelation;
556 : 196 : int arrayidx = 0;
557 : 196 : int last_equal_output_ikey PG_USED_FOR_ASSERTS_ONLY = -1;
558 : :
559 [ - + ]: 196 : Assert(so->qual_ok);
560 : :
561 : : /*
562 : : * Nothing for us to do when _bt_preprocess_array_keys only had to deal
563 : : * with array inequalities
564 : : */
7 565 [ - + ]: 196 : if (so->numArrayKeys == 0)
7 pg@bowt.ie 566 :UNC 0 : return;
567 : :
8 pg@bowt.ie 568 [ + + ]:GNC 588 : for (int output_ikey = 0; output_ikey < so->numberOfKeys; output_ikey++)
569 : : {
570 : 401 : ScanKey outkey = so->keyData + output_ikey;
571 : : int input_ikey;
572 : 401 : bool found PG_USED_FOR_ASSERTS_ONLY = false;
573 : :
574 [ - + ]: 401 : Assert(outkey->sk_strategy != InvalidStrategy);
575 : :
576 [ + + ]: 401 : if (outkey->sk_strategy != BTEqualStrategyNumber)
577 : 9 : continue;
578 : :
579 : 392 : input_ikey = keyDataMap[output_ikey];
580 : :
581 [ - + ]: 392 : Assert(last_equal_output_ikey < output_ikey);
582 [ - + ]: 392 : Assert(last_equal_output_ikey < input_ikey);
583 : 392 : last_equal_output_ikey = output_ikey;
584 : :
585 : : /*
586 : : * We're lazy about looking up ORDER procs for non-array keys, since
587 : : * not all input keys become output keys. Take care of it now.
588 : : */
589 [ + + ]: 392 : if (!(outkey->sk_flags & SK_SEARCHARRAY))
590 : 75 : {
591 : : Oid elemtype;
592 : :
593 : : /* No need for an ORDER proc given an IS NULL scan key */
594 [ + + ]: 127 : if (outkey->sk_flags & SK_SEARCHNULL)
595 : 3 : continue;
596 : :
597 : : /*
598 : : * A non-required scan key doesn't need an ORDER proc, either
599 : : * (unless it's associated with an array, which this one isn't)
600 : : */
601 [ + + ]: 124 : if (!(outkey->sk_flags & SK_BT_REQFWD))
602 : 49 : continue;
603 : :
604 : 75 : elemtype = outkey->sk_subtype;
605 [ - + ]: 75 : if (elemtype == InvalidOid)
8 pg@bowt.ie 606 :UNC 0 : elemtype = rel->rd_opcintype[outkey->sk_attno - 1];
607 : :
8 pg@bowt.ie 608 :GNC 75 : _bt_setup_array_cmp(scan, outkey, elemtype,
609 : 75 : &so->orderProcs[output_ikey], NULL);
610 : 75 : continue;
611 : : }
612 : :
613 : : /*
614 : : * Reorder existing array scan key so->orderProcs[] entries.
615 : : *
616 : : * Doing this in-place is safe because preprocessing is required to
617 : : * output all equality strategy scan keys in original input order
618 : : * (among each group of entries against the same index attribute).
619 : : * This is also the order that the arrays themselves appear in.
620 : : */
621 : 265 : so->orderProcs[output_ikey] = so->orderProcs[input_ikey];
622 : :
623 : : /* Fix-up array->scan_key references for arrays */
624 [ + - ]: 265 : for (; arrayidx < so->numArrayKeys; arrayidx++)
625 : : {
626 : 265 : BTArrayKeyInfo *array = &so->arrayKeys[arrayidx];
627 : :
628 [ - + ]: 265 : Assert(array->num_elems > 0);
629 : :
630 [ + - ]: 265 : if (array->scan_key == input_ikey)
631 : : {
632 : : /* found it */
633 : 265 : array->scan_key = output_ikey;
634 : 265 : found = true;
635 : :
636 : : /*
637 : : * Transform array scan keys that have exactly 1 element
638 : : * remaining (following all prior preprocessing) into
639 : : * equivalent non-array scan keys.
640 : : */
641 [ + + ]: 265 : if (array->num_elems == 1)
642 : : {
643 : 12 : outkey->sk_flags &= ~SK_SEARCHARRAY;
644 : 12 : outkey->sk_argument = array->elem_values[0];
645 : 12 : so->numArrayKeys--;
646 : :
647 : : /* If we're out of array keys, we can quit right away */
648 [ + + ]: 12 : if (so->numArrayKeys == 0)
649 : 9 : return;
650 : :
651 : : /* Shift other arrays forward */
652 : 3 : memmove(array, array + 1,
653 : : sizeof(BTArrayKeyInfo) *
654 : 3 : (so->numArrayKeys - arrayidx));
655 : :
656 : : /*
657 : : * Don't increment arrayidx (there was an entry that was
658 : : * just shifted forward to the offset at arrayidx, which
659 : : * will still need to be matched)
660 : : */
661 : : }
662 : : else
663 : : {
664 : : /* Match found, so done with this array */
665 : 253 : arrayidx++;
666 : : }
667 : :
668 : 256 : break;
669 : : }
670 : : }
671 : :
672 [ - + ]: 256 : Assert(found);
673 : : }
674 : :
675 : : /*
676 : : * Parallel index scans require space in shared memory to store the
677 : : * current array elements (for arrays kept by preprocessing) to schedule
678 : : * the next primitive index scan. The underlying structure is protected
679 : : * using a spinlock, so defensively limit its size. In practice this can
680 : : * only affect parallel scans that use an incomplete opfamily.
681 : : */
682 [ - + - - ]: 187 : if (scan->parallel_scan && so->numArrayKeys > INDEX_MAX_KEYS)
8 pg@bowt.ie 683 [ # # ]:UNC 0 : ereport(ERROR,
684 : : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
685 : : errmsg_internal("number of array scan keys left by preprocessing (%d) exceeds the maximum allowed by parallel btree index scans (%d)",
686 : : so->numArrayKeys, INDEX_MAX_KEYS)));
687 : : }
688 : :
689 : : /*
690 : : * _bt_setup_array_cmp() -- Set up array comparison functions
691 : : *
692 : : * Sets ORDER proc in caller's orderproc argument, which is used during binary
693 : : * searches of arrays during the index scan. Also sets a same-type ORDER proc
694 : : * in caller's *sortprocp argument, which is used when sorting the array.
695 : : *
696 : : * Preprocessing calls here with all equality strategy scan keys (when scan
697 : : * uses equality array keys), including those not associated with any array.
698 : : * See _bt_advance_array_keys for an explanation of why it'll need to treat
699 : : * simple scalar equality scan keys as degenerate single element arrays.
700 : : *
701 : : * Caller should pass an orderproc pointing to space that'll store the ORDER
702 : : * proc for the scan, and a *sortprocp pointing to its own separate space.
703 : : * When calling here for a non-array scan key, sortprocp arg should be NULL.
704 : : *
705 : : * In the common case where we don't need to deal with cross-type operators,
706 : : * only one ORDER proc is actually required by caller. We'll set *sortprocp
707 : : * to point to the same memory that caller's orderproc continues to point to.
708 : : * Otherwise, *sortprocp will continue to point to caller's own space. Either
709 : : * way, *sortprocp will point to a same-type ORDER proc (since that's the only
710 : : * safe way to sort/deduplicate the array associated with caller's scan key).
711 : : */
712 : : static void
8 pg@bowt.ie 713 :GNC 566 : _bt_setup_array_cmp(IndexScanDesc scan, ScanKey skey, Oid elemtype,
714 : : FmgrInfo *orderproc, FmgrInfo **sortprocp)
715 : : {
716 : 566 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
717 : 566 : Relation rel = scan->indexRelation;
718 : : RegProcedure cmp_proc;
719 : 566 : Oid opcintype = rel->rd_opcintype[skey->sk_attno - 1];
720 : :
721 [ - + ]: 566 : Assert(skey->sk_strategy == BTEqualStrategyNumber);
722 [ - + ]: 566 : Assert(OidIsValid(elemtype));
723 : :
724 : : /*
725 : : * If scankey operator is not a cross-type comparison, we can use the
726 : : * cached comparison function; otherwise gotta look it up in the catalogs
727 : : */
728 [ + + ]: 566 : if (elemtype == opcintype)
729 : : {
730 : : /* Set same-type ORDER procs for caller */
731 : 554 : *orderproc = *index_getprocinfo(rel, skey->sk_attno, BTORDER_PROC);
732 [ + + ]: 554 : if (sortprocp)
733 : 488 : *sortprocp = orderproc;
734 : :
735 : 554 : return;
736 : : }
737 : :
738 : : /*
739 : : * Look up the appropriate cross-type comparison function in the opfamily.
740 : : *
741 : : * Use the opclass input type as the left hand arg type, and the array
742 : : * element type as the right hand arg type (since binary searches use an
743 : : * index tuple's attribute value to search for a matching array element).
744 : : *
745 : : * Note: it's possible that this would fail, if the opfamily is
746 : : * incomplete, but only in cases where it's quite likely that _bt_first
747 : : * would fail in just the same way (had we not failed before it could).
748 : : */
749 : 12 : cmp_proc = get_opfamily_proc(rel->rd_opfamily[skey->sk_attno - 1],
750 : : opcintype, elemtype, BTORDER_PROC);
751 [ - + ]: 12 : if (!RegProcedureIsValid(cmp_proc))
8 pg@bowt.ie 752 [ # # ]:UNC 0 : elog(ERROR, "missing support function %d(%u,%u) for attribute %d of index \"%s\"",
753 : : BTORDER_PROC, opcintype, elemtype, skey->sk_attno,
754 : : RelationGetRelationName(rel));
755 : :
756 : : /* Set cross-type ORDER proc for caller */
8 pg@bowt.ie 757 :GNC 12 : fmgr_info_cxt(cmp_proc, orderproc, so->arrayContext);
758 : :
759 : : /* Done if caller doesn't actually have an array they'll need to sort */
760 [ + + ]: 12 : if (!sortprocp)
761 : 9 : return;
762 : :
763 : : /*
764 : : * Look up the appropriate same-type comparison function in the opfamily.
765 : : *
766 : : * Note: it's possible that this would fail, if the opfamily is
767 : : * incomplete, but it seems quite unlikely that an opfamily would omit
768 : : * non-cross-type comparison procs for any datatype that it supports at
769 : : * all.
770 : : */
771 : 3 : cmp_proc = get_opfamily_proc(rel->rd_opfamily[skey->sk_attno - 1],
772 : : elemtype, elemtype, BTORDER_PROC);
773 [ - + ]: 3 : if (!RegProcedureIsValid(cmp_proc))
8 pg@bowt.ie 774 [ # # ]:UNC 0 : elog(ERROR, "missing support function %d(%u,%u) for attribute %d of index \"%s\"",
775 : : BTORDER_PROC, elemtype, elemtype,
776 : : skey->sk_attno, RelationGetRelationName(rel));
777 : :
778 : : /* Set same-type ORDER proc for caller */
8 pg@bowt.ie 779 :GNC 3 : fmgr_info_cxt(cmp_proc, *sortprocp, so->arrayContext);
780 : : }
781 : :
782 : : /*
783 : : * _bt_find_extreme_element() -- get least or greatest array element
784 : : *
785 : : * scan and skey identify the index column, whose opfamily determines the
786 : : * comparison semantics. strat should be BTLessStrategyNumber to get the
787 : : * least element, or BTGreaterStrategyNumber to get the greatest.
788 : : */
789 : : static Datum
790 : 3 : _bt_find_extreme_element(IndexScanDesc scan, ScanKey skey, Oid elemtype,
791 : : StrategyNumber strat,
792 : : Datum *elems, int nelems)
793 : : {
4564 tgl@sss.pgh.pa.us 794 :CBC 3 : Relation rel = scan->indexRelation;
795 : : Oid cmp_op;
796 : : RegProcedure cmp_proc;
797 : : FmgrInfo flinfo;
798 : : Datum result;
799 : : int i;
800 : :
801 : : /*
802 : : * Look up the appropriate comparison operator in the opfamily.
803 : : *
804 : : * Note: it's possible that this would fail, if the opfamily is
805 : : * incomplete, but it seems quite unlikely that an opfamily would omit
806 : : * non-cross-type comparison operators for any datatype that it supports
807 : : * at all.
808 : : */
8 pg@bowt.ie 809 [ - + ]:GNC 3 : Assert(skey->sk_strategy != BTEqualStrategyNumber);
810 [ - + ]: 3 : Assert(OidIsValid(elemtype));
4564 tgl@sss.pgh.pa.us 811 :CBC 3 : cmp_op = get_opfamily_member(rel->rd_opfamily[skey->sk_attno - 1],
812 : : elemtype,
813 : : elemtype,
814 : : strat);
815 [ - + ]: 3 : if (!OidIsValid(cmp_op))
4564 tgl@sss.pgh.pa.us 816 [ # # ]:UBC 0 : elog(ERROR, "missing operator %d(%u,%u) in opfamily %u",
817 : : strat, elemtype, elemtype,
818 : : rel->rd_opfamily[skey->sk_attno - 1]);
4564 tgl@sss.pgh.pa.us 819 :CBC 3 : cmp_proc = get_opcode(cmp_op);
820 [ - + ]: 3 : if (!RegProcedureIsValid(cmp_proc))
4564 tgl@sss.pgh.pa.us 821 [ # # ]:UBC 0 : elog(ERROR, "missing oprcode for operator %u", cmp_op);
822 : :
4564 tgl@sss.pgh.pa.us 823 :CBC 3 : fmgr_info(cmp_proc, &flinfo);
824 : :
825 [ - + ]: 3 : Assert(nelems > 0);
826 : 3 : result = elems[0];
827 [ + + ]: 6 : for (i = 1; i < nelems; i++)
828 : : {
829 [ - + ]: 3 : if (DatumGetBool(FunctionCall2Coll(&flinfo,
830 : : skey->sk_collation,
831 : 3 : elems[i],
832 : : result)))
4564 tgl@sss.pgh.pa.us 833 :UBC 0 : result = elems[i];
834 : : }
835 : :
4564 tgl@sss.pgh.pa.us 836 :CBC 3 : return result;
837 : : }
838 : :
839 : : /*
840 : : * _bt_sort_array_elements() -- sort and de-dup array elements
841 : : *
842 : : * The array elements are sorted in-place, and the new number of elements
843 : : * after duplicate removal is returned.
844 : : *
845 : : * skey identifies the index column whose opfamily determines the comparison
846 : : * semantics, and sortproc is a corresponding ORDER proc. If reverse is true,
847 : : * we sort in descending order.
848 : : */
849 : : static int
8 pg@bowt.ie 850 :GNC 491 : _bt_sort_array_elements(ScanKey skey, FmgrInfo *sortproc, bool reverse,
851 : : Datum *elems, int nelems)
852 : : {
853 : : BTSortArrayContext cxt;
854 : :
4564 tgl@sss.pgh.pa.us 855 [ + + ]:CBC 491 : if (nelems <= 1)
856 : 6 : return nelems; /* no work to do */
857 : :
858 : : /* Sort the array elements */
8 pg@bowt.ie 859 :GNC 485 : cxt.sortproc = sortproc;
4564 tgl@sss.pgh.pa.us 860 :CBC 485 : cxt.collation = skey->sk_collation;
861 : 485 : cxt.reverse = reverse;
432 peter@eisentraut.org 862 : 485 : qsort_arg(elems, nelems, sizeof(Datum),
863 : : _bt_compare_array_elements, &cxt);
864 : :
865 : : /* Now scan the sorted elements and remove duplicates */
1620 tmunro@postgresql.or 866 : 485 : return qunique_arg(elems, nelems, sizeof(Datum),
867 : : _bt_compare_array_elements, &cxt);
868 : : }
869 : :
870 : : /*
871 : : * _bt_merge_arrays() -- merge next array's elements into an original array
872 : : *
873 : : * Called when preprocessing encounters a pair of array equality scan keys,
874 : : * both against the same index attribute (during initial array preprocessing).
875 : : * Merging reorganizes caller's original array (the left hand arg) in-place,
876 : : * without ever copying elements from one array into the other. (Mixing the
877 : : * elements together like this would be wrong, since they don't necessarily
878 : : * use the same underlying element type, despite all the other similarities.)
879 : : *
880 : : * Both arrays must have already been sorted and deduplicated by calling
881 : : * _bt_sort_array_elements. sortproc is the same-type ORDER proc that was
882 : : * just used to sort and deduplicate caller's "next" array. We'll usually be
883 : : * able to reuse that order PROC to merge the arrays together now. If not,
884 : : * then we'll perform a separate ORDER proc lookup.
885 : : *
886 : : * If the opfamily doesn't supply a complete set of cross-type ORDER procs we
887 : : * may not be able to determine which elements are contradictory. If we have
888 : : * the required ORDER proc then we return true (and validly set *nelems_orig),
889 : : * guaranteeing that at least the next array can be considered redundant. We
890 : : * return false if the required comparisons cannot not be made (caller must
891 : : * keep both arrays when this happens).
892 : : */
893 : : static bool
8 pg@bowt.ie 894 :GNC 6 : _bt_merge_arrays(IndexScanDesc scan, ScanKey skey, FmgrInfo *sortproc,
895 : : bool reverse, Oid origelemtype, Oid nextelemtype,
896 : : Datum *elems_orig, int *nelems_orig,
897 : : Datum *elems_next, int nelems_next)
898 : : {
899 : 6 : Relation rel = scan->indexRelation;
4564 tgl@sss.pgh.pa.us 900 : 6 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
901 : : BTSortArrayContext cxt;
8 pg@bowt.ie 902 : 6 : int nelems_orig_start = *nelems_orig,
903 : 6 : nelems_orig_merged = 0;
904 : 6 : FmgrInfo *mergeproc = sortproc;
905 : : FmgrInfo crosstypeproc;
906 : :
907 [ - + ]: 6 : Assert(skey->sk_strategy == BTEqualStrategyNumber);
908 [ + - - + ]: 6 : Assert(OidIsValid(origelemtype) && OidIsValid(nextelemtype));
909 : :
910 [ + + ]: 6 : if (origelemtype != nextelemtype)
911 : : {
912 : : RegProcedure cmp_proc;
913 : :
914 : : /*
915 : : * Cross-array-element-type merging is required, so can't just reuse
916 : : * sortproc when merging
917 : : */
918 : 3 : cmp_proc = get_opfamily_proc(rel->rd_opfamily[skey->sk_attno - 1],
919 : : origelemtype, nextelemtype, BTORDER_PROC);
920 [ - + ]: 3 : if (!RegProcedureIsValid(cmp_proc))
921 : : {
922 : : /* Can't make the required comparisons */
8 pg@bowt.ie 923 :UNC 0 : return false;
924 : : }
925 : :
926 : : /* We have all we need to determine redundancy/contradictoriness */
8 pg@bowt.ie 927 :GNC 3 : mergeproc = &crosstypeproc;
928 : 3 : fmgr_info_cxt(cmp_proc, mergeproc, so->arrayContext);
929 : : }
930 : :
931 : 6 : cxt.sortproc = mergeproc;
932 : 6 : cxt.collation = skey->sk_collation;
933 : 6 : cxt.reverse = reverse;
934 : :
935 [ + + + + ]: 27 : for (int i = 0, j = 0; i < nelems_orig_start && j < nelems_next;)
936 : : {
937 : 21 : Datum *oelem = elems_orig + i,
938 : 21 : *nelem = elems_next + j;
939 : 21 : int res = _bt_compare_array_elements(oelem, nelem, &cxt);
940 : :
941 [ + + ]: 21 : if (res == 0)
942 : : {
943 : 3 : elems_orig[nelems_orig_merged++] = *oelem;
944 : 3 : i++;
945 : 3 : j++;
946 : : }
947 [ + + ]: 18 : else if (res < 0)
948 : 12 : i++;
949 : : else /* res > 0 */
950 : 6 : j++;
951 : : }
952 : :
953 : 6 : *nelems_orig = nelems_orig_merged;
954 : :
955 : 6 : return true;
956 : : }
957 : :
958 : : /*
959 : : * Compare an array scan key to a scalar scan key, eliminating contradictory
960 : : * array elements such that the scalar scan key becomes redundant.
961 : : *
962 : : * Array elements can be eliminated as contradictory when excluded by some
963 : : * other operator on the same attribute. For example, with an index scan qual
964 : : * "WHERE a IN (1, 2, 3) AND a < 2", all array elements except the value "1"
965 : : * are eliminated, and the < scan key is eliminated as redundant. Cases where
966 : : * every array element is eliminated by a redundant scalar scan key have an
967 : : * unsatisfiable qual, which we handle by setting *qual_ok=false for caller.
968 : : *
969 : : * If the opfamily doesn't supply a complete set of cross-type ORDER procs we
970 : : * may not be able to determine which elements are contradictory. If we have
971 : : * the required ORDER proc then we return true (and validly set *qual_ok),
972 : : * guaranteeing that at least the scalar scan key can be considered redundant.
973 : : * We return false if the comparison could not be made (caller must keep both
974 : : * scan keys when this happens).
975 : : */
976 : : static bool
977 : 15 : _bt_compare_array_scankey_args(IndexScanDesc scan, ScanKey arraysk, ScanKey skey,
978 : : FmgrInfo *orderproc, BTArrayKeyInfo *array,
979 : : bool *qual_ok)
980 : : {
981 : 15 : Relation rel = scan->indexRelation;
982 : 15 : Oid opcintype = rel->rd_opcintype[arraysk->sk_attno - 1];
983 : 15 : int cmpresult = 0,
984 : 15 : cmpexact = 0,
985 : : matchelem,
986 : 15 : new_nelems = 0;
987 : : FmgrInfo crosstypeproc;
988 : 15 : FmgrInfo *orderprocp = orderproc;
989 : :
990 [ - + ]: 15 : Assert(arraysk->sk_attno == skey->sk_attno);
991 [ - + ]: 15 : Assert(array->num_elems > 0);
992 [ - + ]: 15 : Assert(!(arraysk->sk_flags & (SK_ISNULL | SK_ROW_HEADER | SK_ROW_MEMBER)));
993 [ + - - + ]: 15 : Assert((arraysk->sk_flags & SK_SEARCHARRAY) &&
994 : : arraysk->sk_strategy == BTEqualStrategyNumber);
995 [ - + ]: 15 : Assert(!(skey->sk_flags & (SK_ISNULL | SK_ROW_HEADER | SK_ROW_MEMBER)));
996 [ - + - - ]: 15 : Assert(!(skey->sk_flags & SK_SEARCHARRAY) ||
997 : : skey->sk_strategy != BTEqualStrategyNumber);
998 : :
999 : : /*
1000 : : * _bt_binsrch_array_skey searches an array for the entry best matching a
1001 : : * datum of opclass input type for the index's attribute (on-disk type).
1002 : : * We can reuse the array's ORDER proc whenever the non-array scan key's
1003 : : * type is a match for the corresponding attribute's input opclass type.
1004 : : * Otherwise, we have to do another ORDER proc lookup so that our call to
1005 : : * _bt_binsrch_array_skey applies the correct comparator.
1006 : : *
1007 : : * Note: we have to support the convention that sk_subtype == InvalidOid
1008 : : * means the opclass input type; this is a hack to simplify life for
1009 : : * ScanKeyInit().
1010 : : */
1011 [ + + + - ]: 15 : if (skey->sk_subtype != opcintype && skey->sk_subtype != InvalidOid)
1012 : : {
1013 : : RegProcedure cmp_proc;
1014 : : Oid arraysk_elemtype;
1015 : :
1016 : : /*
1017 : : * Need an ORDER proc lookup to detect redundancy/contradictoriness
1018 : : * with this pair of scankeys.
1019 : : *
1020 : : * Scalar scan key's argument will be passed to _bt_compare_array_skey
1021 : : * as its tupdatum/lefthand argument (rhs arg is for array elements).
1022 : : */
1023 : 3 : arraysk_elemtype = arraysk->sk_subtype;
1024 [ - + ]: 3 : if (arraysk_elemtype == InvalidOid)
8 pg@bowt.ie 1025 :UNC 0 : arraysk_elemtype = rel->rd_opcintype[arraysk->sk_attno - 1];
8 pg@bowt.ie 1026 :GNC 3 : cmp_proc = get_opfamily_proc(rel->rd_opfamily[arraysk->sk_attno - 1],
1027 : : skey->sk_subtype, arraysk_elemtype,
1028 : : BTORDER_PROC);
1029 [ - + ]: 3 : if (!RegProcedureIsValid(cmp_proc))
1030 : : {
1031 : : /* Can't make the comparison */
8 pg@bowt.ie 1032 :UNC 0 : *qual_ok = false; /* suppress compiler warnings */
1033 : 0 : return false;
1034 : : }
1035 : :
1036 : : /* We have all we need to determine redundancy/contradictoriness */
8 pg@bowt.ie 1037 :GNC 3 : orderprocp = &crosstypeproc;
1038 : 3 : fmgr_info(cmp_proc, orderprocp);
1039 : : }
1040 : :
1041 : 15 : matchelem = _bt_binsrch_array_skey(orderprocp, false,
1042 : : NoMovementScanDirection,
1043 : : skey->sk_argument, false, array,
1044 : : arraysk, &cmpresult);
1045 : :
1046 [ + - + + : 15 : switch (skey->sk_strategy)
+ - ]
1047 : : {
1048 : 3 : case BTLessStrategyNumber:
1049 : 3 : cmpexact = 1; /* exclude exact match, if any */
1050 : : /* FALL THRU */
1051 : 3 : case BTLessEqualStrategyNumber:
1052 [ - + ]: 3 : if (cmpresult >= cmpexact)
8 pg@bowt.ie 1053 :UNC 0 : matchelem++;
1054 : : /* Resize, keeping elements from the start of the array */
8 pg@bowt.ie 1055 :GNC 3 : new_nelems = matchelem;
1056 : 3 : break;
1057 : 6 : case BTEqualStrategyNumber:
1058 [ + + ]: 6 : if (cmpresult != 0)
1059 : : {
1060 : : /* qual is unsatisfiable */
1061 : 3 : new_nelems = 0;
1062 : : }
1063 : : else
1064 : : {
1065 : : /* Shift matching element to the start of the array, resize */
1066 : 3 : array->elem_values[0] = array->elem_values[matchelem];
1067 : 3 : new_nelems = 1;
1068 : : }
1069 : 6 : break;
1070 : 3 : case BTGreaterEqualStrategyNumber:
1071 : 3 : cmpexact = 1; /* include exact match, if any */
1072 : : /* FALL THRU */
1073 : 6 : case BTGreaterStrategyNumber:
1074 [ + + ]: 6 : if (cmpresult >= cmpexact)
1075 : 3 : matchelem++;
1076 : : /* Shift matching elements to the start of the array, resize */
1077 : 6 : new_nelems = array->num_elems - matchelem;
1078 : 6 : memmove(array->elem_values, array->elem_values + matchelem,
1079 : : sizeof(Datum) * new_nelems);
1080 : 6 : break;
8 pg@bowt.ie 1081 :UNC 0 : default:
1082 [ # # ]: 0 : elog(ERROR, "unrecognized StrategyNumber: %d",
1083 : : (int) skey->sk_strategy);
1084 : : break;
1085 : : }
1086 : :
8 pg@bowt.ie 1087 [ - + ]:GNC 15 : Assert(new_nelems >= 0);
1088 [ - + ]: 15 : Assert(new_nelems <= array->num_elems);
1089 : :
1090 : 15 : array->num_elems = new_nelems;
1091 : 15 : *qual_ok = new_nelems > 0;
1092 : :
1093 : 15 : return true;
1094 : : }
1095 : :
1096 : : /*
1097 : : * qsort_arg comparator for sorting array elements
1098 : : */
1099 : : static int
8 pg@bowt.ie 1100 :CBC 4538 : _bt_compare_array_elements(const void *a, const void *b, void *arg)
1101 : : {
1102 : 4538 : Datum da = *((const Datum *) a);
1103 : 4538 : Datum db = *((const Datum *) b);
1104 : 4538 : BTSortArrayContext *cxt = (BTSortArrayContext *) arg;
1105 : : int32 compare;
1106 : :
8 pg@bowt.ie 1107 :GNC 4538 : compare = DatumGetInt32(FunctionCall2Coll(cxt->sortproc,
1108 : : cxt->collation,
1109 : : da, db));
8 pg@bowt.ie 1110 [ + + ]:CBC 4538 : if (cxt->reverse)
8 pg@bowt.ie 1111 [ - + ]:GBC 15 : INVERT_COMPARE_RESULT(compare);
8 pg@bowt.ie 1112 :CBC 4538 : return compare;
1113 : : }
1114 : :
1115 : : /*
1116 : : * _bt_compare_array_skey() -- apply array comparison function
1117 : : *
1118 : : * Compares caller's tuple attribute value to a scan key/array element.
1119 : : * Helper function used during binary searches of SK_SEARCHARRAY arrays.
1120 : : *
1121 : : * This routine returns:
1122 : : * <0 if tupdatum < arrdatum;
1123 : : * 0 if tupdatum == arrdatum;
1124 : : * >0 if tupdatum > arrdatum.
1125 : : *
1126 : : * This is essentially the same interface as _bt_compare: both functions
1127 : : * compare the value that they're searching for to a binary search pivot.
1128 : : * However, unlike _bt_compare, this function's "tuple argument" comes first,
1129 : : * while its "array/scankey argument" comes second.
1130 : : */
1131 : : static inline int32
8 pg@bowt.ie 1132 :GNC 9033 : _bt_compare_array_skey(FmgrInfo *orderproc,
1133 : : Datum tupdatum, bool tupnull,
1134 : : Datum arrdatum, ScanKey cur)
1135 : : {
1136 : 9033 : int32 result = 0;
1137 : :
1138 [ - + ]: 9033 : Assert(cur->sk_strategy == BTEqualStrategyNumber);
1139 : :
1140 [ + + ]: 9033 : if (tupnull) /* NULL tupdatum */
1141 : : {
1142 [ + - ]: 3 : if (cur->sk_flags & SK_ISNULL)
1143 : 3 : result = 0; /* NULL "=" NULL */
8 pg@bowt.ie 1144 [ # # ]:UNC 0 : else if (cur->sk_flags & SK_BT_NULLS_FIRST)
1145 : 0 : result = -1; /* NULL "<" NOT_NULL */
1146 : : else
1147 : 0 : result = 1; /* NULL ">" NOT_NULL */
1148 : : }
8 pg@bowt.ie 1149 [ + + ]:GNC 9030 : else if (cur->sk_flags & SK_ISNULL) /* NOT_NULL tupdatum, NULL arrdatum */
1150 : : {
1151 [ - + ]: 6 : if (cur->sk_flags & SK_BT_NULLS_FIRST)
8 pg@bowt.ie 1152 :UNC 0 : result = 1; /* NOT_NULL ">" NULL */
1153 : : else
8 pg@bowt.ie 1154 :GNC 6 : result = -1; /* NOT_NULL "<" NULL */
1155 : : }
1156 : : else
1157 : : {
1158 : : /*
1159 : : * Like _bt_compare, we need to be careful of cross-type comparisons,
1160 : : * so the left value has to be the value that came from an index tuple
1161 : : */
1162 : 9024 : result = DatumGetInt32(FunctionCall2Coll(orderproc, cur->sk_collation,
1163 : : tupdatum, arrdatum));
1164 : :
1165 : : /*
1166 : : * We flip the sign by following the obvious rule: flip whenever the
1167 : : * column is a DESC column.
1168 : : *
1169 : : * _bt_compare does it the wrong way around (flip when *ASC*) in order
1170 : : * to compensate for passing its orderproc arguments backwards. We
1171 : : * don't need to play these games because we find it natural to pass
1172 : : * tupdatum as the left value (and arrdatum as the right value).
1173 : : */
1174 [ + + ]: 9024 : if (cur->sk_flags & SK_BT_DESC)
1175 [ + + ]: 21 : INVERT_COMPARE_RESULT(result);
1176 : : }
1177 : :
1178 : 9033 : return result;
1179 : : }
1180 : :
1181 : : /*
1182 : : * _bt_binsrch_array_skey() -- Binary search for next matching array key
1183 : : *
1184 : : * Returns an index to the first array element >= caller's tupdatum argument.
1185 : : * This convention is more natural for forwards scan callers, but that can't
1186 : : * really matter to backwards scan callers. Both callers require handling for
1187 : : * the case where the match we return is < tupdatum, and symmetric handling
1188 : : * for the case where our best match is > tupdatum.
1189 : : *
1190 : : * Also sets *set_elem_result to the result _bt_compare_array_skey returned
1191 : : * when we used it to compare the matching array element to tupdatum/tupnull.
1192 : : *
1193 : : * cur_elem_trig indicates if array advancement was triggered by this array's
1194 : : * scan key, and that the array is for a required scan key. We can apply this
1195 : : * information to find the next matching array element in the current scan
1196 : : * direction using far fewer comparisons (fewer on average, compared to naive
1197 : : * binary search). This scheme takes advantage of an important property of
1198 : : * required arrays: required arrays always advance in lockstep with the index
1199 : : * scan's progress through the index's key space.
1200 : : */
1201 : : static int
1202 : 2033 : _bt_binsrch_array_skey(FmgrInfo *orderproc,
1203 : : bool cur_elem_trig, ScanDirection dir,
1204 : : Datum tupdatum, bool tupnull,
1205 : : BTArrayKeyInfo *array, ScanKey cur,
1206 : : int32 *set_elem_result)
1207 : : {
1208 : 2033 : int low_elem = 0,
1209 : 2033 : mid_elem = -1,
1210 : 2033 : high_elem = array->num_elems - 1,
1211 : 2033 : result = 0;
1212 : : Datum arrdatum;
1213 : :
1214 [ - + ]: 2033 : Assert(cur->sk_flags & SK_SEARCHARRAY);
1215 [ - + ]: 2033 : Assert(cur->sk_strategy == BTEqualStrategyNumber);
1216 : :
1217 [ + + ]: 2033 : if (cur_elem_trig)
1218 : : {
1219 [ - + ]: 1790 : Assert(!ScanDirectionIsNoMovement(dir));
1220 [ - + ]: 1790 : Assert(cur->sk_flags & SK_BT_REQFWD);
1221 : :
1222 : : /*
1223 : : * When the scan key that triggered array advancement is a required
1224 : : * array scan key, it is now certain that the current array element
1225 : : * (plus all prior elements relative to the current scan direction)
1226 : : * cannot possibly be at or ahead of the corresponding tuple value.
1227 : : * (_bt_checkkeys must have called _bt_tuple_before_array_skeys, which
1228 : : * makes sure this is true as a condition of advancing the arrays.)
1229 : : *
1230 : : * This makes it safe to exclude array elements up to and including
1231 : : * the former-current array element from our search.
1232 : : *
1233 : : * Separately, when array advancement was triggered by a required scan
1234 : : * key, the array element immediately after the former-current element
1235 : : * is often either an exact tupdatum match, or a "close by" near-match
1236 : : * (a near-match tupdatum is one whose key space falls _between_ the
1237 : : * former-current and new-current array elements). We'll detect both
1238 : : * cases via an optimistic comparison of the new search lower bound
1239 : : * (or new search upper bound in the case of backwards scans).
1240 : : */
1241 [ + + ]: 1790 : if (ScanDirectionIsForward(dir))
1242 : : {
1243 : 1778 : low_elem = array->cur_elem + 1; /* old cur_elem exhausted */
1244 : :
1245 : : /* Compare prospective new cur_elem (also the new lower bound) */
1246 [ + + ]: 1778 : if (high_elem >= low_elem)
1247 : : {
1248 : 1479 : arrdatum = array->elem_values[low_elem];
1249 : 1479 : result = _bt_compare_array_skey(orderproc, tupdatum, tupnull,
1250 : : arrdatum, cur);
1251 : :
1252 [ + + ]: 1479 : if (result <= 0)
1253 : : {
1254 : : /* Optimistic comparison optimization worked out */
1255 : 1449 : *set_elem_result = result;
1256 : 1449 : return low_elem;
1257 : : }
1258 : 30 : mid_elem = low_elem;
1259 : 30 : low_elem++; /* this cur_elem exhausted, too */
1260 : : }
1261 : :
1262 [ + + ]: 329 : if (high_elem < low_elem)
1263 : : {
1264 : : /* Caller needs to perform "beyond end" array advancement */
1265 : 299 : *set_elem_result = 1;
1266 : 299 : return high_elem;
1267 : : }
1268 : : }
1269 : : else
1270 : : {
1271 : 12 : high_elem = array->cur_elem - 1; /* old cur_elem exhausted */
1272 : :
1273 : : /* Compare prospective new cur_elem (also the new upper bound) */
1274 [ + + ]: 12 : if (high_elem >= low_elem)
1275 : : {
1276 : 9 : arrdatum = array->elem_values[high_elem];
1277 : 9 : result = _bt_compare_array_skey(orderproc, tupdatum, tupnull,
1278 : : arrdatum, cur);
1279 : :
1280 [ + - ]: 9 : if (result >= 0)
1281 : : {
1282 : : /* Optimistic comparison optimization worked out */
1283 : 9 : *set_elem_result = result;
1284 : 9 : return high_elem;
1285 : : }
8 pg@bowt.ie 1286 :UNC 0 : mid_elem = high_elem;
1287 : 0 : high_elem--; /* this cur_elem exhausted, too */
1288 : : }
1289 : :
8 pg@bowt.ie 1290 [ + - ]:GNC 3 : if (high_elem < low_elem)
1291 : : {
1292 : : /* Caller needs to perform "beyond end" array advancement */
1293 : 3 : *set_elem_result = -1;
1294 : 3 : return low_elem;
1295 : : }
1296 : : }
1297 : : }
1298 : :
1299 [ + + ]: 501 : while (high_elem > low_elem)
1300 : : {
1301 : 299 : mid_elem = low_elem + ((high_elem - low_elem) / 2);
1302 : 299 : arrdatum = array->elem_values[mid_elem];
1303 : :
1304 : 299 : result = _bt_compare_array_skey(orderproc, tupdatum, tupnull,
1305 : : arrdatum, cur);
1306 : :
1307 [ + + ]: 299 : if (result == 0)
1308 : : {
1309 : : /*
1310 : : * It's safe to quit as soon as we see an equal array element.
1311 : : * This often saves an extra comparison or two...
1312 : : */
1313 : 71 : low_elem = mid_elem;
4564 tgl@sss.pgh.pa.us 1314 : 71 : break;
1315 : : }
1316 : :
8 pg@bowt.ie 1317 [ + + ]: 228 : if (result > 0)
1318 : 201 : low_elem = mid_elem + 1;
1319 : : else
1320 : 27 : high_elem = mid_elem;
1321 : : }
1322 : :
1323 : : /*
1324 : : * ...but our caller also cares about how its searched-for tuple datum
1325 : : * compares to the low_elem datum. Must always set *set_elem_result with
1326 : : * the result of that comparison specifically.
1327 : : */
1328 [ + + ]: 273 : if (low_elem != mid_elem)
1329 : 181 : result = _bt_compare_array_skey(orderproc, tupdatum, tupnull,
1330 : 181 : array->elem_values[low_elem], cur);
1331 : :
1332 : 273 : *set_elem_result = result;
1333 : :
1334 : 273 : return low_elem;
1335 : : }
1336 : :
1337 : : /*
1338 : : * _bt_start_array_keys() -- Initialize array keys at start of a scan
1339 : : *
1340 : : * Set up the cur_elem counters and fill in the first sk_argument value for
1341 : : * each array scankey.
1342 : : */
1343 : : void
8 pg@bowt.ie 1344 :CBC 724 : _bt_start_array_keys(IndexScanDesc scan, ScanDirection dir)
1345 : : {
4217 tgl@sss.pgh.pa.us 1346 : 724 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
1347 : : int i;
1348 : :
8 pg@bowt.ie 1349 [ - + ]:GNC 724 : Assert(so->numArrayKeys);
1350 [ - + ]: 724 : Assert(so->qual_ok);
1351 : :
4217 tgl@sss.pgh.pa.us 1352 [ + + ]:CBC 1580 : for (i = 0; i < so->numArrayKeys; i++)
1353 : : {
1354 : 856 : BTArrayKeyInfo *curArrayKey = &so->arrayKeys[i];
8 pg@bowt.ie 1355 :GNC 856 : ScanKey skey = &so->keyData[curArrayKey->scan_key];
1356 : :
8 pg@bowt.ie 1357 [ - + ]:CBC 856 : Assert(curArrayKey->num_elems > 0);
8 pg@bowt.ie 1358 [ - + ]:GNC 856 : Assert(skey->sk_flags & SK_SEARCHARRAY);
1359 : :
8 pg@bowt.ie 1360 [ + + ]:CBC 856 : if (ScanDirectionIsBackward(dir))
8 pg@bowt.ie 1361 :GBC 398 : curArrayKey->cur_elem = curArrayKey->num_elems - 1;
1362 : : else
8 pg@bowt.ie 1363 :CBC 458 : curArrayKey->cur_elem = 0;
1364 : 856 : skey->sk_argument = curArrayKey->elem_values[curArrayKey->cur_elem];
1365 : : }
8 pg@bowt.ie 1366 :GNC 724 : so->scanBehind = false;
4217 tgl@sss.pgh.pa.us 1367 :CBC 724 : }
1368 : :
1369 : : /*
1370 : : * _bt_advance_array_keys_increment() -- Advance to next set of array elements
1371 : : *
1372 : : * Advances the array keys by a single increment in the current scan
1373 : : * direction. When there are multiple array keys this can roll over from the
1374 : : * lowest order array to higher order arrays.
1375 : : *
1376 : : * Returns true if there is another set of values to consider, false if not.
1377 : : * On true result, the scankeys are initialized with the next set of values.
1378 : : * On false result, the scankeys stay the same, and the array keys are not
1379 : : * advanced (every array remains at its final element for scan direction).
1380 : : */
1381 : : static bool
8 pg@bowt.ie 1382 :GNC 345 : _bt_advance_array_keys_increment(IndexScanDesc scan, ScanDirection dir)
1383 : : {
4217 tgl@sss.pgh.pa.us 1384 :CBC 345 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
1385 : :
1386 : : /*
1387 : : * We must advance the last array key most quickly, since it will
1388 : : * correspond to the lowest-order index column among the available
1389 : : * qualifications
1390 : : */
8 pg@bowt.ie 1391 [ + + ]:GNC 740 : for (int i = so->numArrayKeys - 1; i >= 0; i--)
1392 : : {
4217 tgl@sss.pgh.pa.us 1393 :CBC 411 : BTArrayKeyInfo *curArrayKey = &so->arrayKeys[i];
8 pg@bowt.ie 1394 :GNC 411 : ScanKey skey = &so->keyData[curArrayKey->scan_key];
8 pg@bowt.ie 1395 :CBC 411 : int cur_elem = curArrayKey->cur_elem;
1396 : 411 : int num_elems = curArrayKey->num_elems;
8 pg@bowt.ie 1397 :GNC 411 : bool rolled = false;
1398 : :
1399 [ + + + + ]: 411 : if (ScanDirectionIsForward(dir) && ++cur_elem >= num_elems)
1400 : : {
1401 : 392 : cur_elem = 0;
1402 : 392 : rolled = true;
1403 : : }
1404 [ + + + - ]: 19 : else if (ScanDirectionIsBackward(dir) && --cur_elem < 0)
1405 : : {
1406 : 3 : cur_elem = num_elems - 1;
1407 : 3 : rolled = true;
1408 : : }
1409 : :
8 pg@bowt.ie 1410 :CBC 411 : curArrayKey->cur_elem = cur_elem;
1411 : 411 : skey->sk_argument = curArrayKey->elem_values[cur_elem];
8 pg@bowt.ie 1412 [ + + ]:GNC 411 : if (!rolled)
1413 : 16 : return true;
1414 : :
1415 : : /* Need to advance next array key, if any */
1416 : : }
1417 : :
1418 : : /*
1419 : : * The array keys are now exhausted. (There isn't actually a distinct
1420 : : * state that represents array exhaustion, since index scans don't always
1421 : : * end after btgettuple returns "false".)
1422 : : *
1423 : : * Restore the array keys to the state they were in immediately before we
1424 : : * were called. This ensures that the arrays only ever ratchet in the
1425 : : * current scan direction. Without this, scans would overlook matching
1426 : : * tuples if and when the scan's direction was subsequently reversed.
1427 : : */
1428 : 329 : _bt_start_array_keys(scan, -dir);
1429 : :
1430 : 329 : return false;
1431 : : }
1432 : :
1433 : : /*
1434 : : * _bt_rewind_nonrequired_arrays() -- Rewind non-required arrays
1435 : : *
1436 : : * Called when _bt_advance_array_keys decides to start a new primitive index
1437 : : * scan on the basis of the current scan position being before the position
1438 : : * that _bt_first is capable of repositioning the scan to by applying an
1439 : : * inequality operator required in the opposite-to-scan direction only.
1440 : : *
1441 : : * Although equality strategy scan keys (for both arrays and non-arrays alike)
1442 : : * are either marked required in both directions or in neither direction,
1443 : : * there is a sense in which non-required arrays behave like required arrays.
1444 : : * With a qual such as "WHERE a IN (100, 200) AND b >= 3 AND c IN (5, 6, 7)",
1445 : : * the scan key on "c" is non-required, but nevertheless enables positioning
1446 : : * the scan at the first tuple >= "(100, 3, 5)" on the leaf level during the
1447 : : * first descent of the tree by _bt_first. Later on, there could also be a
1448 : : * second descent, that places the scan right before tuples >= "(200, 3, 5)".
1449 : : * _bt_first must never be allowed to build an insertion scan key whose "c"
1450 : : * entry is set to a value other than 5, the "c" array's first element/value.
1451 : : * (Actually, it's the first in the current scan direction. This example uses
1452 : : * a forward scan.)
1453 : : *
1454 : : * Calling here resets the array scan key elements for the scan's non-required
1455 : : * arrays. This is strictly necessary for correctness in a subset of cases
1456 : : * involving "required in opposite direction"-triggered primitive index scans.
1457 : : * Not all callers are at risk of _bt_first using a non-required array like
1458 : : * this, but advancement always resets the arrays when another primitive scan
1459 : : * is scheduled, just to keep things simple. Array advancement even makes
1460 : : * sure to reset non-required arrays during scans that have no inequalities.
1461 : : * (Advancement still won't call here when there are no inequalities, though
1462 : : * that's just because it's all handled indirectly instead.)
1463 : : *
1464 : : * Note: _bt_verify_arrays_bt_first is called by an assertion to enforce that
1465 : : * everybody got this right.
1466 : : */
1467 : : static void
8 pg@bowt.ie 1468 :UNC 0 : _bt_rewind_nonrequired_arrays(IndexScanDesc scan, ScanDirection dir)
1469 : : {
8 pg@bowt.ie 1470 :LBC (3) : BTScanOpaque so = (BTScanOpaque) scan->opaque;
8 pg@bowt.ie 1471 :UNC 0 : int arrayidx = 0;
1472 : :
1473 [ # # ]: 0 : for (int ikey = 0; ikey < so->numberOfKeys; ikey++)
1474 : : {
1475 : 0 : ScanKey cur = so->keyData + ikey;
1476 : 0 : BTArrayKeyInfo *array = NULL;
1477 : : int first_elem_dir;
1478 : :
1479 [ # # ]: 0 : if (!(cur->sk_flags & SK_SEARCHARRAY) ||
1480 [ # # ]: 0 : cur->sk_strategy != BTEqualStrategyNumber)
1481 : 0 : continue;
1482 : :
1483 : 0 : array = &so->arrayKeys[arrayidx++];
1484 [ # # ]: 0 : Assert(array->scan_key == ikey);
1485 : :
1486 [ # # ]: 0 : if ((cur->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)))
1487 : 0 : continue;
1488 : :
1489 [ # # ]: 0 : if (ScanDirectionIsForward(dir))
1490 : 0 : first_elem_dir = 0;
1491 : : else
1492 : 0 : first_elem_dir = array->num_elems - 1;
1493 : :
1494 [ # # ]: 0 : if (array->cur_elem != first_elem_dir)
1495 : : {
1496 : 0 : array->cur_elem = first_elem_dir;
1497 : 0 : cur->sk_argument = array->elem_values[first_elem_dir];
1498 : : }
1499 : : }
8 pg@bowt.ie 1500 :LBC (3) : }
1501 : :
1502 : : /*
1503 : : * _bt_tuple_before_array_skeys() -- too early to advance required arrays?
1504 : : *
1505 : : * We always compare the tuple using the current array keys (which we assume
1506 : : * are already set in so->keyData[]). readpagetup indicates if tuple is the
1507 : : * scan's current _bt_readpage-wise tuple.
1508 : : *
1509 : : * readpagetup callers must only call here when _bt_check_compare already set
1510 : : * continuescan=false. We help these callers deal with _bt_check_compare's
1511 : : * inability to distinguishing between the < and > cases (it uses equality
1512 : : * operator scan keys, whereas we use 3-way ORDER procs). These callers pass
1513 : : * a _bt_check_compare-set sktrig value that indicates which scan key
1514 : : * triggered the call (!readpagetup callers just pass us sktrig=0 instead).
1515 : : * This information allows us to avoid wastefully checking earlier scan keys
1516 : : * that were already deemed to have been satisfied inside _bt_check_compare.
1517 : : *
1518 : : * Returns false when caller's tuple is >= the current required equality scan
1519 : : * keys (or <=, in the case of backwards scans). This happens to readpagetup
1520 : : * callers when the scan has reached the point of needing its array keys
1521 : : * advanced; caller will need to advance required and non-required arrays at
1522 : : * scan key offsets >= sktrig, plus scan keys < sktrig iff sktrig rolls over.
1523 : : * (When we return false to readpagetup callers, tuple can only be == current
1524 : : * required equality scan keys when caller's sktrig indicates that the arrays
1525 : : * need to be advanced due to an unsatisfied required inequality key trigger.)
1526 : : *
1527 : : * Returns true when caller passes a tuple that is < the current set of
1528 : : * equality keys for the most significant non-equal required scan key/column
1529 : : * (or > the keys, during backwards scans). This happens to readpagetup
1530 : : * callers when tuple is still before the start of matches for the scan's
1531 : : * required equality strategy scan keys. (sktrig can't have indicated that an
1532 : : * inequality strategy scan key wasn't satisfied in _bt_check_compare when we
1533 : : * return true. In fact, we automatically return false when passed such an
1534 : : * inequality sktrig by readpagetup callers -- _bt_check_compare's initial
1535 : : * continuescan=false doesn't really need to be confirmed here by us.)
1536 : : *
1537 : : * !readpagetup callers optionally pass us *scanBehind, which tracks whether
1538 : : * any missing truncated attributes might have affected array advancement
1539 : : * (compared to what would happen if it was shown the first non-pivot tuple on
1540 : : * the page to the right of caller's finaltup/high key tuple instead). It's
1541 : : * only possible that we'll set *scanBehind to true when caller passes us a
1542 : : * pivot tuple (with truncated -inf attributes) that we return false for.
1543 : : */
1544 : : static bool
8 pg@bowt.ie 1545 :GNC 6675 : _bt_tuple_before_array_skeys(IndexScanDesc scan, ScanDirection dir,
1546 : : IndexTuple tuple, TupleDesc tupdesc, int tupnatts,
1547 : : bool readpagetup, int sktrig, bool *scanBehind)
1548 : : {
8 pg@bowt.ie 1549 :CBC 6675 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
1550 : :
8 pg@bowt.ie 1551 [ - + ]:GNC 6675 : Assert(so->numArrayKeys);
1552 [ - + ]: 6675 : Assert(so->numberOfKeys);
1553 [ + + - + ]: 6675 : Assert(sktrig == 0 || readpagetup);
1554 [ + + - + ]: 6675 : Assert(!readpagetup || scanBehind == NULL);
1555 : :
1556 [ + + ]: 6675 : if (scanBehind)
1557 : 543 : *scanBehind = false;
1558 : :
1559 [ + + ]: 6906 : for (int ikey = sktrig; ikey < so->numberOfKeys; ikey++)
1560 : : {
1561 : 6898 : ScanKey cur = so->keyData + ikey;
1562 : : Datum tupdatum;
1563 : : bool tupnull;
1564 : : int32 result;
1565 : :
1566 : : /* readpagetup calls require one ORDER proc comparison (at most) */
1567 [ + + - + ]: 6898 : Assert(!readpagetup || ikey == sktrig);
1568 : :
1569 : : /*
1570 : : * Once we reach a non-required scan key, we're completely done.
1571 : : *
1572 : : * Note: we deliberately don't consider the scan direction here.
1573 : : * _bt_advance_array_keys caller requires that we track *scanBehind
1574 : : * without concern for scan direction.
1575 : : */
1576 [ + + ]: 6898 : if ((cur->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)) == 0)
1577 : : {
1578 [ - + ]: 3 : Assert(!readpagetup);
1579 [ - + - - ]: 3 : Assert(ikey > sktrig || ikey == 0);
1580 : 6667 : return false;
1581 : : }
1582 : :
1583 [ + + ]: 6895 : if (cur->sk_attno > tupnatts)
1584 : : {
1585 [ - + ]: 3 : Assert(!readpagetup);
1586 : :
1587 : : /*
1588 : : * When we reach a high key's truncated attribute, assume that the
1589 : : * tuple attribute's value is >= the scan's equality constraint
1590 : : * scan keys (but set *scanBehind to let interested callers know
1591 : : * that a truncated attribute might have affected our answer).
1592 : : */
1593 [ + - ]: 3 : if (scanBehind)
1594 : 3 : *scanBehind = true;
1595 : :
1596 : 3 : return false;
1597 : : }
1598 : :
1599 : : /*
1600 : : * Deal with inequality strategy scan keys that _bt_check_compare set
1601 : : * continuescan=false for
1602 : : */
1603 [ + + ]: 6892 : if (cur->sk_strategy != BTEqualStrategyNumber)
1604 : : {
1605 : : /*
1606 : : * When _bt_check_compare indicated that a required inequality
1607 : : * scan key wasn't satisfied, there's no need to verify anything;
1608 : : * caller always calls _bt_advance_array_keys with this sktrig.
1609 : : */
1610 [ + + ]: 9 : if (readpagetup)
1611 : 3 : return false;
1612 : :
1613 : : /*
1614 : : * Otherwise we can't give up, since we must check all required
1615 : : * scan keys (required in either direction) in order to correctly
1616 : : * track *scanBehind for caller
1617 : : */
1618 : 6 : continue;
1619 : : }
1620 : :
1621 : 6883 : tupdatum = index_getattr(tuple, cur->sk_attno, tupdesc, &tupnull);
1622 : :
1623 : 6883 : result = _bt_compare_array_skey(&so->orderProcs[ikey],
1624 : : tupdatum, tupnull,
1625 : : cur->sk_argument, cur);
1626 : :
1627 : : /*
1628 : : * Does this comparison indicate that caller must _not_ advance the
1629 : : * scan's arrays just yet?
1630 : : */
1631 [ + + + + : 6883 : if ((ScanDirectionIsForward(dir) && result < 0) ||
+ + ]
1632 [ + + ]: 54 : (ScanDirectionIsBackward(dir) && result > 0))
1633 : 2495 : return true;
1634 : :
1635 : : /*
1636 : : * Does this comparison indicate that caller should now advance the
1637 : : * scan's arrays? (Must be if we get here during a readpagetup call.)
1638 : : */
1639 [ + + + + ]: 4388 : if (readpagetup || result != 0)
1640 : : {
1641 [ - + ]: 4163 : Assert(result != 0);
1642 : 4163 : return false;
1643 : : }
1644 : :
1645 : : /*
1646 : : * Inconclusive -- need to check later scan keys, too.
1647 : : *
1648 : : * This must be a finaltup precheck, or a call made from an assertion.
1649 : : */
1650 [ - + ]: 225 : Assert(result == 0);
1651 : : }
1652 : :
1653 [ - + ]: 8 : Assert(!readpagetup);
1654 : :
1655 : 8 : return false;
1656 : : }
1657 : :
1658 : : /*
1659 : : * _bt_start_prim_scan() -- start scheduled primitive index scan?
1660 : : *
1661 : : * Returns true if _bt_checkkeys scheduled another primitive index scan, just
1662 : : * as the last one ended. Otherwise returns false, indicating that the array
1663 : : * keys are now fully exhausted.
1664 : : *
1665 : : * Only call here during scans with one or more equality type array scan keys,
1666 : : * after _bt_first or _bt_next return false.
1667 : : */
1668 : : bool
1669 : 689 : _bt_start_prim_scan(IndexScanDesc scan, ScanDirection dir)
1670 : : {
1671 : 689 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
1672 : :
1673 [ - + ]: 689 : Assert(so->numArrayKeys);
1674 : :
1675 : : /* scanBehind flag doesn't persist across primitive index scans - reset */
1676 : 689 : so->scanBehind = false;
1677 : :
1678 : : /*
1679 : : * Array keys are advanced within _bt_checkkeys when the scan reaches the
1680 : : * leaf level (more precisely, they're advanced when the scan reaches the
1681 : : * end of each distinct set of array elements). This process avoids
1682 : : * repeat access to leaf pages (across multiple primitive index scans) by
1683 : : * advancing the scan's array keys when it allows the primitive index scan
1684 : : * to find nearby matching tuples (or when it eliminates ranges of array
1685 : : * key space that can't possibly be satisfied by any index tuple).
1686 : : *
1687 : : * _bt_checkkeys sets a simple flag variable to schedule another primitive
1688 : : * index scan. The flag tells us what to do.
1689 : : *
1690 : : * We cannot rely on _bt_first always reaching _bt_checkkeys. There are
1691 : : * various cases where that won't happen. For example, if the index is
1692 : : * completely empty, then _bt_first won't call _bt_readpage/_bt_checkkeys.
1693 : : * We also don't expect a call to _bt_checkkeys during searches for a
1694 : : * non-existent value that happens to be lower/higher than any existing
1695 : : * value in the index.
1696 : : *
1697 : : * We don't require special handling for these cases -- we don't need to
1698 : : * be explicitly instructed to _not_ perform another primitive index scan.
1699 : : * It's up to code under the control of _bt_first to always set the flag
1700 : : * when another primitive index scan will be required.
1701 : : *
1702 : : * This works correctly, even with the tricky cases listed above, which
1703 : : * all involve access to leaf pages "near the boundaries of the key space"
1704 : : * (whether it's from a leftmost/rightmost page, or an imaginary empty
1705 : : * leaf root page). If _bt_checkkeys cannot be reached by a primitive
1706 : : * index scan for one set of array keys, then it also won't be reached for
1707 : : * any later set ("later" in terms of the direction that we scan the index
1708 : : * and advance the arrays). The array keys won't have advanced in these
1709 : : * cases, but that's the correct behavior (even _bt_advance_array_keys
1710 : : * won't always advance the arrays at the point they become "exhausted").
1711 : : */
1712 [ + + ]: 689 : if (so->needPrimScan)
1713 : : {
1714 [ - + ]: 285 : Assert(_bt_verify_arrays_bt_first(scan, dir));
1715 : :
1716 : : /*
1717 : : * Flag was set -- must call _bt_first again, which will reset the
1718 : : * scan's needPrimScan flag
1719 : : */
1720 : 285 : return true;
1721 : : }
1722 : :
1723 : : /* The top-level index scan ran out of tuples in this scan direction */
1724 [ + + ]: 404 : if (scan->parallel_scan != NULL)
1725 : 15 : _bt_parallel_done(scan);
1726 : :
1727 : 404 : return false;
1728 : : }
1729 : :
1730 : : /*
1731 : : * _bt_advance_array_keys() -- Advance array elements using a tuple
1732 : : *
1733 : : * The scan always gets a new qual as a consequence of calling here (except
1734 : : * when we determine that the top-level scan has run out of matching tuples).
1735 : : * All later _bt_check_compare calls also use the same new qual that was first
1736 : : * used here (at least until the next call here advances the keys once again).
1737 : : * It's convenient to structure _bt_check_compare rechecks of caller's tuple
1738 : : * (using the new qual) as one the steps of advancing the scan's array keys,
1739 : : * so this function works as a wrapper around _bt_check_compare.
1740 : : *
1741 : : * Like _bt_check_compare, we'll set pstate.continuescan on behalf of the
1742 : : * caller, and return a boolean indicating if caller's tuple satisfies the
1743 : : * scan's new qual. But unlike _bt_check_compare, we set so->needPrimScan
1744 : : * when we set continuescan=false, indicating if a new primitive index scan
1745 : : * has been scheduled (otherwise, the top-level scan has run out of tuples in
1746 : : * the current scan direction).
1747 : : *
1748 : : * Caller must use _bt_tuple_before_array_skeys to determine if the current
1749 : : * place in the scan is >= the current array keys _before_ calling here.
1750 : : * We're responsible for ensuring that caller's tuple is <= the newly advanced
1751 : : * required array keys once we return. We try to find an exact match, but
1752 : : * failing that we'll advance the array keys to whatever set of array elements
1753 : : * comes next in the key space for the current scan direction. Required array
1754 : : * keys "ratchet forwards" (or backwards). They can only advance as the scan
1755 : : * itself advances through the index/key space.
1756 : : *
1757 : : * (The rules are the same for backwards scans, except that the operators are
1758 : : * flipped: just replace the precondition's >= operator with a <=, and the
1759 : : * postcondition's <= operator with with a >=. In other words, just swap the
1760 : : * precondition with the postcondition.)
1761 : : *
1762 : : * We also deal with "advancing" non-required arrays here. Callers whose
1763 : : * sktrig scan key is non-required specify sktrig_required=false. These calls
1764 : : * are the only exception to the general rule about always advancing the
1765 : : * required array keys (the scan may not even have a required array). These
1766 : : * callers should just pass a NULL pstate (since there is never any question
1767 : : * of stopping the scan). No call to _bt_tuple_before_array_skeys is required
1768 : : * ahead of these calls (it's already clear that any required scan keys must
1769 : : * be satisfied by caller's tuple).
1770 : : *
1771 : : * Note that we deal with non-array required equality strategy scan keys as
1772 : : * degenerate single element arrays here. Obviously, they can never really
1773 : : * advance in the way that real arrays can, but they must still affect how we
1774 : : * advance real array scan keys (exactly like true array equality scan keys).
1775 : : * We have to keep around a 3-way ORDER proc for these (using the "=" operator
1776 : : * won't do), since in general whether the tuple is < or > _any_ unsatisfied
1777 : : * required equality key influences how the scan's real arrays must advance.
1778 : : *
1779 : : * Note also that we may sometimes need to advance the array keys when the
1780 : : * existing required array keys (and other required equality keys) are already
1781 : : * an exact match for every corresponding value from caller's tuple. We must
1782 : : * do this for inequalities that _bt_check_compare set continuescan=false for.
1783 : : * They'll advance the array keys here, just like any other scan key that
1784 : : * _bt_check_compare stops on. (This can even happen _after_ we advance the
1785 : : * array keys, in which case we'll advance the array keys a second time. That
1786 : : * way _bt_checkkeys caller always has its required arrays advance to the
1787 : : * maximum possible extent that its tuple will allow.)
1788 : : */
1789 : : static bool
1790 : 1941 : _bt_advance_array_keys(IndexScanDesc scan, BTReadPageState *pstate,
1791 : : IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
1792 : : int sktrig, bool sktrig_required)
1793 : : {
1794 : 1941 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
1795 : 1941 : Relation rel = scan->indexRelation;
1796 [ + + ]: 1941 : ScanDirection dir = pstate ? pstate->dir : ForwardScanDirection;
1797 : 1941 : int arrayidx = 0;
1798 : 1941 : bool beyond_end_advance = false,
1799 : 1941 : has_required_opposite_direction_only = false,
1800 : 1941 : oppodir_inequality_sktrig = false,
1801 : 1941 : all_required_satisfied = true,
1802 : 1941 : all_satisfied = true;
1803 : :
1804 [ + + ]: 1941 : if (sktrig_required)
1805 : : {
1806 : : /*
1807 : : * Precondition array state assertion
1808 : : */
1809 [ - + ]: 1821 : Assert(!_bt_tuple_before_array_skeys(scan, dir, tuple, tupdesc,
1810 : : tupnatts, false, 0, NULL));
1811 : :
1812 : 1821 : so->scanBehind = false; /* reset */
1813 : :
1814 : : /*
1815 : : * Required scan key wasn't satisfied, so required arrays will have to
1816 : : * advance. Invalidate page-level state that tracks whether the
1817 : : * scan's required-in-opposite-direction-only keys are known to be
1818 : : * satisfied by page's remaining tuples.
1819 : : */
1820 : 1821 : pstate->firstmatch = false;
1821 : :
1822 : : /* Shouldn't have to invalidate 'prechecked', though */
1823 [ - + ]: 1821 : Assert(!pstate->prechecked);
1824 : :
1825 : : /*
1826 : : * Once we return we'll have a new set of required array keys, so
1827 : : * reset state used by "look ahead" optimization
1828 : : */
1829 : 1821 : pstate->rechecks = 0;
1830 : 1821 : pstate->targetdistance = 0;
1831 : : }
1832 : :
1833 [ - + ]: 1941 : Assert(_bt_verify_keys_with_arraykeys(scan));
1834 : :
1835 [ + + ]: 5330 : for (int ikey = 0; ikey < so->numberOfKeys; ikey++)
1836 : : {
1837 : 3497 : ScanKey cur = so->keyData + ikey;
1838 : 3497 : BTArrayKeyInfo *array = NULL;
1839 : : Datum tupdatum;
1840 : 3497 : bool required = false,
1841 : 3497 : required_opposite_direction_only = false,
1842 : : tupnull;
1843 : : int32 result;
1844 : 3497 : int set_elem = 0;
1845 : :
1846 [ + + ]: 3497 : if (cur->sk_strategy == BTEqualStrategyNumber)
1847 : : {
1848 : : /* Manage array state */
1849 [ + + ]: 3365 : if (cur->sk_flags & SK_SEARCHARRAY)
1850 : : {
1851 : 2337 : array = &so->arrayKeys[arrayidx++];
1852 [ - + ]: 2337 : Assert(array->scan_key == ikey);
1853 : : }
1854 : : }
1855 : : else
1856 : : {
1857 : : /*
1858 : : * Are any inequalities required in the opposite direction only
1859 : : * present here?
1860 : : */
1861 [ + - ]: 132 : if (((ScanDirectionIsForward(dir) &&
1862 [ + + - + ]: 132 : (cur->sk_flags & (SK_BT_REQBKWD))) ||
8 pg@bowt.ie 1863 :UNC 0 : (ScanDirectionIsBackward(dir) &&
1864 [ # # ]: 0 : (cur->sk_flags & (SK_BT_REQFWD)))))
8 pg@bowt.ie 1865 :GNC 9 : has_required_opposite_direction_only =
1866 : 9 : required_opposite_direction_only = true;
1867 : : }
1868 : :
1869 : : /* Optimization: skip over known-satisfied scan keys */
1870 [ + + ]: 3497 : if (ikey < sktrig)
1871 : 1297 : continue;
1872 : :
1873 [ + + ]: 3214 : if (cur->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD))
1874 : : {
1875 [ - + ]: 2454 : Assert(sktrig_required);
1876 : :
1877 : 2454 : required = true;
1878 : :
1879 [ - + ]: 2454 : if (cur->sk_attno > tupnatts)
1880 : : {
1881 : : /* Set this just like _bt_tuple_before_array_skeys */
8 pg@bowt.ie 1882 [ # # ]:UNC 0 : Assert(sktrig < ikey);
1883 : 0 : so->scanBehind = true;
1884 : : }
1885 : : }
1886 : :
1887 : : /*
1888 : : * Handle a required non-array scan key that the initial call to
1889 : : * _bt_check_compare indicated triggered array advancement, if any.
1890 : : *
1891 : : * The non-array scan key's strategy will be <, <=, or = during a
1892 : : * forwards scan (or any one of =, >=, or > during a backwards scan).
1893 : : * It follows that the corresponding tuple attribute's value must now
1894 : : * be either > or >= the scan key value (for backwards scans it must
1895 : : * be either < or <= that value).
1896 : : *
1897 : : * If this is a required equality strategy scan key, this is just an
1898 : : * optimization; _bt_tuple_before_array_skeys already confirmed that
1899 : : * this scan key places us ahead of caller's tuple. There's no need
1900 : : * to repeat that work now. (The same underlying principle also gets
1901 : : * applied by the cur_elem_trig optimization used to speed up searches
1902 : : * for the next array element.)
1903 : : *
1904 : : * If this is a required inequality strategy scan key, we _must_ rely
1905 : : * on _bt_check_compare like this; we aren't capable of directly
1906 : : * evaluating required inequality strategy scan keys here, on our own.
1907 : : */
8 pg@bowt.ie 1908 [ + + + + ]:GNC 3214 : if (ikey == sktrig && !array)
1909 : : {
1910 [ + - + - : 31 : Assert(sktrig_required && required && all_required_satisfied);
- + ]
1911 : :
1912 : : /* Use "beyond end" advancement. See below for an explanation. */
1913 : 31 : beyond_end_advance = true;
1914 : 31 : all_satisfied = all_required_satisfied = false;
1915 : :
1916 : : /*
1917 : : * Set a flag that remembers that this was an inequality required
1918 : : * in the opposite scan direction only, that nevertheless
1919 : : * triggered the call here.
1920 : : *
1921 : : * This only happens when an inequality operator (which must be
1922 : : * strict) encounters a group of NULLs that indicate the end of
1923 : : * non-NULL values for tuples in the current scan direction.
1924 : : */
1925 [ - + ]: 31 : if (unlikely(required_opposite_direction_only))
8 pg@bowt.ie 1926 :UNC 0 : oppodir_inequality_sktrig = true;
1927 : :
8 pg@bowt.ie 1928 :GNC 31 : continue;
1929 : : }
1930 : :
1931 : : /*
1932 : : * Nothing more for us to do with an inequality strategy scan key that
1933 : : * wasn't the one that _bt_check_compare stopped on, though.
1934 : : *
1935 : : * Note: if our later call to _bt_check_compare (to recheck caller's
1936 : : * tuple) sets continuescan=false due to finding this same inequality
1937 : : * unsatisfied (possible when it's required in the scan direction),
1938 : : * we'll deal with it via a recursive "second pass" call.
1939 : : */
1940 [ + + ]: 3183 : else if (cur->sk_strategy != BTEqualStrategyNumber)
1941 : 9 : continue;
1942 : :
1943 : : /*
1944 : : * Nothing for us to do with an equality strategy scan key that isn't
1945 : : * marked required, either -- unless it's a non-required array
1946 : : */
1947 [ + + + + ]: 3174 : else if (!required && !array)
1948 : 637 : continue;
1949 : :
1950 : : /*
1951 : : * Here we perform steps for all array scan keys after a required
1952 : : * array scan key whose binary search triggered "beyond end of array
1953 : : * element" array advancement due to encountering a tuple attribute
1954 : : * value > the closest matching array key (or < for backwards scans).
1955 : : */
1956 [ + + ]: 2537 : if (beyond_end_advance)
1957 : 132 : {
1958 : : int final_elem_dir;
1959 : :
1960 [ + - + + ]: 132 : if (ScanDirectionIsBackward(dir) || !array)
1961 : 48 : final_elem_dir = 0;
1962 : : else
1963 : 84 : final_elem_dir = array->num_elems - 1;
1964 : :
1965 [ + + + + ]: 132 : if (array && array->cur_elem != final_elem_dir)
1966 : : {
1967 : 24 : array->cur_elem = final_elem_dir;
1968 : 24 : cur->sk_argument = array->elem_values[final_elem_dir];
1969 : : }
1970 : :
1971 : 132 : continue;
1972 : : }
1973 : :
1974 : : /*
1975 : : * Here we perform steps for all array scan keys after a required
1976 : : * array scan key whose tuple attribute was < the closest matching
1977 : : * array key when we dealt with it (or > for backwards scans).
1978 : : *
1979 : : * This earlier required array key already puts us ahead of caller's
1980 : : * tuple in the key space (for the current scan direction). We must
1981 : : * make sure that subsequent lower-order array keys do not put us too
1982 : : * far ahead (ahead of tuples that have yet to be seen by our caller).
1983 : : * For example, when a tuple "(a, b) = (42, 5)" advances the array
1984 : : * keys on "a" from 40 to 45, we must also set "b" to whatever the
1985 : : * first array element for "b" is. It would be wrong to allow "b" to
1986 : : * be set based on the tuple value.
1987 : : *
1988 : : * Perform the same steps with truncated high key attributes. You can
1989 : : * think of this as a "binary search" for the element closest to the
1990 : : * value -inf. Again, the arrays must never get ahead of the scan.
1991 : : */
1992 [ + + - + ]: 2405 : if (!all_required_satisfied || cur->sk_attno > tupnatts)
1993 : 205 : {
1994 : : int first_elem_dir;
1995 : :
1996 [ - + - - ]: 205 : if (ScanDirectionIsForward(dir) || !array)
1997 : 205 : first_elem_dir = 0;
1998 : : else
8 pg@bowt.ie 1999 :UNC 0 : first_elem_dir = array->num_elems - 1;
2000 : :
8 pg@bowt.ie 2001 [ + + + + ]:GNC 205 : if (array && array->cur_elem != first_elem_dir)
2002 : : {
2003 : 90 : array->cur_elem = first_elem_dir;
2004 : 90 : cur->sk_argument = array->elem_values[first_elem_dir];
2005 : : }
2006 : :
2007 : 205 : continue;
2008 : : }
2009 : :
2010 : : /*
2011 : : * Search in scankey's array for the corresponding tuple attribute
2012 : : * value from caller's tuple
2013 : : */
2014 : 2200 : tupdatum = index_getattr(tuple, cur->sk_attno, tupdesc, &tupnull);
2015 : :
2016 [ + + ]: 2200 : if (array)
2017 : : {
2018 [ + + + + ]: 2018 : bool cur_elem_trig = (sktrig_required && ikey == sktrig);
2019 : :
2020 : : /*
2021 : : * Binary search for closest match that's available from the array
2022 : : */
2023 : 2018 : set_elem = _bt_binsrch_array_skey(&so->orderProcs[ikey],
2024 : : cur_elem_trig, dir,
2025 : : tupdatum, tupnull, array, cur,
2026 : : &result);
2027 : :
2028 [ + - - + ]: 2018 : Assert(set_elem >= 0 && set_elem < array->num_elems);
2029 : : }
2030 : : else
2031 : : {
2032 [ + - - + ]: 182 : Assert(sktrig_required && required);
2033 : :
2034 : : /*
2035 : : * This is a required non-array equality strategy scan key, which
2036 : : * we'll treat as a degenerate single element array.
2037 : : *
2038 : : * This scan key's imaginary "array" can't really advance, but it
2039 : : * can still roll over like any other array. (Actually, this is
2040 : : * no different to real single value arrays, which never advance
2041 : : * without rolling over -- they can never truly advance, either.)
2042 : : */
2043 : 182 : result = _bt_compare_array_skey(&so->orderProcs[ikey],
2044 : : tupdatum, tupnull,
2045 : : cur->sk_argument, cur);
2046 : : }
2047 : :
2048 : : /*
2049 : : * Consider "beyond end of array element" array advancement.
2050 : : *
2051 : : * When the tuple attribute value is > the closest matching array key
2052 : : * (or < in the backwards scan case), we need to ratchet this array
2053 : : * forward (backward) by one increment, so that caller's tuple ends up
2054 : : * being < final array value instead (or > final array value instead).
2055 : : * This process has to work for all of the arrays, not just this one:
2056 : : * it must "carry" to higher-order arrays when the set_elem that we
2057 : : * just found happens to be the final one for the scan's direction.
2058 : : * Incrementing (decrementing) set_elem itself isn't good enough.
2059 : : *
2060 : : * Our approach is to provisionally use set_elem as if it was an exact
2061 : : * match now, then set each later/less significant array to whatever
2062 : : * its final element is. Once outside the loop we'll then "increment
2063 : : * this array's set_elem" by calling _bt_advance_array_keys_increment.
2064 : : * That way the process rolls over to higher order arrays as needed.
2065 : : *
2066 : : * Under this scheme any required arrays only ever ratchet forwards
2067 : : * (or backwards), and always do so to the maximum possible extent
2068 : : * that we can know will be safe without seeing the scan's next tuple.
2069 : : * We don't need any special handling for required scan keys that lack
2070 : : * a real array to advance, nor for redundant scan keys that couldn't
2071 : : * be eliminated by _bt_preprocess_keys. It won't matter if some of
2072 : : * our "true" array scan keys (or even all of them) are non-required.
2073 : : */
2074 [ + + + + ]: 2200 : if (required &&
2075 [ + + + + ]: 2080 : ((ScanDirectionIsForward(dir) && result > 0) ||
2076 [ + + ]: 12 : (ScanDirectionIsBackward(dir) && result < 0)))
2077 : 314 : beyond_end_advance = true;
2078 : :
2079 [ + - - + ]: 2200 : Assert(all_required_satisfied && all_satisfied);
2080 [ + + ]: 2200 : if (result != 0)
2081 : : {
2082 : : /*
2083 : : * Track whether caller's tuple satisfies our new post-advancement
2084 : : * qual, for required scan keys, as well as for the entire set of
2085 : : * interesting scan keys (all required scan keys plus non-required
2086 : : * array scan keys are considered interesting.)
2087 : : */
2088 : 1001 : all_satisfied = false;
2089 [ + + ]: 1001 : if (required)
2090 : 893 : all_required_satisfied = false;
2091 : : else
2092 : : {
2093 : : /*
2094 : : * There's no need to advance the arrays using the best
2095 : : * available match for a non-required array. Give up now.
2096 : : * (Though note that sktrig_required calls still have to do
2097 : : * all the usual post-advancement steps, including the recheck
2098 : : * call to _bt_check_compare.)
2099 : : */
2100 : 108 : break;
2101 : : }
2102 : : }
2103 : :
2104 : : /* Advance array keys, even when set_elem isn't an exact match */
2105 [ + + + + ]: 2092 : if (array && array->cur_elem != set_elem)
2106 : : {
2107 : 1596 : array->cur_elem = set_elem;
2108 : 1596 : cur->sk_argument = array->elem_values[set_elem];
2109 : : }
2110 : : }
2111 : :
2112 : : /*
2113 : : * Advance the array keys incrementally whenever "beyond end of array
2114 : : * element" array advancement happens, so that advancement will carry to
2115 : : * higher-order arrays (might exhaust all the scan's arrays instead, which
2116 : : * ends the top-level scan).
2117 : : */
2118 [ + + + + ]: 1941 : if (beyond_end_advance && !_bt_advance_array_keys_increment(scan, dir))
2119 : 329 : goto end_toplevel_scan;
2120 : :
2121 [ - + ]: 1612 : Assert(_bt_verify_keys_with_arraykeys(scan));
2122 : :
2123 : : /*
2124 : : * Does tuple now satisfy our new qual? Recheck with _bt_check_compare.
2125 : : *
2126 : : * Calls triggered by an unsatisfied required scan key, whose tuple now
2127 : : * satisfies all required scan keys, but not all nonrequired array keys,
2128 : : * will still require a recheck call to _bt_check_compare. They'll still
2129 : : * need its "second pass" handling of required inequality scan keys.
2130 : : * (Might have missed a still-unsatisfied required inequality scan key
2131 : : * that caller didn't detect as the sktrig scan key during its initial
2132 : : * _bt_check_compare call that used the old/original qual.)
2133 : : *
2134 : : * Calls triggered by an unsatisfied nonrequired array scan key never need
2135 : : * "second pass" handling of required inequalities (nor any other handling
2136 : : * of any required scan key). All that matters is whether caller's tuple
2137 : : * satisfies the new qual, so it's safe to just skip the _bt_check_compare
2138 : : * recheck when we've already determined that it can only return 'false'.
2139 : : */
2140 [ + + + + ]: 1612 : if ((sktrig_required && all_required_satisfied) ||
2141 [ + + + + ]: 715 : (!sktrig_required && all_satisfied))
2142 : : {
2143 : 909 : int nsktrig = sktrig + 1;
2144 : : bool continuescan;
2145 : :
2146 [ - + ]: 909 : Assert(all_required_satisfied);
2147 : :
2148 : : /* Recheck _bt_check_compare on behalf of caller */
2149 [ + + ]: 909 : if (_bt_check_compare(scan, dir, tuple, tupnatts, tupdesc,
2150 : : false, false, false,
2151 : 906 : &continuescan, &nsktrig) &&
2152 [ + - ]: 906 : !so->scanBehind)
2153 : : {
2154 : : /* This tuple satisfies the new qual */
2155 [ + - - + ]: 906 : Assert(all_satisfied && continuescan);
2156 : :
2157 [ + + ]: 906 : if (pstate)
2158 : 894 : pstate->continuescan = true;
2159 : :
2160 : 906 : return true;
2161 : : }
2162 : :
2163 : : /*
2164 : : * Consider "second pass" handling of required inequalities.
2165 : : *
2166 : : * It's possible that our _bt_check_compare call indicated that the
2167 : : * scan should end due to some unsatisfied inequality that wasn't
2168 : : * initially recognized as such by us. Handle this by calling
2169 : : * ourselves recursively, this time indicating that the trigger is the
2170 : : * inequality that we missed first time around (and using a set of
2171 : : * required array/equality keys that are now exact matches for tuple).
2172 : : *
2173 : : * We make a strong, general guarantee that every _bt_checkkeys call
2174 : : * here will advance the array keys to the maximum possible extent
2175 : : * that we can know to be safe based on caller's tuple alone. If we
2176 : : * didn't perform this step, then that guarantee wouldn't quite hold.
2177 : : */
2178 [ - + ]: 3 : if (unlikely(!continuescan))
2179 : : {
2180 : : bool satisfied PG_USED_FOR_ASSERTS_ONLY;
2181 : :
8 pg@bowt.ie 2182 [ # # ]:UNC 0 : Assert(sktrig_required);
2183 [ # # ]: 0 : Assert(so->keyData[nsktrig].sk_strategy != BTEqualStrategyNumber);
2184 : :
2185 : : /*
2186 : : * The tuple must use "beyond end" advancement during the
2187 : : * recursive call, so we cannot possibly end up back here when
2188 : : * recursing. We'll consume a small, fixed amount of stack space.
2189 : : */
2190 [ # # ]: 0 : Assert(!beyond_end_advance);
2191 : :
2192 : : /* Advance the array keys a second time using same tuple */
2193 : 0 : satisfied = _bt_advance_array_keys(scan, pstate, tuple, tupnatts,
2194 : : tupdesc, nsktrig, true);
2195 : :
2196 : : /* This tuple doesn't satisfy the inequality */
2197 [ # # ]: 0 : Assert(!satisfied);
2198 : 0 : return false;
2199 : : }
2200 : :
2201 : : /*
2202 : : * Some non-required scan key (from new qual) still not satisfied.
2203 : : *
2204 : : * All scan keys required in the current scan direction must still be
2205 : : * satisfied, though, so we can trust all_required_satisfied below.
2206 : : */
2207 : : }
2208 : :
2209 : : /*
2210 : : * When we were called just to deal with "advancing" non-required arrays,
2211 : : * this is as far as we can go (cannot stop the scan for these callers)
2212 : : */
8 pg@bowt.ie 2213 [ + + ]:GNC 706 : if (!sktrig_required)
2214 : : {
2215 : : /* Caller's tuple doesn't match any qual */
2216 : 108 : return false;
2217 : : }
2218 : :
2219 : : /*
2220 : : * Postcondition array state assertion (for still-unsatisfied tuples).
2221 : : *
2222 : : * By here we have established that the scan's required arrays (scan must
2223 : : * have at least one required array) advanced, without becoming exhausted.
2224 : : *
2225 : : * Caller's tuple is now < the newly advanced array keys (or > when this
2226 : : * is a backwards scan), except in the case where we only got this far due
2227 : : * to an unsatisfied non-required scan key. Verify that with an assert.
2228 : : *
2229 : : * Note: we don't just quit at this point when all required scan keys were
2230 : : * found to be satisfied because we need to consider edge-cases involving
2231 : : * scan keys required in the opposite direction only; those aren't tracked
2232 : : * by all_required_satisfied. (Actually, oppodir_inequality_sktrig trigger
2233 : : * scan keys are tracked by all_required_satisfied, since it's convenient
2234 : : * for _bt_check_compare to behave as if they are required in the current
2235 : : * scan direction to deal with NULLs. We'll account for that separately.)
2236 : : */
2237 [ - + ]: 598 : Assert(_bt_tuple_before_array_skeys(scan, dir, tuple, tupdesc, tupnatts,
2238 : : false, 0, NULL) ==
2239 : : !all_required_satisfied);
2240 : :
2241 : : /*
2242 : : * We generally permit primitive index scans to continue onto the next
2243 : : * sibling page when the page's finaltup satisfies all required scan keys
2244 : : * at the point where we're between pages.
2245 : : *
2246 : : * If caller's tuple is also the page's finaltup, and we see that required
2247 : : * scan keys still aren't satisfied, start a new primitive index scan.
2248 : : */
2249 [ + + - + ]: 598 : if (!all_required_satisfied && pstate->finaltup == tuple)
8 pg@bowt.ie 2250 :UNC 0 : goto new_prim_scan;
2251 : :
2252 : : /*
2253 : : * Proactively check finaltup (don't wait until finaltup is reached by the
2254 : : * scan) when it might well turn out to not be satisfied later on.
2255 : : *
2256 : : * Note: if so->scanBehind hasn't already been set for finaltup by us,
2257 : : * it'll be set during this call to _bt_tuple_before_array_skeys. Either
2258 : : * way, it'll be set correctly (for the whole page) after this point.
2259 : : */
8 pg@bowt.ie 2260 [ + + + + :GNC 1141 : if (!all_required_satisfied && pstate->finaltup &&
+ + ]
2261 [ + + ]: 1086 : _bt_tuple_before_array_skeys(scan, dir, pstate->finaltup, tupdesc,
2262 : 1086 : BTreeTupleGetNAtts(pstate->finaltup, rel),
2263 : : false, 0, &so->scanBehind))
2264 : 285 : goto new_prim_scan;
2265 : :
2266 : : /*
2267 : : * When we encounter a truncated finaltup high key attribute, we're
2268 : : * optimistic about the chances of its corresponding required scan key
2269 : : * being satisfied when we go on to check it against tuples from this
2270 : : * page's right sibling leaf page. We consider truncated attributes to be
2271 : : * satisfied by required scan keys, which allows the primitive index scan
2272 : : * to continue to the next leaf page. We must set so->scanBehind to true
2273 : : * to remember that the last page's finaltup had "satisfied" required scan
2274 : : * keys for one or more truncated attribute values (scan keys required in
2275 : : * _either_ scan direction).
2276 : : *
2277 : : * There is a chance that _bt_checkkeys (which checks so->scanBehind) will
2278 : : * find that even the sibling leaf page's finaltup is < the new array
2279 : : * keys. When that happens, our optimistic policy will have incurred a
2280 : : * single extra leaf page access that could have been avoided.
2281 : : *
2282 : : * A pessimistic policy would give backward scans a gratuitous advantage
2283 : : * over forward scans. We'd punish forward scans for applying more
2284 : : * accurate information from the high key, rather than just using the
2285 : : * final non-pivot tuple as finaltup, in the style of backward scans.
2286 : : * Being pessimistic would also give some scans with non-required arrays a
2287 : : * perverse advantage over similar scans that use required arrays instead.
2288 : : *
2289 : : * You can think of this as a speculative bet on what the scan is likely
2290 : : * to find on the next page. It's not much of a gamble, though, since the
2291 : : * untruncated prefix of attributes must strictly satisfy the new qual
2292 : : * (though it's okay if any non-required scan keys fail to be satisfied).
2293 : : */
2294 [ + + - + ]: 313 : if (so->scanBehind && has_required_opposite_direction_only)
2295 : : {
2296 : : /*
2297 : : * However, we avoid this behavior whenever the scan involves a scan
2298 : : * key required in the opposite direction to the scan only, along with
2299 : : * a finaltup with at least one truncated attribute that's associated
2300 : : * with a scan key marked required (required in either direction).
2301 : : *
2302 : : * _bt_check_compare simply won't stop the scan for a scan key that's
2303 : : * marked required in the opposite scan direction only. That leaves
2304 : : * us without any reliable way of reconsidering any opposite-direction
2305 : : * inequalities if it turns out that starting a new primitive index
2306 : : * scan will allow _bt_first to skip ahead by a great many leaf pages
2307 : : * (see next section for details of how that works).
2308 : : */
8 pg@bowt.ie 2309 :UNC 0 : goto new_prim_scan;
2310 : : }
2311 : :
2312 : : /*
2313 : : * Handle inequalities marked required in the opposite scan direction.
2314 : : * They can also signal that we should start a new primitive index scan.
2315 : : *
2316 : : * It's possible that the scan is now positioned where "matching" tuples
2317 : : * begin, and that caller's tuple satisfies all scan keys required in the
2318 : : * current scan direction. But if caller's tuple still doesn't satisfy
2319 : : * other scan keys that are required in the opposite scan direction only
2320 : : * (e.g., a required >= strategy scan key when scan direction is forward),
2321 : : * it's still possible that there are many leaf pages before the page that
2322 : : * _bt_first could skip straight to. Groveling through all those pages
2323 : : * will always give correct answers, but it can be very inefficient. We
2324 : : * must avoid needlessly scanning extra pages.
2325 : : *
2326 : : * Separately, it's possible that _bt_check_compare set continuescan=false
2327 : : * for a scan key that's required in the opposite direction only. This is
2328 : : * a special case, that happens only when _bt_check_compare sees that the
2329 : : * inequality encountered a NULL value. This signals the end of non-NULL
2330 : : * values in the current scan direction, which is reason enough to end the
2331 : : * (primitive) scan. If this happens at the start of a large group of
2332 : : * NULL values, then we shouldn't expect to be called again until after
2333 : : * the scan has already read indefinitely-many leaf pages full of tuples
2334 : : * with NULL suffix values. We need a separate test for this case so that
2335 : : * we don't miss our only opportunity to skip over such a group of pages.
2336 : : * (_bt_first is expected to skip over the group of NULLs by applying a
2337 : : * similar "deduce NOT NULL" rule, where it finishes its insertion scan
2338 : : * key by consing up an explicit SK_SEARCHNOTNULL key.)
2339 : : *
2340 : : * Apply a test against finaltup to detect and recover from these problem:
2341 : : * if even finaltup doesn't satisfy such an inequality, we just skip by
2342 : : * starting a new primitive index scan. When we skip, we know for sure
2343 : : * that all of the tuples on the current page following caller's tuple are
2344 : : * also before the _bt_first-wise start of tuples for our new qual. That
2345 : : * at least suggests many more skippable pages beyond the current page.
2346 : : */
8 pg@bowt.ie 2347 [ + + - + :GNC 313 : if (has_required_opposite_direction_only && pstate->finaltup &&
- - ]
8 pg@bowt.ie 2348 [ # # ]:UNC 0 : (all_required_satisfied || oppodir_inequality_sktrig))
2349 : : {
2350 [ # # ]: 0 : int nfinaltupatts = BTreeTupleGetNAtts(pstate->finaltup, rel);
2351 : : ScanDirection flipped;
2352 : : bool continuescanflip;
2353 : : int opsktrig;
2354 : :
2355 : : /*
2356 : : * We're checking finaltup (which is usually not caller's tuple), so
2357 : : * cannot reuse work from caller's earlier _bt_check_compare call.
2358 : : *
2359 : : * Flip the scan direction when calling _bt_check_compare this time,
2360 : : * so that it will set continuescanflip=false when it encounters an
2361 : : * inequality required in the opposite scan direction.
2362 : : */
2363 [ # # ]: 0 : Assert(!so->scanBehind);
2364 : 0 : opsktrig = 0;
2365 : 0 : flipped = -dir;
2366 : 0 : _bt_check_compare(scan, flipped,
2367 : : pstate->finaltup, nfinaltupatts, tupdesc,
2368 : : false, false, false,
2369 : : &continuescanflip, &opsktrig);
2370 : :
2371 : : /*
2372 : : * If we ended up here due to the all_required_satisfied criteria,
2373 : : * test opsktrig in a way that ensures that finaltup contains the same
2374 : : * prefix of key columns as caller's tuple (a prefix that satisfies
2375 : : * earlier required-in-current-direction scan keys).
2376 : : *
2377 : : * If we ended up here due to the oppodir_inequality_sktrig criteria,
2378 : : * test opsktrig in a way that ensures that the same scan key that our
2379 : : * caller found to be unsatisfied (by the scan's tuple) was also the
2380 : : * one unsatisfied just now (by finaltup). That way we'll only start
2381 : : * a new primitive scan when we're sure that both tuples _don't_ share
2382 : : * the same prefix of satisfied equality-constrained attribute values,
2383 : : * and that finaltup has a non-NULL attribute value indicated by the
2384 : : * unsatisfied scan key at offset opsktrig/sktrig. (This depends on
2385 : : * _bt_check_compare not caring about the direction that inequalities
2386 : : * are required in whenever NULL attribute values are unsatisfied. It
2387 : : * only cares about the scan direction, and its relationship to
2388 : : * whether NULLs are stored first or last relative to non-NULLs.)
2389 : : */
2390 [ # # ]: 0 : Assert(all_required_satisfied != oppodir_inequality_sktrig);
2391 [ # # # # : 0 : if (unlikely(!continuescanflip &&
# # # # #
# # # ]
2392 : : ((all_required_satisfied && opsktrig > sktrig) ||
2393 : : (oppodir_inequality_sktrig && opsktrig >= sktrig))))
2394 : : {
2395 [ # # ]: 0 : Assert(so->keyData[opsktrig].sk_strategy != BTEqualStrategyNumber);
2396 : :
2397 : : /*
2398 : : * Make sure that any non-required arrays are set to the first
2399 : : * array element for the current scan direction
2400 : : */
2401 : 0 : _bt_rewind_nonrequired_arrays(scan, dir);
2402 : :
2403 : 0 : goto new_prim_scan;
2404 : : }
2405 : : }
2406 : :
2407 : : /*
2408 : : * Stick with the ongoing primitive index scan for now.
2409 : : *
2410 : : * It's possible that later tuples will also turn out to have values that
2411 : : * are still < the now-current array keys (or > the current array keys).
2412 : : * Our caller will handle this by performing what amounts to a linear
2413 : : * search of the page, implemented by calling _bt_check_compare and then
2414 : : * _bt_tuple_before_array_skeys for each tuple.
2415 : : *
2416 : : * This approach has various advantages over a binary search of the page.
2417 : : * Repeated binary searches of the page (one binary search for every array
2418 : : * advancement) won't outperform a continuous linear search. While there
2419 : : * are workloads that a naive linear search won't handle well, our caller
2420 : : * has a "look ahead" fallback mechanism to deal with that problem.
2421 : : */
8 pg@bowt.ie 2422 :GNC 313 : pstate->continuescan = true; /* Override _bt_check_compare */
2423 : 313 : so->needPrimScan = false; /* _bt_readpage has more tuples to check */
2424 : :
2425 [ + + ]: 313 : if (so->scanBehind)
2426 : : {
2427 : : /* Optimization: skip by setting "look ahead" mechanism's offnum */
2428 [ - + ]: 3 : Assert(ScanDirectionIsForward(dir));
2429 : 3 : pstate->skip = pstate->maxoff + 1;
2430 : : }
2431 : :
2432 : : /* Caller's tuple doesn't match the new qual */
2433 : 313 : return false;
2434 : :
2435 : 285 : new_prim_scan:
2436 : :
2437 : : /*
2438 : : * End this primitive index scan, but schedule another.
2439 : : *
2440 : : * Note: If the scan direction happens to change, this scheduled primitive
2441 : : * index scan won't go ahead after all.
2442 : : */
2443 : 285 : pstate->continuescan = false; /* Tell _bt_readpage we're done... */
2444 : 285 : so->needPrimScan = true; /* ...but call _bt_first again */
2445 : :
2446 [ + + ]: 285 : if (scan->parallel_scan)
2447 : 18 : _bt_parallel_primscan_schedule(scan, pstate->prev_scan_page);
2448 : :
2449 : : /* Caller's tuple doesn't match the new qual */
2450 : 285 : return false;
2451 : :
2452 : 329 : end_toplevel_scan:
2453 : :
2454 : : /*
2455 : : * End the current primitive index scan, but don't schedule another.
2456 : : *
2457 : : * This ends the entire top-level scan in the current scan direction.
2458 : : *
2459 : : * Note: The scan's arrays (including any non-required arrays) are now in
2460 : : * their final positions for the current scan direction. If the scan
2461 : : * direction happens to change, then the arrays will already be in their
2462 : : * first positions for what will then be the current scan direction.
2463 : : */
2464 : 329 : pstate->continuescan = false; /* Tell _bt_readpage we're done... */
2465 : 329 : so->needPrimScan = false; /* ...don't call _bt_first again, though */
2466 : :
2467 : : /* Caller's tuple doesn't match any qual */
2468 : 329 : return false;
2469 : : }
2470 : :
2471 : : /*
2472 : : * _bt_preprocess_keys() -- Preprocess scan keys
2473 : : *
2474 : : * The given search-type keys (taken from scan->keyData[])
2475 : : * are copied to so->keyData[] with possible transformation.
2476 : : * scan->numberOfKeys is the number of input keys, so->numberOfKeys gets
2477 : : * the number of output keys (possibly less, never greater).
2478 : : *
2479 : : * The output keys are marked with additional sk_flags bits beyond the
2480 : : * system-standard bits supplied by the caller. The DESC and NULLS_FIRST
2481 : : * indoption bits for the relevant index attribute are copied into the flags.
2482 : : * Also, for a DESC column, we commute (flip) all the sk_strategy numbers
2483 : : * so that the index sorts in the desired direction.
2484 : : *
2485 : : * One key purpose of this routine is to discover which scan keys must be
2486 : : * satisfied to continue the scan. It also attempts to eliminate redundant
2487 : : * keys and detect contradictory keys. (If the index opfamily provides
2488 : : * incomplete sets of cross-type operators, we may fail to detect redundant
2489 : : * or contradictory keys, but we can survive that.)
2490 : : *
2491 : : * The output keys must be sorted by index attribute. Presently we expect
2492 : : * (but verify) that the input keys are already so sorted --- this is done
2493 : : * by match_clauses_to_index() in indxpath.c. Some reordering of the keys
2494 : : * within each attribute may be done as a byproduct of the processing here.
2495 : : * That process must leave array scan keys (within an attribute) in the same
2496 : : * order as corresponding entries from the scan's BTArrayKeyInfo array info.
2497 : : *
2498 : : * The output keys are marked with flags SK_BT_REQFWD and/or SK_BT_REQBKWD
2499 : : * if they must be satisfied in order to continue the scan forward or backward
2500 : : * respectively. _bt_checkkeys uses these flags. For example, if the quals
2501 : : * are "x = 1 AND y < 4 AND z < 5", then _bt_checkkeys will reject a tuple
2502 : : * (1,2,7), but we must continue the scan in case there are tuples (1,3,z).
2503 : : * But once we reach tuples like (1,4,z) we can stop scanning because no
2504 : : * later tuples could match. This is reflected by marking the x and y keys,
2505 : : * but not the z key, with SK_BT_REQFWD. In general, the keys for leading
2506 : : * attributes with "=" keys are marked both SK_BT_REQFWD and SK_BT_REQBKWD.
2507 : : * For the first attribute without an "=" key, any "<" and "<=" keys are
2508 : : * marked SK_BT_REQFWD while any ">" and ">=" keys are marked SK_BT_REQBKWD.
2509 : : * This can be seen to be correct by considering the above example. Note
2510 : : * in particular that if there are no keys for a given attribute, the keys for
2511 : : * subsequent attributes can never be required; for instance "WHERE y = 4"
2512 : : * requires a full-index scan.
2513 : : *
2514 : : * If possible, redundant keys are eliminated: we keep only the tightest
2515 : : * >/>= bound and the tightest </<= bound, and if there's an = key then
2516 : : * that's the only one returned. (So, we return either a single = key,
2517 : : * or one or two boundary-condition keys for each attr.) However, if we
2518 : : * cannot compare two keys for lack of a suitable cross-type operator,
2519 : : * we cannot eliminate either. If there are two such keys of the same
2520 : : * operator strategy, the second one is just pushed into the output array
2521 : : * without further processing here. We may also emit both >/>= or both
2522 : : * </<= keys if we can't compare them. The logic about required keys still
2523 : : * works if we don't eliminate redundant keys.
2524 : : *
2525 : : * Note that one reason we need direction-sensitive required-key flags is
2526 : : * precisely that we may not be able to eliminate redundant keys. Suppose
2527 : : * we have "x > 4::int AND x > 10::bigint", and we are unable to determine
2528 : : * which key is more restrictive for lack of a suitable cross-type operator.
2529 : : * _bt_first will arbitrarily pick one of the keys to do the initial
2530 : : * positioning with. If it picks x > 4, then the x > 10 condition will fail
2531 : : * until we reach index entries > 10; but we can't stop the scan just because
2532 : : * x > 10 is failing. On the other hand, if we are scanning backwards, then
2533 : : * failure of either key is indeed enough to stop the scan. (In general, when
2534 : : * inequality keys are present, the initial-positioning code only promises to
2535 : : * position before the first possible match, not exactly at the first match,
2536 : : * for a forward scan; or after the last match for a backward scan.)
2537 : : *
2538 : : * As a byproduct of this work, we can detect contradictory quals such
2539 : : * as "x = 1 AND x > 2". If we see that, we return so->qual_ok = false,
2540 : : * indicating the scan need not be run at all since no tuples can match.
2541 : : * (In this case we do not bother completing the output key array!)
2542 : : * Again, missing cross-type operators might cause us to fail to prove the
2543 : : * quals contradictory when they really are, but the scan will work correctly.
2544 : : *
2545 : : * Row comparison keys are currently also treated without any smarts:
2546 : : * we just transfer them into the preprocessed array without any
2547 : : * editorialization. We can treat them the same as an ordinary inequality
2548 : : * comparison on the row's first index column, for the purposes of the logic
2549 : : * about required keys.
2550 : : *
2551 : : * Note: the reason we have to copy the preprocessed scan keys into private
2552 : : * storage is that we are modifying the array based on comparisons of the
2553 : : * key argument values, which could change on a rescan. Therefore we can't
2554 : : * overwrite the source data.
2555 : : */
2556 : : void
7459 tgl@sss.pgh.pa.us 2557 :CBC 6420890 : _bt_preprocess_keys(IndexScanDesc scan)
2558 : : {
7996 2559 : 6420890 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
7459 2560 : 6420890 : int numberOfKeys = scan->numberOfKeys;
6305 2561 : 6420890 : int16 *indoption = scan->indexRelation->rd_indoption;
2562 : : int new_numberOfKeys;
2563 : : int numberOfEqualCols;
2564 : : ScanKey inkeys;
2565 : : ScanKey outkeys;
2566 : : ScanKey cur;
2567 : : BTScanKeyPreproc xform[BTMaxStrategyNumber];
2568 : : bool test_result;
2569 : : int i,
2570 : : j;
2571 : : AttrNumber attno;
2572 : : ScanKey arrayKeyData;
8 pg@bowt.ie 2573 :GNC 6420890 : int *keyDataMap = NULL;
2574 : 6420890 : int arrayidx = 0;
2575 : :
2576 : : /*
2577 : : * We're called at the start of each primitive index scan during scans
2578 : : * that use equality array keys. We can just reuse the scan keys that
2579 : : * were output at the start of the scan's first primitive index scan.
2580 : : */
2581 [ + + ]: 6420890 : if (so->numberOfKeys > 0)
2582 : : {
2583 : : /*
2584 : : * An earlier call to _bt_advance_array_keys already set everything up
2585 : : * already. Just assert that the scan's existing output scan keys are
2586 : : * consistent with its current array elements.
2587 : : */
2588 [ - + ]: 285 : Assert(so->numArrayKeys);
2589 [ - + ]: 285 : Assert(_bt_verify_keys_with_arraykeys(scan));
2590 : 3242235 : return;
2591 : : }
2592 : :
2593 : : /* initialize result variables */
8664 tgl@sss.pgh.pa.us 2594 :CBC 6420605 : so->qual_ok = true;
7459 2595 : 6420605 : so->numberOfKeys = 0;
2596 : :
9716 bruce@momjian.us 2597 [ + + ]: 6420605 : if (numberOfKeys < 1)
8664 tgl@sss.pgh.pa.us 2598 : 5362 : return; /* done if qual-less scan */
2599 : :
2600 : : /* If any keys are SK_SEARCHARRAY type, set up array-key info */
8 pg@bowt.ie 2601 :GNC 6415243 : arrayKeyData = _bt_preprocess_array_keys(scan);
2602 [ + + ]: 6415243 : if (!so->qual_ok)
2603 : : {
2604 : : /* unmatchable array, so give up */
2605 : 3 : return;
2606 : : }
2607 : :
2608 : : /*
2609 : : * Treat arrayKeyData[] (a partially preprocessed copy of scan->keyData[])
2610 : : * as our input if _bt_preprocess_array_keys just allocated it, else just
2611 : : * use scan->keyData[]
2612 : : */
2613 [ + + ]: 6415240 : if (arrayKeyData)
2614 : : {
2615 : 416 : inkeys = arrayKeyData;
2616 : :
2617 : : /* Also maintain keyDataMap for remapping so->orderProc[] later */
2618 : 416 : keyDataMap = MemoryContextAlloc(so->arrayContext,
2619 : : numberOfKeys * sizeof(int));
2620 : : }
2621 : : else
4564 tgl@sss.pgh.pa.us 2622 :CBC 6414824 : inkeys = scan->keyData;
2623 : :
7459 2624 : 6415240 : outkeys = so->keyData;
2625 : 6415240 : cur = &inkeys[0];
2626 : : /* we check that input keys are correctly ordered */
6880 2627 [ - + ]: 6415240 : if (cur->sk_attno < 1)
6880 tgl@sss.pgh.pa.us 2628 [ # # ]:UBC 0 : elog(ERROR, "btree index keys must be ordered by attribute");
2629 : :
2630 : : /* We can short-circuit most of the work if there's just one key */
9716 bruce@momjian.us 2631 [ + + ]:CBC 6415240 : if (numberOfKeys == 1)
2632 : : {
2633 : : /* Apply indoption to scankey (might change sk_strategy!) */
5215 tgl@sss.pgh.pa.us 2634 [ + + ]: 3236555 : if (!_bt_fix_scankey_strategy(cur, indoption))
2635 : 1354 : so->qual_ok = false;
6305 2636 : 3236555 : memcpy(outkeys, cur, sizeof(ScanKeyData));
7459 2637 : 3236555 : so->numberOfKeys = 1;
2638 : : /* We can mark the qual as required if it's for first index col */
6654 2639 [ + + ]: 3236555 : if (cur->sk_attno == 1)
2640 : 3235232 : _bt_mark_scankey_required(outkeys);
8 pg@bowt.ie 2641 [ + + ]:GNC 3236555 : if (arrayKeyData)
2642 : : {
2643 : : /*
2644 : : * Don't call _bt_preprocess_array_keys_final in this fast path
2645 : : * (we'll miss out on the single value array transformation, but
2646 : : * that's not nearly as important when there's only one scan key)
2647 : : */
2648 [ - + ]: 211 : Assert(cur->sk_flags & SK_SEARCHARRAY);
2649 [ + + + - : 211 : Assert(cur->sk_strategy != BTEqualStrategyNumber ||
- + ]
2650 : : (so->arrayKeys[0].scan_key == 0 &&
2651 : : OidIsValid(so->orderProcs[0].fn_oid)));
2652 : : }
2653 : :
9716 bruce@momjian.us 2654 :CBC 3236555 : return;
2655 : : }
2656 : :
2657 : : /*
2658 : : * Otherwise, do the full set of pushups.
2659 : : */
8664 tgl@sss.pgh.pa.us 2660 : 3178685 : new_numberOfKeys = 0;
7060 2661 : 3178685 : numberOfEqualCols = 0;
2662 : :
2663 : : /*
2664 : : * Initialize for processing of keys for attr 1.
2665 : : *
2666 : : * xform[i] points to the currently best scan key of strategy type i+1; it
2667 : : * is NULL if we haven't yet found such a key for this attr.
2668 : : */
8664 2669 : 3178685 : attno = 1;
7459 2670 : 3178685 : memset(xform, 0, sizeof(xform));
2671 : :
2672 : : /*
2673 : : * Loop iterates from 0 to numberOfKeys inclusive; we use the last pass to
2674 : : * handle after-last-key processing. Actual exit from the loop is at the
2675 : : * "break" statement below.
2676 : : */
8424 bruce@momjian.us 2677 : 10300380 : for (i = 0;; cur++, i++)
2678 : : {
9716 2679 [ + + ]: 10300380 : if (i < numberOfKeys)
2680 : : {
2681 : : /* Apply indoption to scankey (might change sk_strategy!) */
5215 tgl@sss.pgh.pa.us 2682 [ + + ]: 7121707 : if (!_bt_fix_scankey_strategy(cur, indoption))
2683 : : {
2684 : : /* NULL can't be matched, so give up */
5215 tgl@sss.pgh.pa.us 2685 :GBC 9 : so->qual_ok = false;
2686 : 9 : return;
2687 : : }
2688 : : }
2689 : :
2690 : : /*
2691 : : * If we are at the end of the keys for a particular attr, finish up
2692 : : * processing and emit the cleaned-up keys.
2693 : : */
9716 bruce@momjian.us 2694 [ + + + + ]:CBC 10300371 : if (i == numberOfKeys || cur->sk_attno != attno)
2695 : : {
7060 tgl@sss.pgh.pa.us 2696 : 7120701 : int priorNumberOfEqualCols = numberOfEqualCols;
2697 : :
2698 : : /* check input keys are correctly ordered */
6880 2699 [ + + - + ]: 7120701 : if (i < numberOfKeys && cur->sk_attno < attno)
6880 tgl@sss.pgh.pa.us 2700 [ # # ]:UBC 0 : elog(ERROR, "btree index keys must be ordered by attribute");
2701 : :
2702 : : /*
2703 : : * If = has been specified, all other keys can be eliminated as
2704 : : * redundant. If we have a case like key = 1 AND key > 2, we can
2705 : : * set qual_ok to false and abandon further processing.
2706 : : *
2707 : : * We also have to deal with the case of "key IS NULL", which is
2708 : : * unsatisfiable in combination with any other index condition. By
2709 : : * the time we get here, that's been classified as an equality
2710 : : * check, and we've rejected any combination of it with a regular
2711 : : * equality condition; but not with other types of conditions.
2712 : : */
8 pg@bowt.ie 2713 [ + + ]:GNC 7120701 : if (xform[BTEqualStrategyNumber - 1].skey)
2714 : : {
2715 : 6474113 : ScanKey eq = xform[BTEqualStrategyNumber - 1].skey;
2716 : 6474113 : BTArrayKeyInfo *array = NULL;
2717 : 6474113 : FmgrInfo *orderproc = NULL;
2718 : :
2719 [ + + + + ]: 6474113 : if (arrayKeyData && (eq->sk_flags & SK_SEARCHARRAY))
2720 : : {
2721 : : int eq_in_ikey,
2722 : : eq_arrayidx;
2723 : :
2724 : 271 : eq_in_ikey = xform[BTEqualStrategyNumber - 1].ikey;
2725 : 271 : eq_arrayidx = xform[BTEqualStrategyNumber - 1].arrayidx;
2726 : 271 : array = &so->arrayKeys[eq_arrayidx - 1];
2727 : 271 : orderproc = so->orderProcs + eq_in_ikey;
2728 : :
2729 [ - + ]: 271 : Assert(array->scan_key == eq_in_ikey);
2730 [ - + ]: 271 : Assert(OidIsValid(orderproc->fn_oid));
2731 : : }
2732 : :
9716 bruce@momjian.us 2733 [ + + ]:CBC 38844600 : for (j = BTMaxStrategyNumber; --j >= 0;)
2734 : : {
8 pg@bowt.ie 2735 :GNC 32370505 : ScanKey chk = xform[j].skey;
2736 : :
7459 tgl@sss.pgh.pa.us 2737 [ + + + + ]:CBC 32370505 : if (!chk || j == (BTEqualStrategyNumber - 1))
9716 bruce@momjian.us 2738 : 32370450 : continue;
2739 : :
4673 tgl@sss.pgh.pa.us 2740 [ + + ]: 55 : if (eq->sk_flags & SK_SEARCHNULL)
2741 : : {
2742 : : /* IS NULL is contradictory to anything else */
2743 : 12 : so->qual_ok = false;
2744 : 12 : return;
2745 : : }
2746 : :
6317 2747 [ + - ]: 43 : if (_bt_compare_scankey_args(scan, chk, eq, chk,
2748 : : array, orderproc,
2749 : : &test_result))
2750 : : {
2751 [ + + ]: 43 : if (!test_result)
2752 : : {
2753 : : /* keys proven mutually contradictory */
6317 tgl@sss.pgh.pa.us 2754 :GBC 6 : so->qual_ok = false;
2755 : 6 : return;
2756 : : }
2757 : : /* else discard the redundant non-equality key */
8 pg@bowt.ie 2758 [ + + - + ]:GNC 37 : Assert(!array || array->num_elems > 0);
2759 : 37 : xform[j].skey = NULL;
2760 : 37 : xform[j].ikey = -1;
2761 : : }
2762 : : /* else, cannot determine redundancy, keep both keys */
2763 : : }
2764 : : /* track number of attrs for which we have "=" keys */
7060 tgl@sss.pgh.pa.us 2765 :CBC 6474095 : numberOfEqualCols++;
2766 : : }
2767 : :
2768 : : /* try to keep only one of <, <= */
8 pg@bowt.ie 2769 [ + + ]:GNC 7120683 : if (xform[BTLessStrategyNumber - 1].skey
2770 [ + + ]: 981 : && xform[BTLessEqualStrategyNumber - 1].skey)
2771 : : {
2772 : 3 : ScanKey lt = xform[BTLessStrategyNumber - 1].skey;
2773 : 3 : ScanKey le = xform[BTLessEqualStrategyNumber - 1].skey;
2774 : :
2775 [ + - ]: 3 : if (_bt_compare_scankey_args(scan, le, lt, le, NULL, NULL,
2776 : : &test_result))
2777 : : {
6317 tgl@sss.pgh.pa.us 2778 [ + - ]:GBC 3 : if (test_result)
8 pg@bowt.ie 2779 :GNC 3 : xform[BTLessEqualStrategyNumber - 1].skey = NULL;
2780 : : else
8 pg@bowt.ie 2781 :UNC 0 : xform[BTLessStrategyNumber - 1].skey = NULL;
2782 : : }
2783 : : }
2784 : :
2785 : : /* try to keep only one of >, >= */
8 pg@bowt.ie 2786 [ + + ]:GNC 7120683 : if (xform[BTGreaterStrategyNumber - 1].skey
2787 [ + + ]: 644409 : && xform[BTGreaterEqualStrategyNumber - 1].skey)
2788 : : {
2789 : 3 : ScanKey gt = xform[BTGreaterStrategyNumber - 1].skey;
2790 : 3 : ScanKey ge = xform[BTGreaterEqualStrategyNumber - 1].skey;
2791 : :
2792 [ + - ]: 3 : if (_bt_compare_scankey_args(scan, ge, gt, ge, NULL, NULL,
2793 : : &test_result))
2794 : : {
6317 tgl@sss.pgh.pa.us 2795 [ - + ]:GBC 3 : if (test_result)
8 pg@bowt.ie 2796 :UNC 0 : xform[BTGreaterEqualStrategyNumber - 1].skey = NULL;
2797 : : else
8 pg@bowt.ie 2798 :GNC 3 : xform[BTGreaterStrategyNumber - 1].skey = NULL;
2799 : : }
2800 : : }
2801 : :
2802 : : /*
2803 : : * Emit the cleaned-up keys into the outkeys[] array, and then
2804 : : * mark them if they are required. They are required (possibly
2805 : : * only in one direction) if all attrs before this one had "=".
2806 : : */
9716 bruce@momjian.us 2807 [ + + ]:CBC 42724098 : for (j = BTMaxStrategyNumber; --j >= 0;)
2808 : : {
8 pg@bowt.ie 2809 [ + + ]:GNC 35603415 : if (xform[j].skey)
2810 : : {
6656 tgl@sss.pgh.pa.us 2811 :CBC 7121580 : ScanKey outkey = &outkeys[new_numberOfKeys++];
2812 : :
8 pg@bowt.ie 2813 :GNC 7121580 : memcpy(outkey, xform[j].skey, sizeof(ScanKeyData));
2814 [ + + ]: 7121580 : if (arrayKeyData)
2815 : 401 : keyDataMap[new_numberOfKeys - 1] = xform[j].ikey;
6656 tgl@sss.pgh.pa.us 2816 [ + + ]:CBC 7121580 : if (priorNumberOfEqualCols == attno - 1)
6654 2817 : 7121264 : _bt_mark_scankey_required(outkey);
2818 : : }
2819 : : }
2820 : :
2821 : : /*
2822 : : * Exit loop here if done.
2823 : : */
9716 bruce@momjian.us 2824 [ + + ]: 7120683 : if (i == numberOfKeys)
2825 : 3178655 : break;
2826 : :
2827 : : /* Re-initialize for new attno */
2828 : 3942028 : attno = cur->sk_attno;
7459 tgl@sss.pgh.pa.us 2829 : 3942028 : memset(xform, 0, sizeof(xform));
2830 : : }
2831 : :
2832 : : /* check strategy this key's operator corresponds to */
7462 2833 : 7121698 : j = cur->sk_strategy - 1;
2834 : :
2835 : : /* if row comparison, push it directly to the output array */
6317 2836 [ - + ]: 7121698 : if (cur->sk_flags & SK_ROW_HEADER)
7459 tgl@sss.pgh.pa.us 2837 :UBC 0 : {
6654 2838 : 0 : ScanKey outkey = &outkeys[new_numberOfKeys++];
2839 : :
2840 : 0 : memcpy(outkey, cur, sizeof(ScanKeyData));
8 pg@bowt.ie 2841 [ # # ]:UNC 0 : if (arrayKeyData)
2842 : 0 : keyDataMap[new_numberOfKeys - 1] = i;
6654 tgl@sss.pgh.pa.us 2843 [ # # ]:UBC 0 : if (numberOfEqualCols == attno - 1)
2844 : 0 : _bt_mark_scankey_required(outkey);
2845 : :
2846 : : /*
2847 : : * We don't support RowCompare using equality; such a qual would
2848 : : * mess up the numberOfEqualCols tracking.
2849 : : */
6317 2850 [ # # ]: 0 : Assert(j != (BTEqualStrategyNumber - 1));
7459 2851 : 0 : continue;
2852 : : }
2853 : :
2854 : : /*
2855 : : * Does this input scan key require further processing as an array?
2856 : : */
8 pg@bowt.ie 2857 [ + + ]:GNC 7121698 : if (cur->sk_strategy == InvalidStrategy)
2858 : : {
2859 : : /* _bt_preprocess_array_keys marked this array key redundant */
2860 [ - + ]: 3 : Assert(arrayKeyData);
2861 [ - + ]: 3 : Assert(cur->sk_flags & SK_SEARCHARRAY);
2862 : 3 : continue;
2863 : : }
2864 : :
2865 [ + + ]: 7121695 : if (cur->sk_strategy == BTEqualStrategyNumber &&
2866 [ + + ]: 6474134 : (cur->sk_flags & SK_SEARCHARRAY))
2867 : : {
2868 : : /* _bt_preprocess_array_keys kept this array key */
2869 [ - + ]: 274 : Assert(arrayKeyData);
2870 : 274 : arrayidx++;
2871 : : }
2872 : :
2873 : : /*
2874 : : * have we seen a scan key for this same attribute and using this same
2875 : : * operator strategy before now?
2876 : : */
2877 [ + + ]: 7121695 : if (xform[j].skey == NULL)
2878 : : {
2879 : : /* nope, so this scan key wins by default (at least for now) */
2880 : 7121671 : xform[j].skey = cur;
2881 : 7121671 : xform[j].ikey = i;
2882 : 7121671 : xform[j].arrayidx = arrayidx;
2883 : : }
2884 : : else
2885 : : {
2886 : 24 : FmgrInfo *orderproc = NULL;
2887 : 24 : BTArrayKeyInfo *array = NULL;
2888 : :
2889 : : /*
2890 : : * Seen one of these before, so keep only the more restrictive key
2891 : : * if possible
2892 : : */
2893 [ + + + + ]: 24 : if (j == (BTEqualStrategyNumber - 1) && arrayKeyData)
2894 : : {
2895 : : /*
2896 : : * Have to set up array keys
2897 : : */
2898 [ - + ]: 6 : if ((cur->sk_flags & SK_SEARCHARRAY))
2899 : : {
8 pg@bowt.ie 2900 :UNC 0 : array = &so->arrayKeys[arrayidx - 1];
2901 : 0 : orderproc = so->orderProcs + i;
2902 : :
2903 [ # # ]: 0 : Assert(array->scan_key == i);
2904 [ # # ]: 0 : Assert(OidIsValid(orderproc->fn_oid));
2905 : : }
8 pg@bowt.ie 2906 [ + - ]:GNC 6 : else if ((xform[j].skey->sk_flags & SK_SEARCHARRAY))
2907 : : {
2908 : 6 : array = &so->arrayKeys[xform[j].arrayidx - 1];
2909 : 6 : orderproc = so->orderProcs + xform[j].ikey;
2910 : :
2911 [ - + ]: 6 : Assert(array->scan_key == xform[j].ikey);
2912 [ - + ]: 6 : Assert(OidIsValid(orderproc->fn_oid));
2913 : : }
2914 : :
2915 : : /*
2916 : : * Both scan keys might have arrays, in which case we'll
2917 : : * arbitrarily pass only one of the arrays. That won't
2918 : : * matter, since _bt_compare_scankey_args is aware that two
2919 : : * SEARCHARRAY scan keys mean that _bt_preprocess_array_keys
2920 : : * failed to eliminate redundant arrays through array merging.
2921 : : * _bt_compare_scankey_args just returns false when it sees
2922 : : * this; it won't even try to examine either array.
2923 : : */
2924 : : }
2925 : :
2926 [ + - ]: 24 : if (_bt_compare_scankey_args(scan, cur, cur, xform[j].skey,
2927 : : array, orderproc, &test_result))
2928 : : {
2929 : : /* Have all we need to determine redundancy */
6317 tgl@sss.pgh.pa.us 2930 [ + + ]:CBC 24 : if (test_result)
2931 : : {
8 pg@bowt.ie 2932 [ + + - + ]:GNC 21 : Assert(!array || array->num_elems > 0);
2933 : :
2934 : : /*
2935 : : * New key is more restrictive, and so replaces old key...
2936 : : */
2937 [ + + ]: 21 : if (j != (BTEqualStrategyNumber - 1) ||
2938 [ + + ]: 6 : !(xform[j].skey->sk_flags & SK_SEARCHARRAY))
2939 : : {
2940 : 18 : xform[j].skey = cur;
2941 : 18 : xform[j].ikey = i;
2942 : 18 : xform[j].arrayidx = arrayidx;
2943 : : }
2944 : : else
2945 : : {
2946 : : /*
2947 : : * ...unless we have to keep the old key because it's
2948 : : * an array that rendered the new key redundant. We
2949 : : * need to make sure that we don't throw away an array
2950 : : * scan key. _bt_compare_scankey_args expects us to
2951 : : * always keep arrays (and discard non-arrays).
2952 : : */
2953 [ - + ]: 3 : Assert(!(cur->sk_flags & SK_SEARCHARRAY));
2954 : : }
2955 : : }
6317 tgl@sss.pgh.pa.us 2956 [ + - ]:CBC 3 : else if (j == (BTEqualStrategyNumber - 1))
2957 : : {
2958 : : /* key == a && key == b, but a != b */
6317 tgl@sss.pgh.pa.us 2959 :GBC 3 : so->qual_ok = false;
2960 : 3 : return;
2961 : : }
2962 : : /* else old key is more restrictive, keep it */
2963 : : }
2964 : : else
2965 : : {
2966 : : /*
2967 : : * We can't determine which key is more restrictive. Push
2968 : : * xform[j] directly to the output array, then set xform[j] to
2969 : : * the new scan key.
2970 : : *
2971 : : * Note: We do things this way around so that our arrays are
2972 : : * always in the same order as their corresponding scan keys,
2973 : : * even with incomplete opfamilies. _bt_advance_array_keys
2974 : : * depends on this.
2975 : : */
6317 tgl@sss.pgh.pa.us 2976 :UBC 0 : ScanKey outkey = &outkeys[new_numberOfKeys++];
2977 : :
8 pg@bowt.ie 2978 :UNC 0 : memcpy(outkey, xform[j].skey, sizeof(ScanKeyData));
2979 [ # # ]: 0 : if (arrayKeyData)
2980 : 0 : keyDataMap[new_numberOfKeys - 1] = xform[j].ikey;
6317 tgl@sss.pgh.pa.us 2981 [ # # ]:UBC 0 : if (numberOfEqualCols == attno - 1)
2982 : 0 : _bt_mark_scankey_required(outkey);
8 pg@bowt.ie 2983 :UNC 0 : xform[j].skey = cur;
2984 : 0 : xform[j].ikey = i;
2985 : 0 : xform[j].arrayidx = arrayidx;
2986 : : }
2987 : : }
2988 : : }
2989 : :
9716 bruce@momjian.us 2990 :CBC 3178655 : so->numberOfKeys = new_numberOfKeys;
2991 : :
2992 : : /*
2993 : : * Now that we've built a temporary mapping from so->keyData[] (output
2994 : : * scan keys) to scan->keyData[] (input scan keys), fix array->scan_key
2995 : : * references. Also consolidate the so->orderProc[] array such that it
2996 : : * can be subscripted using so->keyData[]-wise offsets.
2997 : : */
8 pg@bowt.ie 2998 [ + + ]:GNC 3178655 : if (arrayKeyData)
2999 : 196 : _bt_preprocess_array_keys_final(scan, keyDataMap);
3000 : :
3001 : : /* Could pfree arrayKeyData/keyDataMap now, but not worth the cycles */
3002 : : }
3003 : :
3004 : : #ifdef USE_ASSERT_CHECKING
3005 : : /*
3006 : : * Verify that the scan's qual state matches what we expect at the point that
3007 : : * _bt_start_prim_scan is about to start a just-scheduled new primitive scan.
3008 : : *
3009 : : * We enforce a rule against non-required array scan keys: they must start out
3010 : : * with whatever element is the first for the scan's current scan direction.
3011 : : * See _bt_rewind_nonrequired_arrays comments for an explanation.
3012 : : */
3013 : : static bool
3014 : 285 : _bt_verify_arrays_bt_first(IndexScanDesc scan, ScanDirection dir)
3015 : : {
3016 : 285 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
3017 : 285 : int arrayidx = 0;
3018 : :
3019 [ + + ]: 830 : for (int ikey = 0; ikey < so->numberOfKeys; ikey++)
3020 : : {
3021 : 545 : ScanKey cur = so->keyData + ikey;
3022 : 545 : BTArrayKeyInfo *array = NULL;
3023 : : int first_elem_dir;
3024 : :
3025 [ + + ]: 545 : if (!(cur->sk_flags & SK_SEARCHARRAY) ||
3026 [ - + ]: 435 : cur->sk_strategy != BTEqualStrategyNumber)
3027 : 110 : continue;
3028 : :
3029 : 435 : array = &so->arrayKeys[arrayidx++];
3030 : :
3031 [ + - + + ]: 435 : if (((cur->sk_flags & SK_BT_REQFWD) && ScanDirectionIsForward(dir)) ||
3032 [ + - + - ]: 3 : ((cur->sk_flags & SK_BT_REQBKWD) && ScanDirectionIsBackward(dir)))
3033 : 435 : continue;
3034 : :
8 pg@bowt.ie 3035 [ # # ]:UNC 0 : if (ScanDirectionIsForward(dir))
3036 : 0 : first_elem_dir = 0;
3037 : : else
3038 : 0 : first_elem_dir = array->num_elems - 1;
3039 : :
3040 [ # # ]: 0 : if (array->cur_elem != first_elem_dir)
3041 : 0 : return false;
3042 : : }
3043 : :
8 pg@bowt.ie 3044 :GNC 285 : return _bt_verify_keys_with_arraykeys(scan);
3045 : : }
3046 : :
3047 : : /*
3048 : : * Verify that the scan's "so->keyData[]" scan keys are in agreement with
3049 : : * its array key state
3050 : : */
3051 : : static bool
3052 : 4123 : _bt_verify_keys_with_arraykeys(IndexScanDesc scan)
3053 : : {
3054 : 4123 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
3055 : 4123 : int last_sk_attno = InvalidAttrNumber,
3056 : 4123 : arrayidx = 0;
3057 : :
3058 [ - + ]: 4123 : if (!so->qual_ok)
8 pg@bowt.ie 3059 :UNC 0 : return false;
3060 : :
8 pg@bowt.ie 3061 [ + + ]:GNC 11679 : for (int ikey = 0; ikey < so->numberOfKeys; ikey++)
3062 : : {
3063 : 7556 : ScanKey cur = so->keyData + ikey;
3064 : : BTArrayKeyInfo *array;
3065 : :
3066 [ + + ]: 7556 : if (cur->sk_strategy != BTEqualStrategyNumber ||
3067 [ + + ]: 7298 : !(cur->sk_flags & SK_SEARCHARRAY))
3068 : 2407 : continue;
3069 : :
3070 : 5149 : array = &so->arrayKeys[arrayidx++];
3071 [ - + ]: 5149 : if (array->scan_key != ikey)
8 pg@bowt.ie 3072 :UNC 0 : return false;
3073 : :
8 pg@bowt.ie 3074 [ - + ]:GNC 5149 : if (array->num_elems <= 0)
8 pg@bowt.ie 3075 :UNC 0 : return false;
3076 : :
8 pg@bowt.ie 3077 [ - + ]:GNC 5149 : if (cur->sk_argument != array->elem_values[array->cur_elem])
8 pg@bowt.ie 3078 :UNC 0 : return false;
8 pg@bowt.ie 3079 [ - + ]:GNC 5149 : if (last_sk_attno > cur->sk_attno)
8 pg@bowt.ie 3080 :UNC 0 : return false;
8 pg@bowt.ie 3081 :GNC 5149 : last_sk_attno = cur->sk_attno;
3082 : : }
3083 : :
3084 [ - + ]: 4123 : if (arrayidx != so->numArrayKeys)
8 pg@bowt.ie 3085 :UNC 0 : return false;
3086 : :
8 pg@bowt.ie 3087 :GNC 4123 : return true;
3088 : : }
3089 : : #endif
3090 : :
3091 : : /*
3092 : : * Compare two scankey values using a specified operator.
3093 : : *
3094 : : * The test we want to perform is logically "leftarg op rightarg", where
3095 : : * leftarg and rightarg are the sk_argument values in those ScanKeys, and
3096 : : * the comparison operator is the one in the op ScanKey. However, in
3097 : : * cross-data-type situations we may need to look up the correct operator in
3098 : : * the index's opfamily: it is the one having amopstrategy = op->sk_strategy
3099 : : * and amoplefttype/amoprighttype equal to the two argument datatypes.
3100 : : *
3101 : : * If the opfamily doesn't supply a complete set of cross-type operators we
3102 : : * may not be able to make the comparison. If we can make the comparison
3103 : : * we store the operator result in *result and return true. We return false
3104 : : * if the comparison could not be made.
3105 : : *
3106 : : * If either leftarg or rightarg are an array, we'll apply array-specific
3107 : : * rules to determine which array elements are redundant on behalf of caller.
3108 : : * It is up to our caller to save whichever of the two scan keys is the array,
3109 : : * and discard the non-array scan key (the non-array scan key is guaranteed to
3110 : : * be redundant with any complete opfamily). Caller isn't expected to call
3111 : : * here with a pair of array scan keys provided we're dealing with a complete
3112 : : * opfamily (_bt_preprocess_array_keys will merge array keys together to make
3113 : : * sure of that).
3114 : : *
3115 : : * Note: we'll also shrink caller's array as needed to eliminate redundant
3116 : : * array elements. One reason why caller should prefer to discard non-array
3117 : : * scan keys is so that we'll have the opportunity to shrink the array
3118 : : * multiple times, in multiple calls (for each of several other scan keys on
3119 : : * the same index attribute).
3120 : : *
3121 : : * Note: op always points at the same ScanKey as either leftarg or rightarg.
3122 : : * Since we don't scribble on the scankeys themselves, this aliasing should
3123 : : * cause no trouble.
3124 : : *
3125 : : * Note: this routine needs to be insensitive to any DESC option applied
3126 : : * to the index column. For example, "x < 4" is a tighter constraint than
3127 : : * "x < 5" regardless of which way the index is sorted.
3128 : : */
3129 : : static bool
6317 tgl@sss.pgh.pa.us 3130 :CBC 73 : _bt_compare_scankey_args(IndexScanDesc scan, ScanKey op,
3131 : : ScanKey leftarg, ScanKey rightarg,
3132 : : BTArrayKeyInfo *array, FmgrInfo *orderproc,
3133 : : bool *result)
3134 : : {
3135 : 73 : Relation rel = scan->indexRelation;
3136 : : Oid lefttype,
3137 : : righttype,
3138 : : optype,
3139 : : opcintype,
3140 : : cmp_op;
3141 : : StrategyNumber strat;
3142 : :
3143 : : /*
3144 : : * First, deal with cases where one or both args are NULL. This should
3145 : : * only happen when the scankeys represent IS NULL/NOT NULL conditions.
3146 : : */
5217 3147 [ + + ]: 73 : if ((leftarg->sk_flags | rightarg->sk_flags) & SK_ISNULL)
3148 : : {
3149 : : bool leftnull,
3150 : : rightnull;
3151 : :
3152 [ - + ]: 6 : if (leftarg->sk_flags & SK_ISNULL)
3153 : : {
5217 tgl@sss.pgh.pa.us 3154 [ # # ]:LBC (3) : Assert(leftarg->sk_flags & (SK_SEARCHNULL | SK_SEARCHNOTNULL));
3155 : (3) : leftnull = true;
3156 : : }
3157 : : else
5217 tgl@sss.pgh.pa.us 3158 :CBC 6 : leftnull = false;
3159 [ + - ]: 6 : if (rightarg->sk_flags & SK_ISNULL)
3160 : : {
3161 [ - + ]: 6 : Assert(rightarg->sk_flags & (SK_SEARCHNULL | SK_SEARCHNOTNULL));
3162 : 6 : rightnull = true;
3163 : : }
3164 : : else
5217 tgl@sss.pgh.pa.us 3165 :LBC (3) : rightnull = false;
3166 : :
3167 : : /*
3168 : : * We treat NULL as either greater than or less than all other values.
3169 : : * Since true > false, the tests below work correctly for NULLS LAST
3170 : : * logic. If the index is NULLS FIRST, we need to flip the strategy.
3171 : : */
5217 tgl@sss.pgh.pa.us 3172 :CBC 6 : strat = op->sk_strategy;
3173 [ - + ]: 6 : if (op->sk_flags & SK_BT_NULLS_FIRST)
5217 tgl@sss.pgh.pa.us 3174 :UBC 0 : strat = BTCommuteStrategyNumber(strat);
3175 : :
5217 tgl@sss.pgh.pa.us 3176 [ + - - - :CBC 6 : switch (strat)
- - ]
3177 : : {
3178 : 6 : case BTLessStrategyNumber:
3179 : 6 : *result = (leftnull < rightnull);
3180 : 6 : break;
5217 tgl@sss.pgh.pa.us 3181 :UBC 0 : case BTLessEqualStrategyNumber:
3182 : 0 : *result = (leftnull <= rightnull);
3183 : 0 : break;
3184 : 0 : case BTEqualStrategyNumber:
3185 : 0 : *result = (leftnull == rightnull);
3186 : 0 : break;
3187 : 0 : case BTGreaterEqualStrategyNumber:
3188 : 0 : *result = (leftnull >= rightnull);
3189 : 0 : break;
3190 : 0 : case BTGreaterStrategyNumber:
3191 : 0 : *result = (leftnull > rightnull);
3192 : 0 : break;
3193 : 0 : default:
3194 [ # # ]: 0 : elog(ERROR, "unrecognized StrategyNumber: %d", (int) strat);
3195 : : *result = false; /* keep compiler quiet */
3196 : : break;
3197 : : }
5217 tgl@sss.pgh.pa.us 3198 :CBC 6 : return true;
3199 : : }
3200 : :
3201 : : /*
3202 : : * If either leftarg or rightarg are equality-type array scankeys, we need
3203 : : * specialized handling (since by now we know that IS NULL wasn't used)
3204 : : */
8 pg@bowt.ie 3205 [ + + ]:GNC 67 : if (array)
3206 : : {
3207 : : bool leftarray,
3208 : : rightarray;
3209 : :
3210 [ + + ]: 24 : leftarray = ((leftarg->sk_flags & SK_SEARCHARRAY) &&
3211 [ + - ]: 9 : leftarg->sk_strategy == BTEqualStrategyNumber);
3212 [ + + ]: 21 : rightarray = ((rightarg->sk_flags & SK_SEARCHARRAY) &&
3213 [ + - ]: 6 : rightarg->sk_strategy == BTEqualStrategyNumber);
3214 : :
3215 : : /*
3216 : : * _bt_preprocess_array_keys is responsible for merging together array
3217 : : * scan keys, and will do so whenever the opfamily has the required
3218 : : * cross-type support. If it failed to do that, we handle it just
3219 : : * like the case where we can't make the comparison ourselves.
3220 : : */
3221 [ + + - + ]: 15 : if (leftarray && rightarray)
3222 : : {
3223 : : /* Can't make the comparison */
8 pg@bowt.ie 3224 :UNC 0 : *result = false; /* suppress compiler warnings */
3225 : 0 : return false;
3226 : : }
3227 : :
3228 : : /*
3229 : : * Otherwise we need to determine if either one of leftarg or rightarg
3230 : : * uses an array, then pass this through to a dedicated helper
3231 : : * function.
3232 : : */
8 pg@bowt.ie 3233 [ + + ]:GNC 15 : if (leftarray)
3234 : 9 : return _bt_compare_array_scankey_args(scan, leftarg, rightarg,
3235 : : orderproc, array, result);
3236 [ + - ]: 6 : else if (rightarray)
3237 : 6 : return _bt_compare_array_scankey_args(scan, rightarg, leftarg,
3238 : : orderproc, array, result);
3239 : :
3240 : : /* FALL THRU */
3241 : : }
3242 : :
3243 : : /*
3244 : : * The opfamily we need to worry about is identified by the index column.
3245 : : */
6317 tgl@sss.pgh.pa.us 3246 [ - + ]:CBC 52 : Assert(leftarg->sk_attno == rightarg->sk_attno);
3247 : :
3248 : 52 : opcintype = rel->rd_opcintype[leftarg->sk_attno - 1];
3249 : :
3250 : : /*
3251 : : * Determine the actual datatypes of the ScanKey arguments. We have to
3252 : : * support the convention that sk_subtype == InvalidOid means the opclass
3253 : : * input type; this is a hack to simplify life for ScanKeyInit().
3254 : : */
3255 : 52 : lefttype = leftarg->sk_subtype;
3256 [ - + ]: 52 : if (lefttype == InvalidOid)
6317 tgl@sss.pgh.pa.us 3257 :UBC 0 : lefttype = opcintype;
6317 tgl@sss.pgh.pa.us 3258 :CBC 52 : righttype = rightarg->sk_subtype;
3259 [ - + ]: 52 : if (righttype == InvalidOid)
6317 tgl@sss.pgh.pa.us 3260 :UBC 0 : righttype = opcintype;
6317 tgl@sss.pgh.pa.us 3261 :CBC 52 : optype = op->sk_subtype;
3262 [ - + ]: 52 : if (optype == InvalidOid)
6317 tgl@sss.pgh.pa.us 3263 :UBC 0 : optype = opcintype;
3264 : :
3265 : : /*
3266 : : * If leftarg and rightarg match the types expected for the "op" scankey,
3267 : : * we can use its already-looked-up comparison function.
3268 : : */
6317 tgl@sss.pgh.pa.us 3269 [ + + + - ]:CBC 52 : if (lefttype == opcintype && righttype == optype)
3270 : : {
4751 3271 : 49 : *result = DatumGetBool(FunctionCall2Coll(&op->sk_func,
3272 : : op->sk_collation,
3273 : : leftarg->sk_argument,
3274 : : rightarg->sk_argument));
6317 3275 : 49 : return true;
3276 : : }
3277 : :
3278 : : /*
3279 : : * Otherwise, we need to go to the syscache to find the appropriate
3280 : : * operator. (This cannot result in infinite recursion, since no
3281 : : * indexscan initiated by syscache lookup will use cross-data-type
3282 : : * operators.)
3283 : : *
3284 : : * If the sk_strategy was flipped by _bt_fix_scankey_strategy, we have to
3285 : : * un-flip it to get the correct opfamily member.
3286 : : */
6305 tgl@sss.pgh.pa.us 3287 :GBC 3 : strat = op->sk_strategy;
3288 [ - + ]: 3 : if (op->sk_flags & SK_BT_DESC)
6305 tgl@sss.pgh.pa.us 3289 :UBC 0 : strat = BTCommuteStrategyNumber(strat);
3290 : :
6317 tgl@sss.pgh.pa.us 3291 :GBC 3 : cmp_op = get_opfamily_member(rel->rd_opfamily[leftarg->sk_attno - 1],
3292 : : lefttype,
3293 : : righttype,
3294 : : strat);
3295 [ + - ]: 3 : if (OidIsValid(cmp_op))
3296 : : {
3297 : 3 : RegProcedure cmp_proc = get_opcode(cmp_op);
3298 : :
3299 [ + - ]: 3 : if (RegProcedureIsValid(cmp_proc))
3300 : : {
4751 3301 : 3 : *result = DatumGetBool(OidFunctionCall2Coll(cmp_proc,
3302 : : op->sk_collation,
3303 : : leftarg->sk_argument,
3304 : : rightarg->sk_argument));
6317 3305 : 3 : return true;
3306 : : }
3307 : : }
3308 : :
3309 : : /* Can't make the comparison */
6317 tgl@sss.pgh.pa.us 3310 :UBC 0 : *result = false; /* suppress compiler warnings */
3311 : 0 : return false;
3312 : : }
3313 : :
3314 : : /*
3315 : : * Adjust a scankey's strategy and flags setting as needed for indoptions.
3316 : : *
3317 : : * We copy the appropriate indoption value into the scankey sk_flags
3318 : : * (shifting to avoid clobbering system-defined flag bits). Also, if
3319 : : * the DESC option is set, commute (flip) the operator strategy number.
3320 : : *
3321 : : * A secondary purpose is to check for IS NULL/NOT NULL scankeys and set up
3322 : : * the strategy field correctly for them.
3323 : : *
3324 : : * Lastly, for ordinary scankeys (not IS NULL/NOT NULL), we check for a
3325 : : * NULL comparison value. Since all btree operators are assumed strict,
3326 : : * a NULL means that the qual cannot be satisfied. We return true if the
3327 : : * comparison value isn't NULL, or false if the scan should be abandoned.
3328 : : *
3329 : : * This function is applied to the *input* scankey structure; therefore
3330 : : * on a rescan we will be looking at already-processed scankeys. Hence
3331 : : * we have to be careful not to re-commute the strategy if we already did it.
3332 : : * It's a bit ugly to modify the caller's copy of the scankey but in practice
3333 : : * there shouldn't be any problem, since the index's indoptions are certainly
3334 : : * not going to change while the scankey survives.
3335 : : */
3336 : : static bool
5215 tgl@sss.pgh.pa.us 3337 :CBC 10358262 : _bt_fix_scankey_strategy(ScanKey skey, int16 *indoption)
3338 : : {
3339 : : int addflags;
3340 : :
6305 3341 : 10358262 : addflags = indoption[skey->sk_attno - 1] << SK_BT_INDOPTION_SHIFT;
3342 : :
3343 : : /*
3344 : : * We treat all btree operators as strict (even if they're not so marked
3345 : : * in pg_proc). This means that it is impossible for an operator condition
3346 : : * with a NULL comparison constant to succeed, and we can reject it right
3347 : : * away.
3348 : : *
3349 : : * However, we now also support "x IS NULL" clauses as search conditions,
3350 : : * so in that case keep going. The planner has not filled in any
3351 : : * particular strategy in this case, so set it to BTEqualStrategyNumber
3352 : : * --- we can treat IS NULL as an equality operator for purposes of search
3353 : : * strategy.
3354 : : *
3355 : : * Likewise, "x IS NOT NULL" is supported. We treat that as either "less
3356 : : * than NULL" in a NULLS LAST index, or "greater than NULL" in a NULLS
3357 : : * FIRST index.
3358 : : *
3359 : : * Note: someday we might have to fill in sk_collation from the index
3360 : : * column's collation. At the moment this is a non-issue because we'll
3361 : : * never actually call the comparison operator on a NULL.
3362 : : */
5215 3363 [ + + ]: 10358262 : if (skey->sk_flags & SK_ISNULL)
3364 : : {
3365 : : /* SK_ISNULL shouldn't be set in a row header scankey */
3366 [ - + ]: 56875 : Assert(!(skey->sk_flags & SK_ROW_HEADER));
3367 : :
3368 : : /* Set indoption flags in scankey (might be done already) */
3369 : 56875 : skey->sk_flags |= addflags;
3370 : :
3371 : : /* Set correct strategy for IS NULL or NOT NULL search */
3372 [ + + ]: 56875 : if (skey->sk_flags & SK_SEARCHNULL)
3373 : : {
3374 : 70 : skey->sk_strategy = BTEqualStrategyNumber;
3375 : 70 : skey->sk_subtype = InvalidOid;
4751 3376 : 70 : skey->sk_collation = InvalidOid;
3377 : : }
5215 3378 [ + + ]: 56805 : else if (skey->sk_flags & SK_SEARCHNOTNULL)
3379 : : {
3380 [ + + ]: 55442 : if (skey->sk_flags & SK_BT_NULLS_FIRST)
3381 : 18 : skey->sk_strategy = BTGreaterStrategyNumber;
3382 : : else
3383 : 55424 : skey->sk_strategy = BTLessStrategyNumber;
3384 : 55442 : skey->sk_subtype = InvalidOid;
4751 3385 : 55442 : skey->sk_collation = InvalidOid;
3386 : : }
3387 : : else
3388 : : {
3389 : : /* regular qual, so it cannot be satisfied */
5215 3390 : 1363 : return false;
3391 : : }
3392 : :
3393 : : /* Needn't do the rest */
3394 : 55512 : return true;
3395 : : }
3396 : :
3397 : : /* Adjust strategy for DESC, if we didn't already */
6305 3398 [ + + + - ]: 10301387 : if ((addflags & SK_BT_DESC) && !(skey->sk_flags & SK_BT_DESC))
6305 tgl@sss.pgh.pa.us 3399 :GBC 3 : skey->sk_strategy = BTCommuteStrategyNumber(skey->sk_strategy);
6305 tgl@sss.pgh.pa.us 3400 :CBC 10301387 : skey->sk_flags |= addflags;
3401 : :
3402 : : /* If it's a row header, fix row member flags and strategies similarly */
3403 [ + + ]: 10301387 : if (skey->sk_flags & SK_ROW_HEADER)
3404 : : {
3405 : 18 : ScanKey subkey = (ScanKey) DatumGetPointer(skey->sk_argument);
3406 : :
3407 : : for (;;)
3408 : : {
3409 [ - + ]: 36 : Assert(subkey->sk_flags & SK_ROW_MEMBER);
3410 : 36 : addflags = indoption[subkey->sk_attno - 1] << SK_BT_INDOPTION_SHIFT;
3411 [ - + - - ]: 36 : if ((addflags & SK_BT_DESC) && !(subkey->sk_flags & SK_BT_DESC))
6305 tgl@sss.pgh.pa.us 3412 :UBC 0 : subkey->sk_strategy = BTCommuteStrategyNumber(subkey->sk_strategy);
6305 tgl@sss.pgh.pa.us 3413 :CBC 36 : subkey->sk_flags |= addflags;
3414 [ + + ]: 36 : if (subkey->sk_flags & SK_ROW_END)
3415 : 18 : break;
3416 : 18 : subkey++;
3417 : : }
3418 : : }
3419 : :
5215 3420 : 10301387 : return true;
3421 : : }
3422 : :
3423 : : /*
3424 : : * Mark a scankey as "required to continue the scan".
3425 : : *
3426 : : * Depending on the operator type, the key may be required for both scan
3427 : : * directions or just one. Also, if the key is a row comparison header,
3428 : : * we have to mark its first subsidiary ScanKey as required. (Subsequent
3429 : : * subsidiary ScanKeys are normally for lower-order columns, and thus
3430 : : * cannot be required, since they're after the first non-equality scankey.)
3431 : : *
3432 : : * Note: when we set required-key flag bits in a subsidiary scankey, we are
3433 : : * scribbling on a data structure belonging to the index AM's caller, not on
3434 : : * our private copy. This should be OK because the marking will not change
3435 : : * from scan to scan within a query, and so we'd just re-mark the same way
3436 : : * anyway on a rescan. Something to keep an eye on though.
3437 : : */
3438 : : static void
6654 3439 : 10356496 : _bt_mark_scankey_required(ScanKey skey)
3440 : : {
3441 : : int addflags;
3442 : :
3443 [ + + + - ]: 10356496 : switch (skey->sk_strategy)
3444 : : {
3445 : 56916 : case BTLessStrategyNumber:
3446 : : case BTLessEqualStrategyNumber:
3447 : 56916 : addflags = SK_BT_REQFWD;
3448 : 56916 : break;
3449 : 9652566 : case BTEqualStrategyNumber:
3450 : 9652566 : addflags = SK_BT_REQFWD | SK_BT_REQBKWD;
3451 : 9652566 : break;
3452 : 647014 : case BTGreaterEqualStrategyNumber:
3453 : : case BTGreaterStrategyNumber:
3454 : 647014 : addflags = SK_BT_REQBKWD;
3455 : 647014 : break;
6654 tgl@sss.pgh.pa.us 3456 :UBC 0 : default:
3457 [ # # ]: 0 : elog(ERROR, "unrecognized StrategyNumber: %d",
3458 : : (int) skey->sk_strategy);
3459 : : addflags = 0; /* keep compiler quiet */
3460 : : break;
3461 : : }
3462 : :
6654 tgl@sss.pgh.pa.us 3463 :CBC 10356496 : skey->sk_flags |= addflags;
3464 : :
3465 [ + + ]: 10356496 : if (skey->sk_flags & SK_ROW_HEADER)
3466 : : {
6402 bruce@momjian.us 3467 : 18 : ScanKey subkey = (ScanKey) DatumGetPointer(skey->sk_argument);
3468 : :
3469 : : /* First subkey should be same column/operator as the header */
2958 tgl@sss.pgh.pa.us 3470 [ - + ]: 18 : Assert(subkey->sk_flags & SK_ROW_MEMBER);
3471 [ - + ]: 18 : Assert(subkey->sk_attno == skey->sk_attno);
3472 [ - + ]: 18 : Assert(subkey->sk_strategy == skey->sk_strategy);
3473 : 18 : subkey->sk_flags |= addflags;
3474 : : }
6654 3475 : 10356496 : }
3476 : :
3477 : : /*
3478 : : * Test whether an indextuple satisfies all the scankey conditions.
3479 : : *
3480 : : * Return true if so, false if not. If the tuple fails to pass the qual,
3481 : : * we also determine whether there's any need to continue the scan beyond
3482 : : * this tuple, and set pstate.continuescan accordingly. See comments for
3483 : : * _bt_preprocess_keys(), above, about how this is done.
3484 : : *
3485 : : * Forward scan callers can pass a high key tuple in the hopes of having
3486 : : * us set *continuescan to false, and avoiding an unnecessary visit to
3487 : : * the page to the right.
3488 : : *
3489 : : * Advances the scan's array keys when necessary for arrayKeys=true callers.
3490 : : * Caller can avoid all array related side-effects when calling just to do a
3491 : : * page continuescan precheck -- pass arrayKeys=false for that. Scans without
3492 : : * any arrays keys must always pass arrayKeys=false.
3493 : : *
3494 : : * Also stops and starts primitive index scans for arrayKeys=true callers.
3495 : : * Scans with array keys are required to set up page state that helps us with
3496 : : * this. The page's finaltup tuple (the page high key for a forward scan, or
3497 : : * the page's first non-pivot tuple for a backward scan) must be set in
3498 : : * pstate.finaltup ahead of the first call here for the page (or possibly the
3499 : : * first call after an initial continuescan-setting page precheck call). Set
3500 : : * this to NULL for rightmost page (or the leftmost page for backwards scans).
3501 : : *
3502 : : * scan: index scan descriptor (containing a search-type scankey)
3503 : : * pstate: page level input and output parameters
3504 : : * arrayKeys: should we advance the scan's array keys if necessary?
3505 : : * tuple: index tuple to test
3506 : : * tupnatts: number of attributes in tupnatts (high key may be truncated)
3507 : : */
3508 : : bool
8 pg@bowt.ie 3509 :GNC 26089089 : _bt_checkkeys(IndexScanDesc scan, BTReadPageState *pstate, bool arrayKeys,
3510 : : IndexTuple tuple, int tupnatts)
3511 : : {
3512 : 26089089 : TupleDesc tupdesc = RelationGetDescr(scan->indexRelation);
3513 : 26089089 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
3514 : 26089089 : ScanDirection dir = pstate->dir;
3515 : 26089089 : int ikey = 0;
3516 : : bool res;
3517 : :
1849 pg@bowt.ie 3518 [ + + - + ]:CBC 26089089 : Assert(BTreeTupleGetNAtts(tuple, scan->indexRelation) == tupnatts);
3519 : :
8 pg@bowt.ie 3520 :GNC 26089089 : res = _bt_check_compare(scan, dir, tuple, tupnatts, tupdesc,
3521 : 26089089 : arrayKeys, pstate->prechecked, pstate->firstmatch,
3522 : : &pstate->continuescan, &ikey);
3523 : :
3524 : : #ifdef USE_ASSERT_CHECKING
3525 [ + + + + ]: 26089089 : if (!arrayKeys && so->numArrayKeys)
3526 : : {
3527 : : /*
3528 : : * This is a continuescan precheck call for a scan with array keys.
3529 : : *
3530 : : * Assert that the scan isn't in danger of becoming confused.
3531 : : */
3532 [ + - + - : 9 : Assert(!so->scanBehind && !pstate->prechecked && !pstate->firstmatch);
- + ]
3533 [ - + ]: 9 : Assert(!_bt_tuple_before_array_skeys(scan, dir, tuple, tupdesc,
3534 : : tupnatts, false, 0, NULL));
3535 : : }
3536 [ + + + + ]: 26089089 : if (pstate->prechecked || pstate->firstmatch)
3537 : : {
3538 : : bool dcontinuescan;
3539 : 20716788 : int dikey = 0;
3540 : :
3541 : : /*
3542 : : * Call relied on continuescan/firstmatch prechecks -- assert that we
3543 : : * get the same answer without those optimizations
3544 : : */
3545 [ - + ]: 20716788 : Assert(res == _bt_check_compare(scan, dir, tuple, tupnatts, tupdesc,
3546 : : false, false, false,
3547 : : &dcontinuescan, &dikey));
3548 [ - + ]: 20716788 : Assert(pstate->continuescan == dcontinuescan);
3549 : : }
3550 : : #endif
3551 : :
3552 : : /*
3553 : : * Only one _bt_check_compare call is required in the common case where
3554 : : * there are no equality strategy array scan keys. Otherwise we can only
3555 : : * accept _bt_check_compare's answer unreservedly when it didn't set
3556 : : * pstate.continuescan=false.
3557 : : */
3558 [ + + + + ]: 26089089 : if (!arrayKeys || pstate->continuescan)
3559 : 26085928 : return res;
3560 : :
3561 : : /*
3562 : : * _bt_check_compare call set continuescan=false in the presence of
3563 : : * equality type array keys. This could mean that the tuple is just past
3564 : : * the end of matches for the current array keys.
3565 : : *
3566 : : * It's also possible that the scan is still _before_ the _start_ of
3567 : : * tuples matching the current set of array keys. Check for that first.
3568 : : */
3569 [ + + ]: 3161 : if (_bt_tuple_before_array_skeys(scan, dir, tuple, tupdesc, tupnatts, true,
3570 : : ikey, NULL))
3571 : : {
3572 : : /*
3573 : : * Tuple is still before the start of matches according to the scan's
3574 : : * required array keys (according to _all_ of its required equality
3575 : : * strategy keys, actually).
3576 : : *
3577 : : * _bt_advance_array_keys occasionally sets so->scanBehind to signal
3578 : : * that the scan's current position/tuples might be significantly
3579 : : * behind (multiple pages behind) its current array keys. When this
3580 : : * happens, we need to be prepared to recover by starting a new
3581 : : * primitive index scan here, on our own.
3582 : : */
3583 [ + + - + ]: 1340 : Assert(!so->scanBehind ||
3584 : : so->keyData[ikey].sk_strategy == BTEqualStrategyNumber);
3585 [ + + + - : 1355 : if (unlikely(so->scanBehind) && pstate->finaltup &&
- + ]
3586 [ + - ]: 30 : _bt_tuple_before_array_skeys(scan, dir, pstate->finaltup, tupdesc,
8 pg@bowt.ie 3587 :UNC 0 : BTreeTupleGetNAtts(pstate->finaltup,
3588 : : scan->indexRelation),
3589 : : false, 0, NULL))
3590 : : {
3591 : : /* Cut our losses -- start a new primitive index scan now */
3592 : 0 : pstate->continuescan = false;
3593 : 0 : so->needPrimScan = true;
3594 : : }
3595 : : else
3596 : : {
3597 : : /* Override _bt_check_compare, continue primitive scan */
8 pg@bowt.ie 3598 :GNC 1340 : pstate->continuescan = true;
3599 : :
3600 : : /*
3601 : : * We will end up here repeatedly given a group of tuples > the
3602 : : * previous array keys and < the now-current keys (for a backwards
3603 : : * scan it's just the same, though the operators swap positions).
3604 : : *
3605 : : * We must avoid allowing this linear search process to scan very
3606 : : * many tuples from well before the start of tuples matching the
3607 : : * current array keys (or from well before the point where we'll
3608 : : * once again have to advance the scan's array keys).
3609 : : *
3610 : : * We keep the overhead under control by speculatively "looking
3611 : : * ahead" to later still-unscanned items from this same leaf page.
3612 : : * We'll only attempt this once the number of tuples that the
3613 : : * linear search process has examined starts to get out of hand.
3614 : : */
3615 : 1340 : pstate->rechecks++;
3616 [ + + ]: 1340 : if (pstate->rechecks >= LOOK_AHEAD_REQUIRED_RECHECKS)
3617 : : {
3618 : : /* See if we should skip ahead within the current leaf page */
3619 : 529 : _bt_checkkeys_look_ahead(scan, pstate, tupnatts, tupdesc);
3620 : :
3621 : : /*
3622 : : * Might have set pstate.skip to a later page offset. When
3623 : : * that happens then _bt_readpage caller will inexpensively
3624 : : * skip ahead to a later tuple from the same page (the one
3625 : : * just after the tuple we successfully "looked ahead" to).
3626 : : */
3627 : : }
3628 : : }
3629 : :
3630 : : /* This indextuple doesn't match the current qual, in any case */
3631 : 1340 : return false;
3632 : : }
3633 : :
3634 : : /*
3635 : : * Caller's tuple is >= the current set of array keys and other equality
3636 : : * constraint scan keys (or <= if this is a backwards scan). It's now
3637 : : * clear that we _must_ advance any required array keys in lockstep with
3638 : : * the scan.
3639 : : */
3640 : 1821 : return _bt_advance_array_keys(scan, pstate, tuple, tupnatts, tupdesc,
3641 : : ikey, true);
3642 : : }
3643 : :
3644 : : /*
3645 : : * Test whether an indextuple satisfies current scan condition.
3646 : : *
3647 : : * Return true if so, false if not. If not, also sets *continuescan to false
3648 : : * when it's also not possible for any later tuples to pass the current qual
3649 : : * (with the scan's current set of array keys, in the current scan direction),
3650 : : * in addition to setting *ikey to the so->keyData[] subscript/offset for the
3651 : : * unsatisfied scan key (needed when caller must consider advancing the scan's
3652 : : * array keys).
3653 : : *
3654 : : * This is a subroutine for _bt_checkkeys. We provisionally assume that
3655 : : * reaching the end of the current set of required keys (in particular the
3656 : : * current required array keys) ends the ongoing (primitive) index scan.
3657 : : * Callers without array keys should just end the scan right away when they
3658 : : * find that continuescan has been set to false here by us. Things are more
3659 : : * complicated for callers with array keys.
3660 : : *
3661 : : * Callers with array keys must first consider advancing the arrays when
3662 : : * continuescan has been set to false here by us. They must then consider if
3663 : : * it really does make sense to end the current (primitive) index scan, in
3664 : : * light of everything that is known at that point. (In general when we set
3665 : : * continuescan=false for these callers it must be treated as provisional.)
3666 : : *
3667 : : * We deal with advancing unsatisfied non-required arrays directly, though.
3668 : : * This is safe, since by definition non-required keys can't end the scan.
3669 : : * This is just how we determine if non-required arrays are just unsatisfied
3670 : : * by the current array key, or if they're truly unsatisfied (that is, if
3671 : : * they're unsatisfied by every possible array key).
3672 : : *
3673 : : * Though we advance non-required array keys on our own, that shouldn't have
3674 : : * any lasting consequences for the scan. By definition, non-required arrays
3675 : : * have no fixed relationship with the scan's progress. (There are delicate
3676 : : * considerations for non-required arrays when the arrays need to be advanced
3677 : : * following our setting continuescan to false, but that doesn't concern us.)
3678 : : *
3679 : : * Pass advancenonrequired=false to avoid all array related side effects.
3680 : : * This allows _bt_advance_array_keys caller to avoid infinite recursion.
3681 : : */
3682 : : static bool
3683 : 46806786 : _bt_check_compare(IndexScanDesc scan, ScanDirection dir,
3684 : : IndexTuple tuple, int tupnatts, TupleDesc tupdesc,
3685 : : bool advancenonrequired, bool prechecked, bool firstmatch,
3686 : : bool *continuescan, int *ikey)
3687 : : {
3688 : 46806786 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
3689 : :
8 pg@bowt.ie 3690 :CBC 46806786 : *continuescan = true; /* default assumption */
3691 : :
8 pg@bowt.ie 3692 [ + + ]:GNC 88619010 : for (; *ikey < so->numberOfKeys; (*ikey)++)
3693 : : {
3694 : 51094027 : ScanKey key = so->keyData + *ikey;
3695 : : Datum datum;
3696 : : bool isNull;
191 akorotkov@postgresql 3697 : 51094027 : bool requiredSameDir = false,
8 pg@bowt.ie 3698 : 51094027 : requiredOppositeDirOnly = false;
3699 : :
3700 : : /*
3701 : : * Check if the key is required in the current scan direction, in the
3702 : : * opposite scan direction _only_, or in neither direction
3703 : : */
191 akorotkov@postgresql 3704 [ + + + + ]: 51094027 : if (((key->sk_flags & SK_BT_REQFWD) && ScanDirectionIsForward(dir)) ||
3705 [ + + + + ]: 12705636 : ((key->sk_flags & SK_BT_REQBKWD) && ScanDirectionIsBackward(dir)))
3706 : 38405577 : requiredSameDir = true;
3707 [ + + - + ]: 12688450 : else if (((key->sk_flags & SK_BT_REQFWD) && ScanDirectionIsBackward(dir)) ||
3708 [ + + + - ]: 5414417 : ((key->sk_flags & SK_BT_REQBKWD) && ScanDirectionIsForward(dir)))
8 pg@bowt.ie 3709 : 12231411 : requiredOppositeDirOnly = true;
3710 : :
3711 : : /*
3712 : : * If the caller told us the *continuescan flag is known to be true
3713 : : * for the last item on the page, then we know the keys required for
3714 : : * the current direction scan should be matched. Otherwise, the
3715 : : * *continuescan flag would be set for the current item and
3716 : : * subsequently the last item on the page accordingly.
3717 : : *
3718 : : * If the key is required for the opposite direction scan, we can skip
3719 : : * the check if the caller tells us there was already at least one
3720 : : * matching item on the page. Also, we require the *continuescan flag
3721 : : * to be true for the last item on the page to know there are no
3722 : : * NULLs.
3723 : : *
3724 : : * Both cases above work except for the row keys, where NULLs could be
3725 : : * found in the middle of matching values.
3726 : : */
3727 [ + + + + ]: 51094027 : if (prechecked &&
3728 [ + + + + ]: 1082501 : (requiredSameDir || (requiredOppositeDirOnly && firstmatch)) &&
3729 [ + - ]: 1063109 : !(key->sk_flags & SK_ROW_HEADER))
191 akorotkov@postgresql 3730 : 19095760 : continue;
3731 : :
1849 pg@bowt.ie 3732 [ + + ]:CBC 50030918 : if (key->sk_attno > tupnatts)
3733 : : {
3734 : : /*
3735 : : * This attribute is truncated (must be high key). The value for
3736 : : * this attribute in the first non-pivot tuple on the page to the
3737 : : * right could be any possible value. Assume that truncated
3738 : : * attribute passes the qual.
3739 : : */
1509 3740 [ - + ]: 1535 : Assert(BTreeTupleIsPivot(tuple));
1849 3741 : 1535 : continue;
3742 : : }
3743 : :
3744 : : /* row-comparison keys need special processing */
6654 tgl@sss.pgh.pa.us 3745 [ + + ]: 50029383 : if (key->sk_flags & SK_ROW_HEADER)
3746 : : {
1849 pg@bowt.ie 3747 [ + + ]: 1962 : if (_bt_check_rowcompare(key, tuple, tupnatts, tupdesc, dir,
3748 : : continuescan))
6654 tgl@sss.pgh.pa.us 3749 : 1908 : continue;
1849 pg@bowt.ie 3750 : 9281803 : return false;
3751 : : }
3752 : :
9716 bruce@momjian.us 3753 : 50027421 : datum = index_getattr(tuple,
8664 tgl@sss.pgh.pa.us 3754 : 50027421 : key->sk_attno,
3755 : : tupdesc,
3756 : : &isNull);
3757 : :
3758 [ + + ]: 50027421 : if (key->sk_flags & SK_ISNULL)
3759 : : {
3760 : : /* Handle IS NULL/NOT NULL tests */
5217 3761 [ + + ]: 18072405 : if (key->sk_flags & SK_SEARCHNULL)
3762 : : {
3763 [ + + ]: 43237 : if (isNull)
5161 bruce@momjian.us 3764 : 100 : continue; /* tuple satisfies this qual */
3765 : : }
3766 : : else
3767 : : {
5217 tgl@sss.pgh.pa.us 3768 [ - + ]: 18029168 : Assert(key->sk_flags & SK_SEARCHNOTNULL);
3769 [ + + ]: 18029168 : if (!isNull)
5161 bruce@momjian.us 3770 : 18029108 : continue; /* tuple satisfies this qual */
3771 : : }
3772 : :
3773 : : /*
3774 : : * Tuple fails this qual. If it's a required qual for the current
3775 : : * scan direction, then we can conclude no further tuples will
3776 : : * pass, either.
3777 : : */
191 akorotkov@postgresql 3778 [ + + ]:GNC 43197 : if (requiredSameDir)
6218 tgl@sss.pgh.pa.us 3779 :GBC 27 : *continuescan = false;
3780 : :
3781 : : /*
3782 : : * In any case, this indextuple doesn't match the qual.
3783 : : */
1849 pg@bowt.ie 3784 :CBC 43197 : return false;
3785 : : }
3786 : :
8664 tgl@sss.pgh.pa.us 3787 [ + + ]: 31955016 : if (isNull)
3788 : : {
4547 3789 [ - + ]: 123 : if (key->sk_flags & SK_BT_NULLS_FIRST)
3790 : : {
3791 : : /*
3792 : : * Since NULLs are sorted before non-NULLs, we know we have
3793 : : * reached the lower limit of the range of values for this
3794 : : * index attr. On a backward scan, we can stop if this qual
3795 : : * is one of the "must match" subset. We can stop regardless
3796 : : * of whether the qual is > or <, so long as it's required,
3797 : : * because it's not possible for any future tuples to pass. On
3798 : : * a forward scan, however, we must keep going, because we may
3799 : : * have initially positioned to the start of the index.
3800 : : * (_bt_advance_array_keys also relies on this behavior during
3801 : : * forward scans.)
3802 : : */
4547 tgl@sss.pgh.pa.us 3803 [ # # # # ]:UBC 0 : if ((key->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)) &&
3804 : : ScanDirectionIsBackward(dir))
3805 : 0 : *continuescan = false;
3806 : : }
3807 : : else
3808 : : {
3809 : : /*
3810 : : * Since NULLs are sorted after non-NULLs, we know we have
3811 : : * reached the upper limit of the range of values for this
3812 : : * index attr. On a forward scan, we can stop if this qual is
3813 : : * one of the "must match" subset. We can stop regardless of
3814 : : * whether the qual is > or <, so long as it's required,
3815 : : * because it's not possible for any future tuples to pass. On
3816 : : * a backward scan, however, we must keep going, because we
3817 : : * may have initially positioned to the end of the index.
3818 : : * (_bt_advance_array_keys also relies on this behavior during
3819 : : * backward scans.)
3820 : : */
4547 tgl@sss.pgh.pa.us 3821 [ + + + - ]:CBC 123 : if ((key->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)) &&
3822 : : ScanDirectionIsForward(dir))
3823 : 78 : *continuescan = false;
3824 : : }
3825 : :
3826 : : /*
3827 : : * In any case, this indextuple doesn't match the qual.
3828 : : */
1849 pg@bowt.ie 3829 : 123 : return false;
3830 : : }
3831 : :
3832 : : /*
3833 : : * Apply the key-checking function, though only if we must.
3834 : : *
3835 : : * When a key is required in the opposite-of-scan direction _only_,
3836 : : * then it must already be satisfied if firstmatch=true indicates that
3837 : : * an earlier tuple from this same page satisfied it earlier on.
3838 : : */
8 pg@bowt.ie 3839 [ + + + + ]:GNC 31954893 : if (!(requiredOppositeDirOnly && firstmatch) &&
3840 [ + + ]: 30344834 : !DatumGetBool(FunctionCall2Coll(&key->sk_func, key->sk_collation,
3841 : : datum, key->sk_argument)))
3842 : : {
3843 : : /*
3844 : : * Tuple fails this qual. If it's a required qual for the current
3845 : : * scan direction, then we can conclude no further tuples will
3846 : : * pass, either.
3847 : : *
3848 : : * Note: because we stop the scan as soon as any required equality
3849 : : * qual fails, it is critical that equality quals be used for the
3850 : : * initial positioning in _bt_first() when they are available. See
3851 : : * comments in _bt_first().
3852 : : */
191 akorotkov@postgresql 3853 [ + + ]: 9238429 : if (requiredSameDir)
6656 tgl@sss.pgh.pa.us 3854 :CBC 8908231 : *continuescan = false;
3855 : :
3856 : : /*
3857 : : * If this is a non-required equality-type array key, the tuple
3858 : : * needs to be checked against every possible array key. Handle
3859 : : * this by "advancing" the scan key's array to a matching value
3860 : : * (if we're successful then the tuple might match the qual).
3861 : : */
8 pg@bowt.ie 3862 [ + + ]:GNC 330198 : else if (advancenonrequired &&
3863 [ + + ]: 153 : key->sk_strategy == BTEqualStrategyNumber &&
3864 [ + - ]: 120 : (key->sk_flags & SK_SEARCHARRAY))
3865 : 120 : return _bt_advance_array_keys(scan, NULL, tuple, tupnatts,
3866 : : tupdesc, *ikey, false);
3867 : :
3868 : : /*
3869 : : * This indextuple doesn't match the qual.
3870 : : */
1849 pg@bowt.ie 3871 :CBC 9238309 : return false;
3872 : : }
3873 : : }
3874 : :
3875 : : /* If we get here, the tuple passes all index quals. */
3876 : 37524983 : return true;
3877 : : }
3878 : :
3879 : : /*
3880 : : * Test whether an indextuple satisfies a row-comparison scan condition.
3881 : : *
3882 : : * Return true if so, false if not. If not, also clear *continuescan if
3883 : : * it's not possible for any future tuples in the current scan direction
3884 : : * to pass the qual.
3885 : : *
3886 : : * This is a subroutine for _bt_checkkeys/_bt_check_compare.
3887 : : */
3888 : : static bool
3889 : 1962 : _bt_check_rowcompare(ScanKey skey, IndexTuple tuple, int tupnatts,
3890 : : TupleDesc tupdesc, ScanDirection dir, bool *continuescan)
3891 : : {
6654 tgl@sss.pgh.pa.us 3892 : 1962 : ScanKey subkey = (ScanKey) DatumGetPointer(skey->sk_argument);
3893 : 1962 : int32 cmpresult = 0;
3894 : : bool result;
3895 : :
3896 : : /* First subkey should be same as the header says */
3897 [ + - ]: 1962 : Assert(subkey->sk_attno == skey->sk_attno);
3898 : :
3899 : : /* Loop over columns of the row condition */
3900 : : for (;;)
3901 : 144 : {
3902 : : Datum datum;
3903 : : bool isNull;
3904 : :
3905 [ - + ]: 2106 : Assert(subkey->sk_flags & SK_ROW_MEMBER);
3906 : :
1849 pg@bowt.ie 3907 [ + + ]: 2106 : if (subkey->sk_attno > tupnatts)
3908 : : {
3909 : : /*
3910 : : * This attribute is truncated (must be high key). The value for
3911 : : * this attribute in the first non-pivot tuple on the page to the
3912 : : * right could be any possible value. Assume that truncated
3913 : : * attribute passes the qual.
3914 : : */
1509 3915 [ - + ]: 6 : Assert(BTreeTupleIsPivot(tuple));
1849 3916 : 6 : cmpresult = 0;
1841 3917 [ + - ]: 6 : if (subkey->sk_flags & SK_ROW_END)
3918 : 6 : break;
1841 pg@bowt.ie 3919 :UBC 0 : subkey++;
1849 3920 : 0 : continue;
3921 : : }
3922 : :
6654 tgl@sss.pgh.pa.us 3923 :CBC 2100 : datum = index_getattr(tuple,
3924 : 2100 : subkey->sk_attno,
3925 : : tupdesc,
3926 : : &isNull);
3927 : :
3928 [ + + ]: 2100 : if (isNull)
3929 : : {
4547 3930 [ - + ]: 48 : if (subkey->sk_flags & SK_BT_NULLS_FIRST)
3931 : : {
3932 : : /*
3933 : : * Since NULLs are sorted before non-NULLs, we know we have
3934 : : * reached the lower limit of the range of values for this
3935 : : * index attr. On a backward scan, we can stop if this qual
3936 : : * is one of the "must match" subset. We can stop regardless
3937 : : * of whether the qual is > or <, so long as it's required,
3938 : : * because it's not possible for any future tuples to pass. On
3939 : : * a forward scan, however, we must keep going, because we may
3940 : : * have initially positioned to the start of the index.
3941 : : * (_bt_advance_array_keys also relies on this behavior during
3942 : : * forward scans.)
3943 : : */
4547 tgl@sss.pgh.pa.us 3944 [ # # # # ]:UBC 0 : if ((subkey->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)) &&
3945 : : ScanDirectionIsBackward(dir))
3946 : 0 : *continuescan = false;
3947 : : }
3948 : : else
3949 : : {
3950 : : /*
3951 : : * Since NULLs are sorted after non-NULLs, we know we have
3952 : : * reached the upper limit of the range of values for this
3953 : : * index attr. On a forward scan, we can stop if this qual is
3954 : : * one of the "must match" subset. We can stop regardless of
3955 : : * whether the qual is > or <, so long as it's required,
3956 : : * because it's not possible for any future tuples to pass. On
3957 : : * a backward scan, however, we must keep going, because we
3958 : : * may have initially positioned to the end of the index.
3959 : : * (_bt_advance_array_keys also relies on this behavior during
3960 : : * backward scans.)
3961 : : */
4547 tgl@sss.pgh.pa.us 3962 [ - + - - ]:CBC 48 : if ((subkey->sk_flags & (SK_BT_REQFWD | SK_BT_REQBKWD)) &&
3963 : : ScanDirectionIsForward(dir))
4547 tgl@sss.pgh.pa.us 3964 :UBC 0 : *continuescan = false;
3965 : : }
3966 : :
3967 : : /*
3968 : : * In any case, this indextuple doesn't match the qual.
3969 : : */
6654 tgl@sss.pgh.pa.us 3970 :CBC 48 : return false;
3971 : : }
3972 : :
3973 [ - + ]: 2052 : if (subkey->sk_flags & SK_ISNULL)
3974 : : {
3975 : : /*
3976 : : * Unlike the simple-scankey case, this isn't a disallowed case.
3977 : : * But it can never match. If all the earlier row comparison
3978 : : * columns are required for the scan direction, we can stop the
3979 : : * scan, because there can't be another tuple that will succeed.
3980 : : */
6654 tgl@sss.pgh.pa.us 3981 [ # # ]:UBC 0 : if (subkey != (ScanKey) DatumGetPointer(skey->sk_argument))
3982 : 0 : subkey--;
3983 [ # # # # ]: 0 : if ((subkey->sk_flags & SK_BT_REQFWD) &&
3984 : : ScanDirectionIsForward(dir))
3985 : 0 : *continuescan = false;
3986 [ # # # # ]: 0 : else if ((subkey->sk_flags & SK_BT_REQBKWD) &&
3987 : : ScanDirectionIsBackward(dir))
3988 : 0 : *continuescan = false;
3989 : 0 : return false;
3990 : : }
3991 : :
3992 : : /* Perform the test --- three-way comparison not bool operator */
4751 tgl@sss.pgh.pa.us 3993 :CBC 2052 : cmpresult = DatumGetInt32(FunctionCall2Coll(&subkey->sk_func,
3994 : : subkey->sk_collation,
3995 : : datum,
3996 : : subkey->sk_argument));
3997 : :
6305 3998 [ - + ]: 2052 : if (subkey->sk_flags & SK_BT_DESC)
2018 tgl@sss.pgh.pa.us 3999 [ # # ]:UBC 0 : INVERT_COMPARE_RESULT(cmpresult);
4000 : :
4001 : : /* Done comparing if unequal, else advance to next column */
6654 tgl@sss.pgh.pa.us 4002 [ + + ]:CBC 2052 : if (cmpresult != 0)
4003 : 1908 : break;
4004 : :
4005 [ - + ]: 144 : if (subkey->sk_flags & SK_ROW_END)
6654 tgl@sss.pgh.pa.us 4006 :UBC 0 : break;
6654 tgl@sss.pgh.pa.us 4007 :CBC 144 : subkey++;
4008 : : }
4009 : :
4010 : : /*
4011 : : * At this point cmpresult indicates the overall result of the row
4012 : : * comparison, and subkey points to the deciding column (or the last
4013 : : * column if the result is "=").
4014 : : */
4015 [ - + + + : 1914 : switch (subkey->sk_strategy)
- ]
4016 : : {
4017 : : /* EQ and NE cases aren't allowed here */
6654 tgl@sss.pgh.pa.us 4018 :UBC 0 : case BTLessStrategyNumber:
4019 : 0 : result = (cmpresult < 0);
4020 : 0 : break;
6654 tgl@sss.pgh.pa.us 4021 :CBC 1584 : case BTLessEqualStrategyNumber:
4022 : 1584 : result = (cmpresult <= 0);
4023 : 1584 : break;
4024 : 234 : case BTGreaterEqualStrategyNumber:
4025 : 234 : result = (cmpresult >= 0);
4026 : 234 : break;
4027 : 96 : case BTGreaterStrategyNumber:
4028 : 96 : result = (cmpresult > 0);
4029 : 96 : break;
6654 tgl@sss.pgh.pa.us 4030 :UBC 0 : default:
4031 [ # # ]: 0 : elog(ERROR, "unrecognized RowCompareType: %d",
4032 : : (int) subkey->sk_strategy);
4033 : : result = 0; /* keep compiler quiet */
4034 : : break;
4035 : : }
4036 : :
6654 tgl@sss.pgh.pa.us 4037 [ + + ]:CBC 1914 : if (!result)
4038 : : {
4039 : : /*
4040 : : * Tuple fails this qual. If it's a required qual for the current
4041 : : * scan direction, then we can conclude no further tuples will pass,
4042 : : * either. Note we have to look at the deciding column, not
4043 : : * necessarily the first or last column of the row condition.
4044 : : */
4045 [ + - + - ]: 6 : if ((subkey->sk_flags & SK_BT_REQFWD) &&
4046 : : ScanDirectionIsForward(dir))
4047 : 6 : *continuescan = false;
6654 tgl@sss.pgh.pa.us 4048 [ # # # # ]:UBC 0 : else if ((subkey->sk_flags & SK_BT_REQBKWD) &&
4049 : : ScanDirectionIsBackward(dir))
4050 : 0 : *continuescan = false;
4051 : : }
4052 : :
6654 tgl@sss.pgh.pa.us 4053 :CBC 1914 : return result;
4054 : : }
4055 : :
4056 : : /*
4057 : : * Determine if a scan with array keys should skip over uninteresting tuples.
4058 : : *
4059 : : * This is a subroutine for _bt_checkkeys. Called when _bt_readpage's linear
4060 : : * search process (started after it finishes reading an initial group of
4061 : : * matching tuples, used to locate the start of the next group of tuples
4062 : : * matching the next set of required array keys) has already scanned an
4063 : : * excessive number of tuples whose key space is "between arrays".
4064 : : *
4065 : : * When we perform look ahead successfully, we'll sets pstate.skip, which
4066 : : * instructs _bt_readpage to skip ahead to that tuple next (could be past the
4067 : : * end of the scan's leaf page). Pages where the optimization is effective
4068 : : * will generally still need to skip several times. Each call here performs
4069 : : * only a single "look ahead" comparison of a later tuple, whose distance from
4070 : : * the current tuple's offset number is determined by applying heuristics.
4071 : : */
4072 : : static void
8 pg@bowt.ie 4073 :GNC 529 : _bt_checkkeys_look_ahead(IndexScanDesc scan, BTReadPageState *pstate,
4074 : : int tupnatts, TupleDesc tupdesc)
4075 : : {
4076 : 529 : ScanDirection dir = pstate->dir;
4077 : : OffsetNumber aheadoffnum;
4078 : : IndexTuple ahead;
4079 : :
4080 : : /* Avoid looking ahead when comparing the page high key */
4081 [ - + ]: 529 : if (pstate->offnum < pstate->minoff)
8 pg@bowt.ie 4082 :UNC 0 : return;
4083 : :
4084 : : /*
4085 : : * Don't look ahead when there aren't enough tuples remaining on the page
4086 : : * (in the current scan direction) for it to be worth our while
4087 : : */
8 pg@bowt.ie 4088 [ + + ]:GNC 529 : if (ScanDirectionIsForward(dir) &&
4089 [ + + ]: 526 : pstate->offnum >= pstate->maxoff - LOOK_AHEAD_DEFAULT_DISTANCE)
4090 : 1 : return;
4091 [ + + ]: 528 : else if (ScanDirectionIsBackward(dir) &&
4092 [ - + ]: 3 : pstate->offnum <= pstate->minoff + LOOK_AHEAD_DEFAULT_DISTANCE)
8 pg@bowt.ie 4093 :UNC 0 : return;
4094 : :
4095 : : /*
4096 : : * The look ahead distance starts small, and ramps up as each call here
4097 : : * allows _bt_readpage to skip over more tuples
4098 : : */
8 pg@bowt.ie 4099 [ + + ]:GNC 528 : if (!pstate->targetdistance)
4100 : 201 : pstate->targetdistance = LOOK_AHEAD_DEFAULT_DISTANCE;
4101 : : else
4102 : 327 : pstate->targetdistance *= 2;
4103 : :
4104 : : /* Don't read past the end (or before the start) of the page, though */
4105 [ + + ]: 528 : if (ScanDirectionIsForward(dir))
4106 : 525 : aheadoffnum = Min((int) pstate->maxoff,
4107 : : (int) pstate->offnum + pstate->targetdistance);
4108 : : else
4109 : 3 : aheadoffnum = Max((int) pstate->minoff,
4110 : : (int) pstate->offnum - pstate->targetdistance);
4111 : :
4112 : 528 : ahead = (IndexTuple) PageGetItem(pstate->page,
4113 : : PageGetItemId(pstate->page, aheadoffnum));
4114 [ + + ]: 528 : if (_bt_tuple_before_array_skeys(scan, dir, ahead, tupdesc, tupnatts,
4115 : : false, 0, NULL))
4116 : : {
4117 : : /*
4118 : : * Success -- instruct _bt_readpage to skip ahead to very next tuple
4119 : : * after the one we determined was still before the current array keys
4120 : : */
4121 [ + + ]: 275 : if (ScanDirectionIsForward(dir))
4122 : 272 : pstate->skip = aheadoffnum + 1;
4123 : : else
4124 : 3 : pstate->skip = aheadoffnum - 1;
4125 : : }
4126 : : else
4127 : : {
4128 : : /*
4129 : : * Failure -- "ahead" tuple is too far ahead (we were too aggresive).
4130 : : *
4131 : : * Reset the number of rechecks, and aggressively reduce the target
4132 : : * distance (we're much more aggressive here than we were when the
4133 : : * distance was initially ramped up).
4134 : : */
4135 : 253 : pstate->rechecks = 0;
4136 [ + + ]: 253 : pstate->targetdistance = Max(pstate->targetdistance / 8, 1);
4137 : : }
4138 : : }
4139 : :
4140 : : /*
4141 : : * _bt_killitems - set LP_DEAD state for items an indexscan caller has
4142 : : * told us were killed
4143 : : *
4144 : : * scan->opaque, referenced locally through so, contains information about the
4145 : : * current page and killed tuples thereon (generally, this should only be
4146 : : * called if so->numKilled > 0).
4147 : : *
4148 : : * The caller does not have a lock on the page and may or may not have the
4149 : : * page pinned in a buffer. Note that read-lock is sufficient for setting
4150 : : * LP_DEAD status (which is only a hint).
4151 : : *
4152 : : * We match items by heap TID before assuming they are the right ones to
4153 : : * delete. We cope with cases where items have moved right due to insertions.
4154 : : * If an item has moved off the current page due to a split, we'll fail to
4155 : : * find it and do nothing (this is not an error case --- we assume the item
4156 : : * will eventually get marked in a future indexscan).
4157 : : *
4158 : : * Note that if we hold a pin on the target page continuously from initially
4159 : : * reading the items until applying this function, VACUUM cannot have deleted
4160 : : * any items from the page, and so there is no need to search left from the
4161 : : * recorded offset. (This observation also guarantees that the item is still
4162 : : * the right one to delete, which might otherwise be questionable since heap
4163 : : * TIDs can get recycled.) This holds true even if the page has been modified
4164 : : * by inserts and page splits, so there is no need to consult the LSN.
4165 : : *
4166 : : * If the pin was released after reading the page, then we re-read it. If it
4167 : : * has been modified since we read it (as determined by the LSN), we dare not
4168 : : * flag any entries because it is possible that the old entry was vacuumed
4169 : : * away and the TID was re-used by a completely different heap tuple.
4170 : : */
4171 : : void
3308 kgrittn@postgresql.o 4172 :CBC 70253 : _bt_killitems(IndexScanDesc scan)
4173 : : {
6552 tgl@sss.pgh.pa.us 4174 : 70253 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
4175 : : Page page;
4176 : : BTPageOpaque opaque;
4177 : : OffsetNumber minoff;
4178 : : OffsetNumber maxoff;
4179 : : int i;
3308 kgrittn@postgresql.o 4180 : 70253 : int numKilled = so->numKilled;
6552 tgl@sss.pgh.pa.us 4181 : 70253 : bool killedsomething = false;
4182 : : bool droppedpin PG_USED_FOR_ASSERTS_ONLY;
4183 : :
3308 kgrittn@postgresql.o 4184 [ - + - - : 70253 : Assert(BTScanPosIsValid(so->currPos));
- + ]
4185 : :
4186 : : /*
4187 : : * Always reset the scan state, so we don't look for same items on other
4188 : : * pages.
4189 : : */
4190 : 70253 : so->numKilled = 0;
4191 : :
4192 [ - + - - : 70253 : if (BTScanPosIsPinned(so->currPos))
+ + ]
4193 : : {
4194 : : /*
4195 : : * We have held the pin on this page since we read the index tuples,
4196 : : * so all we need to do is lock it. The pin will have prevented
4197 : : * re-use of any TID on the page, so there is no need to check the
4198 : : * LSN.
4199 : : */
1469 pg@bowt.ie 4200 : 17307 : droppedpin = false;
1363 4201 : 17307 : _bt_lockbuf(scan->indexRelation, so->currPos.buf, BT_READ);
4202 : :
2916 kgrittn@postgresql.o 4203 : 17307 : page = BufferGetPage(so->currPos.buf);
4204 : : }
4205 : : else
4206 : : {
4207 : : Buffer buf;
4208 : :
1469 pg@bowt.ie 4209 : 52946 : droppedpin = true;
4210 : : /* Attempt to re-read the buffer, getting pin and lock. */
309 4211 : 52946 : buf = _bt_getbuf(scan->indexRelation, so->currPos.currPage, BT_READ);
4212 : :
2916 kgrittn@postgresql.o 4213 : 52946 : page = BufferGetPage(buf);
2287 alvherre@alvh.no-ip. 4214 [ + + ]: 52946 : if (BufferGetLSNAtomic(buf) == so->currPos.lsn)
3308 kgrittn@postgresql.o 4215 : 52861 : so->currPos.buf = buf;
4216 : : else
4217 : : {
4218 : : /* Modified while not pinned means hinting is not safe. */
4219 : 85 : _bt_relbuf(scan->indexRelation, buf);
4220 : 85 : return;
4221 : : }
4222 : : }
4223 : :
744 michael@paquier.xyz 4224 : 70168 : opaque = BTPageGetOpaque(page);
6552 tgl@sss.pgh.pa.us 4225 [ + + ]: 70168 : minoff = P_FIRSTDATAKEY(opaque);
4226 : 70168 : maxoff = PageGetMaxOffsetNumber(page);
4227 : :
3308 kgrittn@postgresql.o 4228 [ + + ]: 261200 : for (i = 0; i < numKilled; i++)
4229 : : {
6402 bruce@momjian.us 4230 : 191032 : int itemIndex = so->killedItems[i];
4231 : 191032 : BTScanPosItem *kitem = &so->currPos.items[itemIndex];
4232 : 191032 : OffsetNumber offnum = kitem->indexOffset;
4233 : :
6552 tgl@sss.pgh.pa.us 4234 [ + - - + ]: 191032 : Assert(itemIndex >= so->currPos.firstItem &&
4235 : : itemIndex <= so->currPos.lastItem);
4236 [ - + ]: 191032 : if (offnum < minoff)
6552 tgl@sss.pgh.pa.us 4237 :UBC 0 : continue; /* pure paranoia */
6552 tgl@sss.pgh.pa.us 4238 [ + + ]:CBC 2436936 : while (offnum <= maxoff)
4239 : : {
4240 : 2417128 : ItemId iid = PageGetItemId(page, offnum);
4241 : 2417128 : IndexTuple ituple = (IndexTuple) PageGetItem(page, iid);
1509 pg@bowt.ie 4242 : 2417128 : bool killtuple = false;
4243 : :
4244 [ + + ]: 2417128 : if (BTreeTupleIsPosting(ituple))
4245 : : {
4246 : 594306 : int pi = i + 1;
4247 : 594306 : int nposting = BTreeTupleGetNPosting(ituple);
4248 : : int j;
4249 : :
4250 : : /*
4251 : : * We rely on the convention that heap TIDs in the scanpos
4252 : : * items array are stored in ascending heap TID order for a
4253 : : * group of TIDs that originally came from a posting list
4254 : : * tuple. This convention even applies during backwards
4255 : : * scans, where returning the TIDs in descending order might
4256 : : * seem more natural. This is about effectiveness, not
4257 : : * correctness.
4258 : : *
4259 : : * Note that the page may have been modified in almost any way
4260 : : * since we first read it (in the !droppedpin case), so it's
4261 : : * possible that this posting list tuple wasn't a posting list
4262 : : * tuple when we first encountered its heap TIDs.
4263 : : */
4264 [ + + ]: 612641 : for (j = 0; j < nposting; j++)
4265 : : {
4266 : 611989 : ItemPointer item = BTreeTupleGetPostingN(ituple, j);
4267 : :
4268 [ + + ]: 611989 : if (!ItemPointerEquals(item, &kitem->heapTid))
4269 : 593654 : break; /* out of posting list loop */
4270 : :
4271 : : /*
4272 : : * kitem must have matching offnum when heap TIDs match,
4273 : : * though only in the common case where the page can't
4274 : : * have been concurrently modified
4275 : : */
1469 4276 [ - + - - ]: 18335 : Assert(kitem->indexOffset == offnum || !droppedpin);
4277 : :
4278 : : /*
4279 : : * Read-ahead to later kitems here.
4280 : : *
4281 : : * We rely on the assumption that not advancing kitem here
4282 : : * will prevent us from considering the posting list tuple
4283 : : * fully dead by not matching its next heap TID in next
4284 : : * loop iteration.
4285 : : *
4286 : : * If, on the other hand, this is the final heap TID in
4287 : : * the posting list tuple, then tuple gets killed
4288 : : * regardless (i.e. we handle the case where the last
4289 : : * kitem is also the last heap TID in the last index tuple
4290 : : * correctly -- posting tuple still gets killed).
4291 : : */
1509 4292 [ + + ]: 18335 : if (pi < numKilled)
4293 : 8393 : kitem = &so->currPos.items[so->killedItems[pi++]];
4294 : : }
4295 : :
4296 : : /*
4297 : : * Don't bother advancing the outermost loop's int iterator to
4298 : : * avoid processing killed items that relate to the same
4299 : : * offnum/posting list tuple. This micro-optimization hardly
4300 : : * seems worth it. (Further iterations of the outermost loop
4301 : : * will fail to match on this same posting list's first heap
4302 : : * TID instead, so we'll advance to the next offnum/index
4303 : : * tuple pretty quickly.)
4304 : : */
4305 [ + + ]: 594306 : if (j == nposting)
4306 : 652 : killtuple = true;
4307 : : }
4308 [ + + ]: 1822822 : else if (ItemPointerEquals(&ituple->t_tid, &kitem->heapTid))
4309 : 170828 : killtuple = true;
4310 : :
4311 : : /*
4312 : : * Mark index item as dead, if it isn't already. Since this
4313 : : * happens while holding a buffer lock possibly in shared mode,
4314 : : * it's possible that multiple processes attempt to do this
4315 : : * simultaneously, leading to multiple full-page images being sent
4316 : : * to WAL (if wal_log_hints or data checksums are enabled), which
4317 : : * is undesirable.
4318 : : */
1430 alvherre@alvh.no-ip. 4319 [ + + + + ]: 2417128 : if (killtuple && !ItemIdIsDead(iid))
4320 : : {
4321 : : /* found the item/all posting list items */
6059 tgl@sss.pgh.pa.us 4322 : 171224 : ItemIdMarkDead(iid);
6552 4323 : 171224 : killedsomething = true;
4324 : 171224 : break; /* out of inner search loop */
4325 : : }
4326 : 2245904 : offnum = OffsetNumberNext(offnum);
4327 : : }
4328 : : }
4329 : :
4330 : : /*
4331 : : * Since this can be redone later if needed, mark as dirty hint.
4332 : : *
4333 : : * Whenever we mark anything LP_DEAD, we also set the page's
4334 : : * BTP_HAS_GARBAGE flag, which is likewise just a hint. (Note that we
4335 : : * only rely on the page-level flag in !heapkeyspace indexes.)
4336 : : */
4337 [ + + ]: 70168 : if (killedsomething)
4338 : : {
6473 4339 : 59720 : opaque->btpo_flags |= BTP_HAS_GARBAGE;
3954 jdavis@postgresql.or 4340 : 59720 : MarkBufferDirtyHint(so->currPos.buf, true);
4341 : : }
4342 : :
1363 pg@bowt.ie 4343 : 70168 : _bt_unlockbuf(scan->indexRelation, so->currPos.buf);
4344 : : }
4345 : :
4346 : :
4347 : : /*
4348 : : * The following routines manage a shared-memory area in which we track
4349 : : * assignment of "vacuum cycle IDs" to currently-active btree vacuuming
4350 : : * operations. There is a single counter which increments each time we
4351 : : * start a vacuum to assign it a cycle ID. Since multiple vacuums could
4352 : : * be active concurrently, we have to track the cycle ID for each active
4353 : : * vacuum; this requires at most MaxBackends entries (usually far fewer).
4354 : : * We assume at most one vacuum can be active for a given index.
4355 : : *
4356 : : * Access to the shared memory area is controlled by BtreeVacuumLock.
4357 : : * In principle we could use a separate lmgr locktag for each index,
4358 : : * but a single LWLock is much cheaper, and given the short time that
4359 : : * the lock is ever held, the concurrency hit should be minimal.
4360 : : */
4361 : :
4362 : : typedef struct BTOneVacInfo
4363 : : {
4364 : : LockRelId relid; /* global identifier of an index */
4365 : : BTCycleId cycleid; /* cycle ID for its active VACUUM */
4366 : : } BTOneVacInfo;
4367 : :
4368 : : typedef struct BTVacInfo
4369 : : {
4370 : : BTCycleId cycle_ctr; /* cycle ID most recently assigned */
4371 : : int num_vacuums; /* number of currently active VACUUMs */
4372 : : int max_vacuums; /* allocated length of vacuums[] array */
4373 : : BTOneVacInfo vacuums[FLEXIBLE_ARRAY_MEMBER];
4374 : : } BTVacInfo;
4375 : :
4376 : : static BTVacInfo *btvacinfo;
4377 : :
4378 : :
4379 : : /*
4380 : : * _bt_vacuum_cycleid --- get the active vacuum cycle ID for an index,
4381 : : * or zero if there is no active VACUUM
4382 : : *
4383 : : * Note: for correct interlocking, the caller must already hold pin and
4384 : : * exclusive lock on each buffer it will store the cycle ID into. This
4385 : : * ensures that even if a VACUUM starts immediately afterwards, it cannot
4386 : : * process those pages until the page split is complete.
4387 : : */
4388 : : BTCycleId
6551 tgl@sss.pgh.pa.us 4389 : 10590 : _bt_vacuum_cycleid(Relation rel)
4390 : : {
4391 : 10590 : BTCycleId result = 0;
4392 : : int i;
4393 : :
4394 : : /* Share lock is enough since this is a read-only operation */
4395 : 10590 : LWLockAcquire(BtreeVacuumLock, LW_SHARED);
4396 : :
4397 [ + + ]: 10601 : for (i = 0; i < btvacinfo->num_vacuums; i++)
4398 : : {
4399 : 12 : BTOneVacInfo *vac = &btvacinfo->vacuums[i];
4400 : :
4401 [ + + ]: 12 : if (vac->relid.relId == rel->rd_lockInfo.lockRelId.relId &&
4402 [ + - ]: 1 : vac->relid.dbId == rel->rd_lockInfo.lockRelId.dbId)
4403 : : {
4404 : 1 : result = vac->cycleid;
4405 : 1 : break;
4406 : : }
4407 : : }
4408 : :
4409 : 10590 : LWLockRelease(BtreeVacuumLock);
4410 : 10590 : return result;
4411 : : }
4412 : :
4413 : : /*
4414 : : * _bt_start_vacuum --- assign a cycle ID to a just-starting VACUUM operation
4415 : : *
4416 : : * Note: the caller must guarantee that it will eventually call
4417 : : * _bt_end_vacuum, else we'll permanently leak an array slot. To ensure
4418 : : * that this happens even in elog(FATAL) scenarios, the appropriate coding
4419 : : * is not just a PG_TRY, but
4420 : : * PG_ENSURE_ERROR_CLEANUP(_bt_end_vacuum_callback, PointerGetDatum(rel))
4421 : : */
4422 : : BTCycleId
4423 : 1256 : _bt_start_vacuum(Relation rel)
4424 : : {
4425 : : BTCycleId result;
4426 : : int i;
4427 : : BTOneVacInfo *vac;
4428 : :
4429 : 1256 : LWLockAcquire(BtreeVacuumLock, LW_EXCLUSIVE);
4430 : :
4431 : : /*
4432 : : * Assign the next cycle ID, being careful to avoid zero as well as the
4433 : : * reserved high values.
4434 : : */
6215 4435 : 1256 : result = ++(btvacinfo->cycle_ctr);
4436 [ + - - + ]: 1256 : if (result == 0 || result > MAX_BT_CYCLE_ID)
6215 tgl@sss.pgh.pa.us 4437 :UBC 0 : result = btvacinfo->cycle_ctr = 1;
4438 : :
4439 : : /* Let's just make sure there's no entry already for this index */
6551 tgl@sss.pgh.pa.us 4440 [ - + ]:CBC 1256 : for (i = 0; i < btvacinfo->num_vacuums; i++)
4441 : : {
6551 tgl@sss.pgh.pa.us 4442 :LBC (6) : vac = &btvacinfo->vacuums[i];
4443 [ # # ]: (6) : if (vac->relid.relId == rel->rd_lockInfo.lockRelId.relId &&
6551 tgl@sss.pgh.pa.us 4444 [ # # ]:UBC 0 : vac->relid.dbId == rel->rd_lockInfo.lockRelId.dbId)
4445 : : {
4446 : : /*
4447 : : * Unlike most places in the backend, we have to explicitly
4448 : : * release our LWLock before throwing an error. This is because
4449 : : * we expect _bt_end_vacuum() to be called before transaction
4450 : : * abort cleanup can run to release LWLocks.
4451 : : */
6225 4452 : 0 : LWLockRelease(BtreeVacuumLock);
6551 4453 [ # # ]: 0 : elog(ERROR, "multiple active vacuums for index \"%s\"",
4454 : : RelationGetRelationName(rel));
4455 : : }
4456 : : }
4457 : :
4458 : : /* OK, add an entry */
6551 tgl@sss.pgh.pa.us 4459 [ - + ]:CBC 1256 : if (btvacinfo->num_vacuums >= btvacinfo->max_vacuums)
4460 : : {
6225 tgl@sss.pgh.pa.us 4461 :UBC 0 : LWLockRelease(BtreeVacuumLock);
6551 4462 [ # # ]: 0 : elog(ERROR, "out of btvacinfo slots");
4463 : : }
6551 tgl@sss.pgh.pa.us 4464 :CBC 1256 : vac = &btvacinfo->vacuums[btvacinfo->num_vacuums];
4465 : 1256 : vac->relid = rel->rd_lockInfo.lockRelId;
4466 : 1256 : vac->cycleid = result;
4467 : 1256 : btvacinfo->num_vacuums++;
4468 : :
4469 : 1256 : LWLockRelease(BtreeVacuumLock);
4470 : 1256 : return result;
4471 : : }
4472 : :
4473 : : /*
4474 : : * _bt_end_vacuum --- mark a btree VACUUM operation as done
4475 : : *
4476 : : * Note: this is deliberately coded not to complain if no entry is found;
4477 : : * this allows the caller to put PG_TRY around the start_vacuum operation.
4478 : : */
4479 : : void
4480 : 1256 : _bt_end_vacuum(Relation rel)
4481 : : {
4482 : : int i;
4483 : :
4484 : 1256 : LWLockAcquire(BtreeVacuumLock, LW_EXCLUSIVE);
4485 : :
4486 : : /* Find the array entry */
4487 [ + - ]: 1256 : for (i = 0; i < btvacinfo->num_vacuums; i++)
4488 : : {
4489 : 1256 : BTOneVacInfo *vac = &btvacinfo->vacuums[i];
4490 : :
4491 [ + - ]: 1256 : if (vac->relid.relId == rel->rd_lockInfo.lockRelId.relId &&
4492 [ + - ]: 1256 : vac->relid.dbId == rel->rd_lockInfo.lockRelId.dbId)
4493 : : {
4494 : : /* Remove it by shifting down the last entry */
4495 : 1256 : *vac = btvacinfo->vacuums[btvacinfo->num_vacuums - 1];
4496 : 1256 : btvacinfo->num_vacuums--;
4497 : 1256 : break;
4498 : : }
4499 : : }
4500 : :
4501 : 1256 : LWLockRelease(BtreeVacuumLock);
4502 : 1256 : }
4503 : :
4504 : : /*
4505 : : * _bt_end_vacuum wrapped as an on_shmem_exit callback function
4506 : : */
4507 : : void
5842 tgl@sss.pgh.pa.us 4508 :UBC 0 : _bt_end_vacuum_callback(int code, Datum arg)
4509 : : {
4510 : 0 : _bt_end_vacuum((Relation) DatumGetPointer(arg));
4511 : 0 : }
4512 : :
4513 : : /*
4514 : : * BTreeShmemSize --- report amount of shared memory space needed
4515 : : */
4516 : : Size
6551 tgl@sss.pgh.pa.us 4517 :CBC 2577 : BTreeShmemSize(void)
4518 : : {
4519 : : Size size;
4520 : :
3341 4521 : 2577 : size = offsetof(BTVacInfo, vacuums);
733 rhaas@postgresql.org 4522 : 2577 : size = add_size(size, mul_size(MaxBackends, sizeof(BTOneVacInfo)));
6551 tgl@sss.pgh.pa.us 4523 : 2577 : return size;
4524 : : }
4525 : :
4526 : : /*
4527 : : * BTreeShmemInit --- initialize this module's shared memory
4528 : : */
4529 : : void
4530 : 898 : BTreeShmemInit(void)
4531 : : {
4532 : : bool found;
4533 : :
4534 : 898 : btvacinfo = (BTVacInfo *) ShmemInitStruct("BTree Vacuum State",
4535 : : BTreeShmemSize(),
4536 : : &found);
4537 : :
4538 [ + - ]: 898 : if (!IsUnderPostmaster)
4539 : : {
4540 : : /* Initialize shared memory area */
4541 [ - + ]: 898 : Assert(!found);
4542 : :
4543 : : /*
4544 : : * It doesn't really matter what the cycle counter starts at, but
4545 : : * having it always start the same doesn't seem good. Seed with
4546 : : * low-order bits of time() instead.
4547 : : */
4548 : 898 : btvacinfo->cycle_ctr = (BTCycleId) time(NULL);
4549 : :
4550 : 898 : btvacinfo->num_vacuums = 0;
733 rhaas@postgresql.org 4551 : 898 : btvacinfo->max_vacuums = MaxBackends;
4552 : : }
4553 : : else
6551 tgl@sss.pgh.pa.us 4554 [ # # ]:UBC 0 : Assert(found);
6551 tgl@sss.pgh.pa.us 4555 :CBC 898 : }
4556 : :
4557 : : bytea *
3010 4558 : 138 : btoptions(Datum reloptions, bool validate)
4559 : : {
4560 : : static const relopt_parse_elt tab[] = {
4561 : : {"fillfactor", RELOPT_TYPE_INT, offsetof(BTOptions, fillfactor)},
4562 : : {"vacuum_cleanup_index_scale_factor", RELOPT_TYPE_REAL,
4563 : : offsetof(BTOptions, vacuum_cleanup_index_scale_factor)},
4564 : : {"deduplicate_items", RELOPT_TYPE_BOOL,
4565 : : offsetof(BTOptions, deduplicate_items)}
4566 : : };
4567 : :
1602 michael@paquier.xyz 4568 : 138 : return (bytea *) build_reloptions(reloptions, validate,
4569 : : RELOPT_KIND_BTREE,
4570 : : sizeof(BTOptions),
4571 : : tab, lengthof(tab));
4572 : : }
4573 : :
4574 : : /*
4575 : : * btproperty() -- Check boolean properties of indexes.
4576 : : *
4577 : : * This is optional, but handling AMPROP_RETURNABLE here saves opening the rel
4578 : : * to call btcanreturn.
4579 : : */
4580 : : bool
2801 tgl@sss.pgh.pa.us 4581 : 378 : btproperty(Oid index_oid, int attno,
4582 : : IndexAMProperty prop, const char *propname,
4583 : : bool *res, bool *isnull)
4584 : : {
4585 [ + + ]: 378 : switch (prop)
4586 : : {
4587 : 21 : case AMPROP_RETURNABLE:
4588 : : /* answer only for columns, not AM or whole index */
4589 [ + + ]: 21 : if (attno == 0)
4590 : 6 : return false;
4591 : : /* otherwise, btree can always return data */
4592 : 15 : *res = true;
4593 : 15 : return true;
4594 : :
4595 : 357 : default:
4596 : 357 : return false; /* punt to generic code */
4597 : : }
4598 : : }
4599 : :
4600 : : /*
4601 : : * btbuildphasename() -- Return name of index build phase.
4602 : : */
4603 : : char *
1839 alvherre@alvh.no-ip. 4604 :UBC 0 : btbuildphasename(int64 phasenum)
4605 : : {
4606 [ # # # # : 0 : switch (phasenum)
# # ]
4607 : : {
4608 : 0 : case PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE:
4609 : 0 : return "initializing";
4610 : 0 : case PROGRESS_BTREE_PHASE_INDEXBUILD_TABLESCAN:
4611 : 0 : return "scanning table";
4612 : 0 : case PROGRESS_BTREE_PHASE_PERFORMSORT_1:
4613 : 0 : return "sorting live tuples";
4614 : 0 : case PROGRESS_BTREE_PHASE_PERFORMSORT_2:
4615 : 0 : return "sorting dead tuples";
4616 : 0 : case PROGRESS_BTREE_PHASE_LEAF_LOAD:
4617 : 0 : return "loading tuples in tree";
4618 : 0 : default:
4619 : 0 : return NULL;
4620 : : }
4621 : : }
4622 : :
4623 : : /*
4624 : : * _bt_truncate() -- create tuple without unneeded suffix attributes.
4625 : : *
4626 : : * Returns truncated pivot index tuple allocated in caller's memory context,
4627 : : * with key attributes copied from caller's firstright argument. If rel is
4628 : : * an INCLUDE index, non-key attributes will definitely be truncated away,
4629 : : * since they're not part of the key space. More aggressive suffix
4630 : : * truncation can take place when it's clear that the returned tuple does not
4631 : : * need one or more suffix key attributes. We only need to keep firstright
4632 : : * attributes up to and including the first non-lastleft-equal attribute.
4633 : : * Caller's insertion scankey is used to compare the tuples; the scankey's
4634 : : * argument values are not considered here.
4635 : : *
4636 : : * Note that returned tuple's t_tid offset will hold the number of attributes
4637 : : * present, so the original item pointer offset is not represented. Caller
4638 : : * should only change truncated tuple's downlink. Note also that truncated
4639 : : * key attributes are treated as containing "minus infinity" values by
4640 : : * _bt_compare().
4641 : : *
4642 : : * In the worst case (when a heap TID must be appended to distinguish lastleft
4643 : : * from firstright), the size of the returned tuple is the size of firstright
4644 : : * plus the size of an additional MAXALIGN()'d item pointer. This guarantee
4645 : : * is important, since callers need to stay under the 1/3 of a page
4646 : : * restriction on tuple size. If this routine is ever taught to truncate
4647 : : * within an attribute/datum, it will need to avoid returning an enlarged
4648 : : * tuple to caller when truncation + TOAST compression ends up enlarging the
4649 : : * final datum.
4650 : : */
4651 : : IndexTuple
1852 pg@bowt.ie 4652 :CBC 29103 : _bt_truncate(Relation rel, IndexTuple lastleft, IndexTuple firstright,
4653 : : BTScanInsert itup_key)
4654 : : {
4655 : 29103 : TupleDesc itupdesc = RelationGetDescr(rel);
4656 : 29103 : int16 nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
4657 : : int keepnatts;
4658 : : IndexTuple pivot;
4659 : : IndexTuple tidpivot;
4660 : : ItemPointer pivotheaptid;
4661 : : Size newsize;
4662 : :
4663 : : /*
4664 : : * We should only ever truncate non-pivot tuples from leaf pages. It's
4665 : : * never okay to truncate when splitting an internal page.
4666 : : */
1509 4667 [ + - - + ]: 29103 : Assert(!BTreeTupleIsPivot(lastleft) && !BTreeTupleIsPivot(firstright));
4668 : :
4669 : : /* Determine how many attributes must be kept in truncated tuple */
1852 4670 : 29103 : keepnatts = _bt_keep_natts(rel, lastleft, firstright, itup_key);
4671 : :
4672 : : #ifdef DEBUG_NO_TRUNCATE
4673 : : /* Force truncation to be ineffective for testing purposes */
4674 : : keepnatts = nkeyatts + 1;
4675 : : #endif
4676 : :
1476 4677 : 29103 : pivot = index_truncate_tuple(itupdesc, firstright,
4678 : : Min(keepnatts, nkeyatts));
4679 : :
4680 [ + + ]: 29103 : if (BTreeTupleIsPosting(pivot))
4681 : : {
4682 : : /*
4683 : : * index_truncate_tuple() just returns a straight copy of firstright
4684 : : * when it has no attributes to truncate. When that happens, we may
4685 : : * need to truncate away a posting list here instead.
4686 : : */
4687 [ + + - + ]: 624 : Assert(keepnatts == nkeyatts || keepnatts == nkeyatts + 1);
4688 [ - + ]: 624 : Assert(IndexRelationGetNumberOfAttributes(rel) == nkeyatts);
4689 : 624 : pivot->t_info &= ~INDEX_SIZE_MASK;
4690 : 624 : pivot->t_info |= MAXALIGN(BTreeTupleGetPostingOffset(firstright));
4691 : : }
4692 : :
4693 : : /*
4694 : : * If there is a distinguishing key attribute within pivot tuple, we're
4695 : : * done
4696 : : */
4697 [ + + ]: 29103 : if (keepnatts <= nkeyatts)
4698 : : {
1468 4699 : 28577 : BTreeTupleSetNAtts(pivot, keepnatts, false);
1476 4700 : 28577 : return pivot;
4701 : : }
4702 : :
4703 : : /*
4704 : : * We have to store a heap TID in the new pivot tuple, since no non-TID
4705 : : * key attribute value in firstright distinguishes the right side of the
4706 : : * split from the left side. nbtree conceptualizes this case as an
4707 : : * inability to truncate away any key attributes, since heap TID is
4708 : : * treated as just another key attribute (despite lacking a pg_attribute
4709 : : * entry).
4710 : : *
4711 : : * Use enlarged space that holds a copy of pivot. We need the extra space
4712 : : * to store a heap TID at the end (using the special pivot tuple
4713 : : * representation). Note that the original pivot already has firstright's
4714 : : * possible posting list/non-key attribute values removed at this point.
4715 : : */
4716 : 526 : newsize = MAXALIGN(IndexTupleSize(pivot)) + MAXALIGN(sizeof(ItemPointerData));
4717 : 526 : tidpivot = palloc0(newsize);
4718 : 526 : memcpy(tidpivot, pivot, MAXALIGN(IndexTupleSize(pivot)));
4719 : : /* Cannot leak memory here */
4720 : 526 : pfree(pivot);
4721 : :
4722 : : /*
4723 : : * Store all of firstright's key attribute values plus a tiebreaker heap
4724 : : * TID value in enlarged pivot tuple
4725 : : */
4726 : 526 : tidpivot->t_info &= ~INDEX_SIZE_MASK;
4727 : 526 : tidpivot->t_info |= newsize;
1468 4728 : 526 : BTreeTupleSetNAtts(tidpivot, nkeyatts, true);
1476 4729 : 526 : pivotheaptid = BTreeTupleGetHeapTID(tidpivot);
4730 : :
4731 : : /*
4732 : : * Lehman & Yao use lastleft as the leaf high key in all cases, but don't
4733 : : * consider suffix truncation. It seems like a good idea to follow that
4734 : : * example in cases where no truncation takes place -- use lastleft's heap
4735 : : * TID. (This is also the closest value to negative infinity that's
4736 : : * legally usable.)
4737 : : */
1509 4738 : 526 : ItemPointerCopy(BTreeTupleGetMaxHeapTID(lastleft), pivotheaptid);
4739 : :
4740 : : /*
4741 : : * We're done. Assert() that heap TID invariants hold before returning.
4742 : : *
4743 : : * Lehman and Yao require that the downlink to the right page, which is to
4744 : : * be inserted into the parent page in the second phase of a page split be
4745 : : * a strict lower bound on items on the right page, and a non-strict upper
4746 : : * bound for items on the left page. Assert that heap TIDs follow these
4747 : : * invariants, since a heap TID value is apparently needed as a
4748 : : * tiebreaker.
4749 : : */
4750 : : #ifndef DEBUG_NO_TRUNCATE
4751 [ - + ]: 526 : Assert(ItemPointerCompare(BTreeTupleGetMaxHeapTID(lastleft),
4752 : : BTreeTupleGetHeapTID(firstright)) < 0);
4753 [ - + ]: 526 : Assert(ItemPointerCompare(pivotheaptid,
4754 : : BTreeTupleGetHeapTID(lastleft)) >= 0);
4755 [ - + ]: 526 : Assert(ItemPointerCompare(pivotheaptid,
4756 : : BTreeTupleGetHeapTID(firstright)) < 0);
4757 : : #else
4758 : :
4759 : : /*
4760 : : * Those invariants aren't guaranteed to hold for lastleft + firstright
4761 : : * heap TID attribute values when they're considered here only because
4762 : : * DEBUG_NO_TRUNCATE is defined (a heap TID is probably not actually
4763 : : * needed as a tiebreaker). DEBUG_NO_TRUNCATE must therefore use a heap
4764 : : * TID value that always works as a strict lower bound for items to the
4765 : : * right. In particular, it must avoid using firstright's leading key
4766 : : * attribute values along with lastleft's heap TID value when lastleft's
4767 : : * TID happens to be greater than firstright's TID.
4768 : : */
4769 : : ItemPointerCopy(BTreeTupleGetHeapTID(firstright), pivotheaptid);
4770 : :
4771 : : /*
4772 : : * Pivot heap TID should never be fully equal to firstright. Note that
4773 : : * the pivot heap TID will still end up equal to lastleft's heap TID when
4774 : : * that's the only usable value.
4775 : : */
4776 : : ItemPointerSetOffsetNumber(pivotheaptid,
4777 : : OffsetNumberPrev(ItemPointerGetOffsetNumber(pivotheaptid)));
4778 : : Assert(ItemPointerCompare(pivotheaptid,
4779 : : BTreeTupleGetHeapTID(firstright)) < 0);
4780 : : #endif
4781 : :
1476 4782 : 526 : return tidpivot;
4783 : : }
4784 : :
4785 : : /*
4786 : : * _bt_keep_natts - how many key attributes to keep when truncating.
4787 : : *
4788 : : * Caller provides two tuples that enclose a split point. Caller's insertion
4789 : : * scankey is used to compare the tuples; the scankey's argument values are
4790 : : * not considered here.
4791 : : *
4792 : : * This can return a number of attributes that is one greater than the
4793 : : * number of key attributes for the index relation. This indicates that the
4794 : : * caller must use a heap TID as a unique-ifier in new pivot tuple.
4795 : : */
4796 : : static int
1852 4797 : 29103 : _bt_keep_natts(Relation rel, IndexTuple lastleft, IndexTuple firstright,
4798 : : BTScanInsert itup_key)
4799 : : {
4800 : 29103 : int nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
4801 : 29103 : TupleDesc itupdesc = RelationGetDescr(rel);
4802 : : int keepnatts;
4803 : : ScanKey scankey;
4804 : :
4805 : : /*
4806 : : * _bt_compare() treats truncated key attributes as having the value minus
4807 : : * infinity, which would break searches within !heapkeyspace indexes. We
4808 : : * must still truncate away non-key attribute values, though.
4809 : : */
4810 [ - + ]: 29103 : if (!itup_key->heapkeyspace)
1852 pg@bowt.ie 4811 :UBC 0 : return nkeyatts;
4812 : :
1852 pg@bowt.ie 4813 :CBC 29103 : scankey = itup_key->scankeys;
4814 : 29103 : keepnatts = 1;
4815 [ + + ]: 35059 : for (int attnum = 1; attnum <= nkeyatts; attnum++, scankey++)
4816 : : {
4817 : : Datum datum1,
4818 : : datum2;
4819 : : bool isNull1,
4820 : : isNull2;
4821 : :
4822 : 34533 : datum1 = index_getattr(lastleft, attnum, itupdesc, &isNull1);
4823 : 34533 : datum2 = index_getattr(firstright, attnum, itupdesc, &isNull2);
4824 : :
4825 [ - + ]: 34533 : if (isNull1 != isNull2)
1852 pg@bowt.ie 4826 :UBC 0 : break;
4827 : :
1852 pg@bowt.ie 4828 [ + + + + ]:CBC 69051 : if (!isNull1 &&
4829 : 34518 : DatumGetInt32(FunctionCall2Coll(&scankey->sk_func,
4830 : : scankey->sk_collation,
4831 : : datum1,
4832 : : datum2)) != 0)
4833 : 28577 : break;
4834 : :
4835 : 5956 : keepnatts++;
4836 : : }
4837 : :
4838 : : /*
4839 : : * Assert that _bt_keep_natts_fast() agrees with us in passing. This is
4840 : : * expected in an allequalimage index.
4841 : : */
1509 4842 [ + + - + ]: 29103 : Assert(!itup_key->allequalimage ||
4843 : : keepnatts == _bt_keep_natts_fast(rel, lastleft, firstright));
4844 : :
1852 4845 : 29103 : return keepnatts;
4846 : : }
4847 : :
4848 : : /*
4849 : : * _bt_keep_natts_fast - fast bitwise variant of _bt_keep_natts.
4850 : : *
4851 : : * This is exported so that a candidate split point can have its effect on
4852 : : * suffix truncation inexpensively evaluated ahead of time when finding a
4853 : : * split location. A naive bitwise approach to datum comparisons is used to
4854 : : * save cycles.
4855 : : *
4856 : : * The approach taken here usually provides the same answer as _bt_keep_natts
4857 : : * will (for the same pair of tuples from a heapkeyspace index), since the
4858 : : * majority of btree opclasses can never indicate that two datums are equal
4859 : : * unless they're bitwise equal after detoasting. When an index only has
4860 : : * "equal image" columns, routine is guaranteed to give the same result as
4861 : : * _bt_keep_natts would.
4862 : : *
4863 : : * Callers can rely on the fact that attributes considered equal here are
4864 : : * definitely also equal according to _bt_keep_natts, even when the index uses
4865 : : * an opclass or collation that is not "allequalimage"/deduplication-safe.
4866 : : * This weaker guarantee is good enough for nbtsplitloc.c caller, since false
4867 : : * negatives generally only have the effect of making leaf page splits use a
4868 : : * more balanced split point.
4869 : : */
4870 : : int
4871 : 6148929 : _bt_keep_natts_fast(Relation rel, IndexTuple lastleft, IndexTuple firstright)
4872 : : {
4873 : 6148929 : TupleDesc itupdesc = RelationGetDescr(rel);
4874 : 6148929 : int keysz = IndexRelationGetNumberOfKeyAttributes(rel);
4875 : : int keepnatts;
4876 : :
4877 : 6148929 : keepnatts = 1;
4878 [ + + ]: 10073174 : for (int attnum = 1; attnum <= keysz; attnum++)
4879 : : {
4880 : : Datum datum1,
4881 : : datum2;
4882 : : bool isNull1,
4883 : : isNull2;
4884 : : Form_pg_attribute att;
4885 : :
4886 : 8986052 : datum1 = index_getattr(lastleft, attnum, itupdesc, &isNull1);
4887 : 8986052 : datum2 = index_getattr(firstright, attnum, itupdesc, &isNull2);
4888 : 8986052 : att = TupleDescAttr(itupdesc, attnum - 1);
4889 : :
4890 [ + + ]: 8986052 : if (isNull1 != isNull2)
4891 : 5061807 : break;
4892 : :
4893 [ + + ]: 8985980 : if (!isNull1 &&
1615 4894 [ + + ]: 8962437 : !datum_image_eq(datum1, datum2, att->attbyval, att->attlen))
1852 4895 : 5061735 : break;
4896 : :
4897 : 3924245 : keepnatts++;
4898 : : }
4899 : :
4900 : 6148929 : return keepnatts;
4901 : : }
4902 : :
4903 : : /*
4904 : : * _bt_check_natts() -- Verify tuple has expected number of attributes.
4905 : : *
4906 : : * Returns value indicating if the expected number of attributes were found
4907 : : * for a particular offset on page. This can be used as a general purpose
4908 : : * sanity check.
4909 : : *
4910 : : * Testing a tuple directly with BTreeTupleGetNAtts() should generally be
4911 : : * preferred to calling here. That's usually more convenient, and is always
4912 : : * more explicit. Call here instead when offnum's tuple may be a negative
4913 : : * infinity tuple that uses the pre-v11 on-disk representation, or when a low
4914 : : * context check is appropriate. This routine is as strict as possible about
4915 : : * what is expected on each version of btree.
4916 : : */
4917 : : bool
4918 : 121516554 : _bt_check_natts(Relation rel, bool heapkeyspace, Page page, OffsetNumber offnum)
4919 : : {
2180 tgl@sss.pgh.pa.us 4920 : 121516554 : int16 natts = IndexRelationGetNumberOfAttributes(rel);
4921 : 121516554 : int16 nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
744 michael@paquier.xyz 4922 : 121516554 : BTPageOpaque opaque = BTPageGetOpaque(page);
4923 : : IndexTuple itup;
4924 : : int tupnatts;
4925 : :
4926 : : /*
4927 : : * We cannot reliably test a deleted or half-dead page, since they have
4928 : : * dummy high keys
4929 : : */
2187 teodor@sigaev.ru 4930 [ - + ]: 121516554 : if (P_IGNORE(opaque))
2187 teodor@sigaev.ru 4931 :UBC 0 : return true;
4932 : :
2187 teodor@sigaev.ru 4933 [ + - - + ]:CBC 121516554 : Assert(offnum >= FirstOffsetNumber &&
4934 : : offnum <= PageGetMaxOffsetNumber(page));
4935 : :
4936 : 121516554 : itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offnum));
1852 pg@bowt.ie 4937 [ + + ]: 121516554 : tupnatts = BTreeTupleGetNAtts(itup, rel);
4938 : :
4939 : : /* !heapkeyspace indexes do not support deduplication */
1509 4940 [ - + - - ]: 121516554 : if (!heapkeyspace && BTreeTupleIsPosting(itup))
1509 pg@bowt.ie 4941 :UBC 0 : return false;
4942 : :
4943 : : /* Posting list tuples should never have "pivot heap TID" bit set */
1509 pg@bowt.ie 4944 [ + + ]:CBC 121516554 : if (BTreeTupleIsPosting(itup) &&
4945 [ - + ]: 1581521 : (ItemPointerGetOffsetNumberNoCheck(&itup->t_tid) &
4946 : : BT_PIVOT_HEAP_TID_ATTR) != 0)
1509 pg@bowt.ie 4947 :UBC 0 : return false;
4948 : :
4949 : : /* INCLUDE indexes do not support deduplication */
1509 pg@bowt.ie 4950 [ + + - + ]:CBC 121516554 : if (natts != nkeyatts && BTreeTupleIsPosting(itup))
1509 pg@bowt.ie 4951 :UBC 0 : return false;
4952 : :
2187 teodor@sigaev.ru 4953 [ + + ]:CBC 121516554 : if (P_ISLEAF(opaque))
4954 : : {
4955 [ + + + + ]: 89499096 : if (offnum >= P_FIRSTDATAKEY(opaque))
4956 : : {
4957 : : /*
4958 : : * Non-pivot tuple should never be explicitly marked as a pivot
4959 : : * tuple
4960 : : */
1509 pg@bowt.ie 4961 [ - + ]: 82191287 : if (BTreeTupleIsPivot(itup))
1852 pg@bowt.ie 4962 :UBC 0 : return false;
4963 : :
4964 : : /*
4965 : : * Leaf tuples that are not the page high key (non-pivot tuples)
4966 : : * should never be truncated. (Note that tupnatts must have been
4967 : : * inferred, even with a posting list tuple, because only pivot
4968 : : * tuples store tupnatts directly.)
4969 : : */
1852 pg@bowt.ie 4970 :CBC 82191287 : return tupnatts == natts;
4971 : : }
4972 : : else
4973 : : {
4974 : : /*
4975 : : * Rightmost page doesn't contain a page high key, so tuple was
4976 : : * checked above as ordinary leaf tuple
4977 : : */
2187 teodor@sigaev.ru 4978 [ - + ]: 7307809 : Assert(!P_RIGHTMOST(opaque));
4979 : :
4980 : : /*
4981 : : * !heapkeyspace high key tuple contains only key attributes. Note
4982 : : * that tupnatts will only have been explicitly represented in
4983 : : * !heapkeyspace indexes that happen to have non-key attributes.
4984 : : */
1852 pg@bowt.ie 4985 [ - + ]: 7307809 : if (!heapkeyspace)
1852 pg@bowt.ie 4986 :UBC 0 : return tupnatts == nkeyatts;
4987 : :
4988 : : /* Use generic heapkeyspace pivot tuple handling */
4989 : : }
4990 : : }
4991 : : else /* !P_ISLEAF(opaque) */
4992 : : {
2187 teodor@sigaev.ru 4993 [ + + + + ]:CBC 32017458 : if (offnum == P_FIRSTDATAKEY(opaque))
4994 : : {
4995 : : /*
4996 : : * The first tuple on any internal page (possibly the first after
4997 : : * its high key) is its negative infinity tuple. Negative
4998 : : * infinity tuples are always truncated to zero attributes. They
4999 : : * are a particular kind of pivot tuple.
5000 : : */
1852 pg@bowt.ie 5001 [ + - ]: 1399389 : if (heapkeyspace)
5002 : 1399389 : return tupnatts == 0;
5003 : :
5004 : : /*
5005 : : * The number of attributes won't be explicitly represented if the
5006 : : * negative infinity tuple was generated during a page split that
5007 : : * occurred with a version of Postgres before v11. There must be
5008 : : * a problem when there is an explicit representation that is
5009 : : * non-zero, or when there is no explicit representation and the
5010 : : * tuple is evidently not a pre-pg_upgrade tuple.
5011 : : *
5012 : : * Prior to v11, downlinks always had P_HIKEY as their offset.
5013 : : * Accept that as an alternative indication of a valid
5014 : : * !heapkeyspace negative infinity tuple.
5015 : : */
1852 pg@bowt.ie 5016 [ # # # # ]:UBC 0 : return tupnatts == 0 ||
1509 5017 : 0 : ItemPointerGetOffsetNumber(&(itup->t_tid)) == P_HIKEY;
5018 : : }
5019 : : else
5020 : : {
5021 : : /*
5022 : : * !heapkeyspace downlink tuple with separator key contains only
5023 : : * key attributes. Note that tupnatts will only have been
5024 : : * explicitly represented in !heapkeyspace indexes that happen to
5025 : : * have non-key attributes.
5026 : : */
1852 pg@bowt.ie 5027 [ - + ]:CBC 30618069 : if (!heapkeyspace)
1852 pg@bowt.ie 5028 :UBC 0 : return tupnatts == nkeyatts;
5029 : :
5030 : : /* Use generic heapkeyspace pivot tuple handling */
5031 : : }
5032 : : }
5033 : :
5034 : : /* Handle heapkeyspace pivot tuples (excluding minus infinity items) */
1852 pg@bowt.ie 5035 [ - + ]:CBC 37925878 : Assert(heapkeyspace);
5036 : :
5037 : : /*
5038 : : * Explicit representation of the number of attributes is mandatory with
5039 : : * heapkeyspace index pivot tuples, regardless of whether or not there are
5040 : : * non-key attributes.
5041 : : */
1509 5042 [ - + ]: 37925878 : if (!BTreeTupleIsPivot(itup))
1509 pg@bowt.ie 5043 :UBC 0 : return false;
5044 : :
5045 : : /* Pivot tuple should not use posting list representation (redundant) */
1509 pg@bowt.ie 5046 [ - + ]:CBC 37925878 : if (BTreeTupleIsPosting(itup))
1852 pg@bowt.ie 5047 :UBC 0 : return false;
5048 : :
5049 : : /*
5050 : : * Heap TID is a tiebreaker key attribute, so it cannot be untruncated
5051 : : * when any other key attribute is truncated
5052 : : */
1852 pg@bowt.ie 5053 [ + + - + ]:CBC 37925878 : if (BTreeTupleGetHeapTID(itup) != NULL && tupnatts != nkeyatts)
1852 pg@bowt.ie 5054 :UBC 0 : return false;
5055 : :
5056 : : /*
5057 : : * Pivot tuple must have at least one untruncated key attribute (minus
5058 : : * infinity pivot tuples are the only exception). Pivot tuples can never
5059 : : * represent that there is a value present for a key attribute that
5060 : : * exceeds pg_index.indnkeyatts for the index.
5061 : : */
1852 pg@bowt.ie 5062 [ + - + - ]:CBC 37925878 : return tupnatts > 0 && tupnatts <= nkeyatts;
5063 : : }
5064 : :
5065 : : /*
5066 : : *
5067 : : * _bt_check_third_page() -- check whether tuple fits on a btree page at all.
5068 : : *
5069 : : * We actually need to be able to fit three items on every page, so restrict
5070 : : * any one item to 1/3 the per-page available space. Note that itemsz should
5071 : : * not include the ItemId overhead.
5072 : : *
5073 : : * It might be useful to apply TOAST methods rather than throw an error here.
5074 : : * Using out of line storage would break assumptions made by suffix truncation
5075 : : * and by contrib/amcheck, though.
5076 : : */
5077 : : void
5078 : 132 : _bt_check_third_page(Relation rel, Relation heap, bool needheaptidspace,
5079 : : Page page, IndexTuple newtup)
5080 : : {
5081 : : Size itemsz;
5082 : : BTPageOpaque opaque;
5083 : :
5084 : 132 : itemsz = MAXALIGN(IndexTupleSize(newtup));
5085 : :
5086 : : /* Double check item size against limit */
5087 [ - + ]: 132 : if (itemsz <= BTMaxItemSize(page))
1852 pg@bowt.ie 5088 :UBC 0 : return;
5089 : :
5090 : : /*
5091 : : * Tuple is probably too large to fit on page, but it's possible that the
5092 : : * index uses version 2 or version 3, or that page is an internal page, in
5093 : : * which case a slightly higher limit applies.
5094 : : */
1852 pg@bowt.ie 5095 [ + - + - ]:CBC 132 : if (!needheaptidspace && itemsz <= BTMaxItemSizeNoHeapTid(page))
5096 : 132 : return;
5097 : :
5098 : : /*
5099 : : * Internal page insertions cannot fail here, because that would mean that
5100 : : * an earlier leaf level insertion that should have failed didn't
5101 : : */
744 michael@paquier.xyz 5102 :UBC 0 : opaque = BTPageGetOpaque(page);
1852 pg@bowt.ie 5103 [ # # ]: 0 : if (!P_ISLEAF(opaque))
5104 [ # # ]: 0 : elog(ERROR, "cannot insert oversized tuple of size %zu on internal page of index \"%s\"",
5105 : : itemsz, RelationGetRelationName(rel));
5106 : :
5107 [ # # # # : 0 : ereport(ERROR,
# # ]
5108 : : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
5109 : : errmsg("index row size %zu exceeds btree version %u maximum %zu for index \"%s\"",
5110 : : itemsz,
5111 : : needheaptidspace ? BTREE_VERSION : BTREE_NOVAC_VERSION,
5112 : : needheaptidspace ? BTMaxItemSize(page) :
5113 : : BTMaxItemSizeNoHeapTid(page),
5114 : : RelationGetRelationName(rel)),
5115 : : errdetail("Index row references tuple (%u,%u) in relation \"%s\".",
5116 : : ItemPointerGetBlockNumber(BTreeTupleGetHeapTID(newtup)),
5117 : : ItemPointerGetOffsetNumber(BTreeTupleGetHeapTID(newtup)),
5118 : : RelationGetRelationName(heap)),
5119 : : errhint("Values larger than 1/3 of a buffer page cannot be indexed.\n"
5120 : : "Consider a function index of an MD5 hash of the value, "
5121 : : "or use full text indexing."),
5122 : : errtableconstraint(heap, RelationGetRelationName(rel))));
5123 : : }
5124 : :
5125 : : /*
5126 : : * Are all attributes in rel "equality is image equality" attributes?
5127 : : *
5128 : : * We use each attribute's BTEQUALIMAGE_PROC opclass procedure. If any
5129 : : * opclass either lacks a BTEQUALIMAGE_PROC procedure or returns false, we
5130 : : * return false; otherwise we return true.
5131 : : *
5132 : : * Returned boolean value is stored in index metapage during index builds.
5133 : : * Deduplication can only be used when we return true.
5134 : : */
5135 : : bool
1509 pg@bowt.ie 5136 :CBC 26628 : _bt_allequalimage(Relation rel, bool debugmessage)
5137 : : {
5138 : 26628 : bool allequalimage = true;
5139 : :
5140 : : /* INCLUDE indexes can never support deduplication */
5141 : 26628 : if (IndexRelationGetNumberOfAttributes(rel) !=
5142 [ + + ]: 26628 : IndexRelationGetNumberOfKeyAttributes(rel))
5143 : 133 : return false;
5144 : :
5145 [ + + ]: 69686 : for (int i = 0; i < IndexRelationGetNumberOfKeyAttributes(rel); i++)
5146 : : {
5147 : 43418 : Oid opfamily = rel->rd_opfamily[i];
5148 : 43418 : Oid opcintype = rel->rd_opcintype[i];
5149 : 43418 : Oid collation = rel->rd_indcollation[i];
5150 : : Oid equalimageproc;
5151 : :
5152 : 43418 : equalimageproc = get_opfamily_proc(opfamily, opcintype, opcintype,
5153 : : BTEQUALIMAGE_PROC);
5154 : :
5155 : : /*
5156 : : * If there is no BTEQUALIMAGE_PROC then deduplication is assumed to
5157 : : * be unsafe. Otherwise, actually call proc and see what it says.
5158 : : */
5159 [ + + ]: 43418 : if (!OidIsValid(equalimageproc) ||
5160 [ + + ]: 43201 : !DatumGetBool(OidFunctionCall1Coll(equalimageproc, collation,
5161 : : ObjectIdGetDatum(opcintype))))
5162 : : {
5163 : 227 : allequalimage = false;
5164 : 227 : break;
5165 : : }
5166 : : }
5167 : :
5168 [ + + ]: 26495 : if (debugmessage)
5169 : : {
5170 [ + + ]: 22447 : if (allequalimage)
5171 [ + + ]: 22220 : elog(DEBUG1, "index \"%s\" can safely use deduplication",
5172 : : RelationGetRelationName(rel));
5173 : : else
5174 [ - + ]: 227 : elog(DEBUG1, "index \"%s\" cannot use deduplication",
5175 : : RelationGetRelationName(rel));
5176 : : }
5177 : :
5178 : 26495 : return allequalimage;
5179 : : }
|
