Line data Source code
1 : /*
2 : * brin_minmax_multi.c
3 : * Implementation of Multi Min/Max opclass for BRIN
4 : *
5 : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
6 : * Portions Copyright (c) 1994, Regents of the University of California
7 : *
8 : *
9 : * Implements a variant of minmax opclass, where the summary is composed of
10 : * multiple smaller intervals. This allows us to handle outliers, which
11 : * usually make the simple minmax opclass inefficient.
12 : *
13 : * Consider for example page range with simple minmax interval [1000,2000],
14 : * and assume a new row gets inserted into the range with value 1000000.
15 : * Due to that the interval gets [1000,1000000]. I.e. the minmax interval
16 : * got 1000x wider and won't be useful to eliminate scan keys between 2001
17 : * and 1000000.
18 : *
19 : * With minmax-multi opclass, we may have [1000,2000] interval initially,
20 : * but after adding the new row we start tracking it as two interval:
21 : *
22 : * [1000,2000] and [1000000,1000000]
23 : *
24 : * This allows us to still eliminate the page range when the scan keys hit
25 : * the gap between 2000 and 1000000, making it useful in cases when the
26 : * simple minmax opclass gets inefficient.
27 : *
28 : * The number of intervals tracked per page range is somewhat flexible.
29 : * What is restricted is the number of values per page range, and the limit
30 : * is currently 32 (see values_per_range reloption). Collapsed intervals
31 : * (with equal minimum and maximum value) are stored as a single value,
32 : * while regular intervals require two values.
33 : *
34 : * When the number of values gets too high (by adding new values to the
35 : * summary), we merge some of the intervals to free space for more values.
36 : * This is done in a greedy way - we simply pick the two closest intervals,
37 : * merge them, and repeat this until the number of values to store gets
38 : * sufficiently low (below 50% of maximum values), but that is mostly
39 : * arbitrary threshold and may be changed easily).
40 : *
41 : * To pick the closest intervals we use the "distance" support procedure,
42 : * which measures space between two ranges (i.e. the length of an interval).
43 : * The computed value may be an approximation - in the worst case we will
44 : * merge two ranges that are slightly less optimal at that step, but the
45 : * index should still produce correct results.
46 : *
47 : * The compactions (reducing the number of values) is fairly expensive, as
48 : * it requires calling the distance functions, sorting etc. So when building
49 : * the summary, we use a significantly larger buffer, and only enforce the
50 : * exact limit at the very end. This improves performance, and it also helps
51 : * with building better ranges (due to the greedy approach).
52 : *
53 : *
54 : * IDENTIFICATION
55 : * src/backend/access/brin/brin_minmax_multi.c
56 : */
57 : #include "postgres.h"
58 :
59 : /* needed for PGSQL_AF_INET */
60 : #include <sys/socket.h>
61 :
62 : #include "access/brin.h"
63 : #include "access/brin_internal.h"
64 : #include "access/brin_tuple.h"
65 : #include "access/genam.h"
66 : #include "access/htup_details.h"
67 : #include "access/reloptions.h"
68 : #include "access/stratnum.h"
69 : #include "catalog/pg_am.h"
70 : #include "catalog/pg_amop.h"
71 : #include "catalog/pg_type.h"
72 : #include "utils/array.h"
73 : #include "utils/builtins.h"
74 : #include "utils/date.h"
75 : #include "utils/datum.h"
76 : #include "utils/float.h"
77 : #include "utils/inet.h"
78 : #include "utils/lsyscache.h"
79 : #include "utils/memutils.h"
80 : #include "utils/pg_lsn.h"
81 : #include "utils/rel.h"
82 : #include "utils/syscache.h"
83 : #include "utils/timestamp.h"
84 : #include "utils/uuid.h"
85 :
86 : /*
87 : * Additional SQL level support functions
88 : *
89 : * Procedure numbers must not use values reserved for BRIN itself; see
90 : * brin_internal.h.
91 : */
92 : #define MINMAX_MAX_PROCNUMS 1 /* maximum support procs we need */
93 : #define PROCNUM_DISTANCE 11 /* required, distance between values */
94 :
95 : /*
96 : * Subtract this from procnum to obtain index in MinmaxMultiOpaque arrays
97 : * (Must be equal to minimum of private procnums).
98 : */
99 : #define PROCNUM_BASE 11
100 :
101 : /*
102 : * Sizing the insert buffer - we use 10x the number of values specified
103 : * in the reloption, but we cap it to 8192 not to get too large. When
104 : * the buffer gets full, we reduce the number of values by half.
105 : */
106 : #define MINMAX_BUFFER_FACTOR 10
107 : #define MINMAX_BUFFER_MIN 256
108 : #define MINMAX_BUFFER_MAX 8192
109 : #define MINMAX_BUFFER_LOAD_FACTOR 0.5
110 :
111 : typedef struct MinmaxMultiOpaque
112 : {
113 : FmgrInfo extra_procinfos[MINMAX_MAX_PROCNUMS];
114 : Oid cached_subtype;
115 : FmgrInfo strategy_procinfos[BTMaxStrategyNumber];
116 : } MinmaxMultiOpaque;
117 :
118 : /*
119 : * Storage type for BRIN's minmax reloptions
120 : */
121 : typedef struct MinMaxMultiOptions
122 : {
123 : int32 vl_len_; /* varlena header (do not touch directly!) */
124 : int valuesPerRange; /* number of values per range */
125 : } MinMaxMultiOptions;
126 :
127 : #define MINMAX_MULTI_DEFAULT_VALUES_PER_PAGE 32
128 :
129 : #define MinMaxMultiGetValuesPerRange(opts) \
130 : ((opts) && (((MinMaxMultiOptions *) (opts))->valuesPerRange != 0) ? \
131 : ((MinMaxMultiOptions *) (opts))->valuesPerRange : \
132 : MINMAX_MULTI_DEFAULT_VALUES_PER_PAGE)
133 :
134 : /*
135 : * The summary of minmax-multi indexes has two representations - Ranges for
136 : * convenient processing, and SerializedRanges for storage in bytea value.
137 : *
138 : * The Ranges struct stores the boundary values in a single array, but we
139 : * treat regular and single-point ranges differently to save space. For
140 : * regular ranges (with different boundary values) we have to store both
141 : * the lower and upper bound of the range, while for "single-point ranges"
142 : * we only need to store a single value.
143 : *
144 : * The 'values' array stores boundary values for regular ranges first (there
145 : * are 2*nranges values to store), and then the nvalues boundary values for
146 : * single-point ranges. That is, we have (2*nranges + nvalues) boundary
147 : * values in the array.
148 : *
149 : * +-------------------------+----------------------------------+
150 : * | ranges (2 * nranges of) | single point values (nvalues of) |
151 : * +-------------------------+----------------------------------+
152 : *
153 : * This allows us to quickly add new values, and store outliers without
154 : * having to widen any of the existing range values.
155 : *
156 : * 'nsorted' denotes how many of 'nvalues' in the values[] array are sorted.
157 : * When nsorted == nvalues, all single point values are sorted.
158 : *
159 : * We never store more than maxvalues values (as set by values_per_range
160 : * reloption). If needed we merge some of the ranges.
161 : *
162 : * To minimize palloc overhead, we always allocate the full array with
163 : * space for maxvalues elements. This should be fine as long as the
164 : * maxvalues is reasonably small (64 seems fine), which is the case
165 : * thanks to values_per_range reloption being limited to 256.
166 : */
167 : typedef struct Ranges
168 : {
169 : /* Cache information that we need quite often. */
170 : Oid typid;
171 : Oid colloid;
172 : AttrNumber attno;
173 : FmgrInfo *cmp;
174 :
175 : /* (2*nranges + nvalues) <= maxvalues */
176 : int nranges; /* number of ranges in the values[] array */
177 : int nsorted; /* number of nvalues which are sorted */
178 : int nvalues; /* number of point values in values[] array */
179 : int maxvalues; /* number of elements in the values[] array */
180 :
181 : /*
182 : * We simply add the values into a large buffer, without any expensive
183 : * steps (sorting, deduplication, ...). The buffer is a multiple of the
184 : * target number of values, so the compaction happens less often,
185 : * amortizing the costs. We keep the actual target and compact to the
186 : * requested number of values at the very end, before serializing to
187 : * on-disk representation.
188 : */
189 : /* requested number of values */
190 : int target_maxvalues;
191 :
192 : /* values stored for this range - either raw values, or ranges */
193 : Datum values[FLEXIBLE_ARRAY_MEMBER];
194 : } Ranges;
195 :
196 : /*
197 : * On-disk the summary is stored as a bytea value, with a simple header
198 : * with basic metadata, followed by the boundary values. It has a varlena
199 : * header, so can be treated as varlena directly.
200 : *
201 : * See brin_range_serialize/brin_range_deserialize for serialization details.
202 : */
203 : typedef struct SerializedRanges
204 : {
205 : /* varlena header (do not touch directly!) */
206 : int32 vl_len_;
207 :
208 : /* type of values stored in the data array */
209 : Oid typid;
210 :
211 : /* (2*nranges + nvalues) <= maxvalues */
212 : int nranges; /* number of ranges in the array (stored) */
213 : int nvalues; /* number of values in the data array (all) */
214 : int maxvalues; /* maximum number of values (reloption) */
215 :
216 : /* contains the actual data */
217 : char data[FLEXIBLE_ARRAY_MEMBER];
218 : } SerializedRanges;
219 :
220 : static SerializedRanges *brin_range_serialize(Ranges *range);
221 :
222 : static Ranges *brin_range_deserialize(int maxvalues,
223 : SerializedRanges *serialized);
224 :
225 :
226 : /*
227 : * Used to represent ranges expanded to make merging and combining easier.
228 : *
229 : * Each expanded range is essentially an interval, represented by min/max
230 : * values, along with a flag whether it's a collapsed range (in which case
231 : * the min and max values are equal). We have the flag to handle by-ref
232 : * data types - we can't simply compare the datums, and this saves some
233 : * calls to the type-specific comparator function.
234 : */
235 : typedef struct ExpandedRange
236 : {
237 : Datum minval; /* lower boundary */
238 : Datum maxval; /* upper boundary */
239 : bool collapsed; /* true if minval==maxval */
240 : } ExpandedRange;
241 :
242 : /*
243 : * Represents a distance between two ranges (identified by index into
244 : * an array of extended ranges).
245 : */
246 : typedef struct DistanceValue
247 : {
248 : int index;
249 : double value;
250 : } DistanceValue;
251 :
252 :
253 : /* Cache for support and strategy procedures. */
254 :
255 : static FmgrInfo *minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno,
256 : uint16 procnum);
257 :
258 : static FmgrInfo *minmax_multi_get_strategy_procinfo(BrinDesc *bdesc,
259 : uint16 attno, Oid subtype,
260 : uint16 strategynum);
261 :
262 : typedef struct compare_context
263 : {
264 : FmgrInfo *cmpFn;
265 : Oid colloid;
266 : } compare_context;
267 :
268 : static int compare_values(const void *a, const void *b, void *arg);
269 :
270 :
271 : #ifdef USE_ASSERT_CHECKING
272 : /*
273 : * Check that the order of the array values is correct, using the cmp
274 : * function (which should be BTLessStrategyNumber).
275 : */
276 : static void
277 : AssertArrayOrder(FmgrInfo *cmp, Oid colloid, const Datum *values, int nvalues)
278 : {
279 : int i;
280 : Datum lt;
281 :
282 : for (i = 0; i < (nvalues - 1); i++)
283 : {
284 : lt = FunctionCall2Coll(cmp, colloid, values[i], values[i + 1]);
285 : Assert(DatumGetBool(lt));
286 : }
287 : }
288 : #endif
289 :
290 : /*
291 : * Comprehensive check of the Ranges structure.
292 : */
293 : static void
294 174554 : AssertCheckRanges(Ranges *ranges, FmgrInfo *cmpFn, Oid colloid)
295 : {
296 : #ifdef USE_ASSERT_CHECKING
297 : int i;
298 :
299 : /* some basic sanity checks */
300 : Assert(ranges->nranges >= 0);
301 : Assert(ranges->nsorted >= 0);
302 : Assert(ranges->nvalues >= ranges->nsorted);
303 : Assert(ranges->maxvalues >= 2 * ranges->nranges + ranges->nvalues);
304 : Assert(ranges->typid != InvalidOid);
305 :
306 : /*
307 : * First the ranges - there are 2*nranges boundary values, and the values
308 : * have to be strictly ordered (equal values would mean the range is
309 : * collapsed, and should be stored as a point). This also guarantees that
310 : * the ranges do not overlap.
311 : */
312 : AssertArrayOrder(cmpFn, colloid, ranges->values, 2 * ranges->nranges);
313 :
314 : /* then the single-point ranges (with nvalues boundary values ) */
315 : AssertArrayOrder(cmpFn, colloid, &ranges->values[2 * ranges->nranges],
316 : ranges->nsorted);
317 :
318 : /*
319 : * Check that none of the values are not covered by ranges (both sorted
320 : * and unsorted)
321 : */
322 : if (ranges->nranges > 0)
323 : {
324 : for (i = 0; i < ranges->nvalues; i++)
325 : {
326 : Datum compar;
327 : int start,
328 : end;
329 : Datum minvalue = ranges->values[0];
330 : Datum maxvalue = ranges->values[2 * ranges->nranges - 1];
331 : Datum value = ranges->values[2 * ranges->nranges + i];
332 :
333 : compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
334 :
335 : /*
336 : * If the value is smaller than the lower bound in the first range
337 : * then it cannot possibly be in any of the ranges.
338 : */
339 : if (DatumGetBool(compar))
340 : continue;
341 :
342 : compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
343 :
344 : /*
345 : * Likewise, if the value is larger than the upper bound of the
346 : * final range, then it cannot possibly be inside any of the
347 : * ranges.
348 : */
349 : if (DatumGetBool(compar))
350 : continue;
351 :
352 : /* bsearch the ranges to see if 'value' fits within any of them */
353 : start = 0; /* first range */
354 : end = ranges->nranges - 1; /* last range */
355 : while (true)
356 : {
357 : int midpoint = (start + end) / 2;
358 :
359 : /* this means we ran out of ranges in the last step */
360 : if (start > end)
361 : break;
362 :
363 : /* copy the min/max values from the ranges */
364 : minvalue = ranges->values[2 * midpoint];
365 : maxvalue = ranges->values[2 * midpoint + 1];
366 :
367 : /*
368 : * Is the value smaller than the minval? If yes, we'll recurse
369 : * to the left side of range array.
370 : */
371 : compar = FunctionCall2Coll(cmpFn, colloid, value, minvalue);
372 :
373 : /* smaller than the smallest value in this range */
374 : if (DatumGetBool(compar))
375 : {
376 : end = (midpoint - 1);
377 : continue;
378 : }
379 :
380 : /*
381 : * Is the value greater than the minval? If yes, we'll recurse
382 : * to the right side of range array.
383 : */
384 : compar = FunctionCall2Coll(cmpFn, colloid, maxvalue, value);
385 :
386 : /* larger than the largest value in this range */
387 : if (DatumGetBool(compar))
388 : {
389 : start = (midpoint + 1);
390 : continue;
391 : }
392 :
393 : /* hey, we found a matching range */
394 : Assert(false);
395 : }
396 : }
397 : }
398 :
399 : /* and values in the unsorted part must not be in the sorted part */
400 : if (ranges->nsorted > 0)
401 : {
402 : compare_context cxt;
403 :
404 : cxt.colloid = ranges->colloid;
405 : cxt.cmpFn = ranges->cmp;
406 :
407 : for (i = ranges->nsorted; i < ranges->nvalues; i++)
408 : {
409 : Datum value = ranges->values[2 * ranges->nranges + i];
410 :
411 : Assert(bsearch_arg(&value, &ranges->values[2 * ranges->nranges],
412 : ranges->nsorted, sizeof(Datum),
413 : compare_values, &cxt) == NULL);
414 : }
415 : }
416 : #endif
417 174554 : }
418 :
419 : /*
420 : * Check that the expanded ranges (built when reducing the number of ranges
421 : * by combining some of them) are correctly sorted and do not overlap.
422 : */
423 : static void
424 328 : AssertCheckExpandedRanges(BrinDesc *bdesc, Oid colloid, AttrNumber attno,
425 : Form_pg_attribute attr, ExpandedRange *ranges,
426 : int nranges)
427 : {
428 : #ifdef USE_ASSERT_CHECKING
429 : int i;
430 : FmgrInfo *eq;
431 : FmgrInfo *lt;
432 :
433 : eq = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
434 : BTEqualStrategyNumber);
435 :
436 : lt = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
437 : BTLessStrategyNumber);
438 :
439 : /*
440 : * Each range independently should be valid, i.e. that for the boundary
441 : * values (lower <= upper).
442 : */
443 : for (i = 0; i < nranges; i++)
444 : {
445 : Datum r;
446 : Datum minval = ranges[i].minval;
447 : Datum maxval = ranges[i].maxval;
448 :
449 : if (ranges[i].collapsed) /* collapsed: minval == maxval */
450 : r = FunctionCall2Coll(eq, colloid, minval, maxval);
451 : else /* non-collapsed: minval < maxval */
452 : r = FunctionCall2Coll(lt, colloid, minval, maxval);
453 :
454 : Assert(DatumGetBool(r));
455 : }
456 :
457 : /*
458 : * And the ranges should be ordered and must not overlap, i.e. upper <
459 : * lower for boundaries of consecutive ranges.
460 : */
461 : for (i = 0; i < nranges - 1; i++)
462 : {
463 : Datum r;
464 : Datum maxval = ranges[i].maxval;
465 : Datum minval = ranges[i + 1].minval;
466 :
467 : r = FunctionCall2Coll(lt, colloid, maxval, minval);
468 :
469 : Assert(DatumGetBool(r));
470 : }
471 : #endif
472 328 : }
473 :
474 :
475 : /*
476 : * minmax_multi_init
477 : * Initialize the deserialized range list, allocate all the memory.
478 : *
479 : * This is only in-memory representation of the ranges, so we allocate
480 : * enough space for the maximum number of values (so as not to have to do
481 : * repallocs as the ranges grow).
482 : */
483 : static Ranges *
484 33789 : minmax_multi_init(int maxvalues)
485 : {
486 : Size len;
487 : Ranges *ranges;
488 :
489 : Assert(maxvalues > 0);
490 :
491 33789 : len = offsetof(Ranges, values); /* fixed header */
492 33789 : len += maxvalues * sizeof(Datum); /* Datum values */
493 :
494 33789 : ranges = (Ranges *) palloc0(len);
495 :
496 33789 : ranges->maxvalues = maxvalues;
497 :
498 33789 : return ranges;
499 : }
500 :
501 :
502 : /*
503 : * range_deduplicate_values
504 : * Deduplicate the part with values in the simple points.
505 : *
506 : * This is meant to be a cheaper way of reducing the size of the ranges. It
507 : * does not touch the ranges, and only sorts the other values - it does not
508 : * call the distance functions, which may be quite expensive, etc.
509 : *
510 : * We do know the values are not duplicate with the ranges, because we check
511 : * that before adding a new value. Same for the sorted part of values.
512 : */
513 : static void
514 12193 : range_deduplicate_values(Ranges *range)
515 : {
516 : int i,
517 : n;
518 : int start;
519 : compare_context cxt;
520 :
521 : /*
522 : * If there are no unsorted values, we're done (this probably can't
523 : * happen, as we're adding values to unsorted part).
524 : */
525 12193 : if (range->nsorted == range->nvalues)
526 12029 : return;
527 :
528 : /* sort the values */
529 164 : cxt.colloid = range->colloid;
530 164 : cxt.cmpFn = range->cmp;
531 :
532 : /* the values start right after the ranges (which are always sorted) */
533 164 : start = 2 * range->nranges;
534 :
535 : /*
536 : * XXX This might do a merge sort, to leverage that the first part of the
537 : * array is already sorted. If the sorted part is large, it might be quite
538 : * a bit faster.
539 : */
540 164 : qsort_arg(&range->values[start],
541 164 : range->nvalues, sizeof(Datum),
542 : compare_values, &cxt);
543 :
544 164 : n = 1;
545 52160 : for (i = 1; i < range->nvalues; i++)
546 : {
547 : /* same as preceding value, so store it */
548 51996 : if (compare_values(&range->values[start + i - 1],
549 51996 : &range->values[start + i],
550 : &cxt) == 0)
551 0 : continue;
552 :
553 51996 : range->values[start + n] = range->values[start + i];
554 :
555 51996 : n++;
556 : }
557 :
558 : /* now all the values are sorted */
559 164 : range->nvalues = n;
560 164 : range->nsorted = n;
561 :
562 164 : AssertCheckRanges(range, range->cmp, range->colloid);
563 : }
564 :
565 :
566 : /*
567 : * brin_range_serialize
568 : * Serialize the in-memory representation into a compact varlena value.
569 : *
570 : * Simply copy the header and then also the individual values, as stored
571 : * in the in-memory value array.
572 : */
573 : static SerializedRanges *
574 12029 : brin_range_serialize(Ranges *range)
575 : {
576 : Size len;
577 : int nvalues;
578 : SerializedRanges *serialized;
579 : Oid typid;
580 : int typlen;
581 : bool typbyval;
582 :
583 : char *ptr;
584 :
585 : /* simple sanity checks */
586 : Assert(range->nranges >= 0);
587 : Assert(range->nsorted >= 0);
588 : Assert(range->nvalues >= 0);
589 : Assert(range->maxvalues > 0);
590 : Assert(range->target_maxvalues > 0);
591 :
592 : /* at this point the range should be compacted to the target size */
593 : Assert(2 * range->nranges + range->nvalues <= range->target_maxvalues);
594 :
595 : Assert(range->target_maxvalues <= range->maxvalues);
596 :
597 : /* range boundaries are always sorted */
598 : Assert(range->nvalues >= range->nsorted);
599 :
600 : /* deduplicate values, if there's unsorted part */
601 12029 : range_deduplicate_values(range);
602 :
603 : /* see how many Datum values we actually have */
604 12029 : nvalues = 2 * range->nranges + range->nvalues;
605 :
606 12029 : typid = range->typid;
607 12029 : typbyval = get_typbyval(typid);
608 12029 : typlen = get_typlen(typid);
609 :
610 : /* header is always needed */
611 12029 : len = offsetof(SerializedRanges, data);
612 :
613 : /*
614 : * The space needed depends on data type - for fixed-length data types
615 : * (by-value and some by-reference) it's pretty simple, just multiply
616 : * (attlen * nvalues) and we're done. For variable-length by-reference
617 : * types we need to actually walk all the values and sum the lengths.
618 : */
619 12029 : if (typlen == -1) /* varlena */
620 : {
621 : int i;
622 :
623 7952 : for (i = 0; i < nvalues; i++)
624 : {
625 6231 : len += VARSIZE_ANY(DatumGetPointer(range->values[i]));
626 : }
627 : }
628 10308 : else if (typlen == -2) /* cstring */
629 : {
630 : int i;
631 :
632 0 : for (i = 0; i < nvalues; i++)
633 : {
634 : /* don't forget to include the null terminator ;-) */
635 0 : len += strlen(DatumGetCString(range->values[i])) + 1;
636 : }
637 : }
638 : else /* fixed-length types (even by-reference) */
639 : {
640 : Assert(typlen > 0);
641 10308 : len += nvalues * typlen;
642 : }
643 :
644 : /*
645 : * Allocate the serialized object, copy the basic information. The
646 : * serialized object is a varlena, so update the header.
647 : */
648 12029 : serialized = (SerializedRanges *) palloc0(len);
649 12029 : SET_VARSIZE(serialized, len);
650 :
651 12029 : serialized->typid = typid;
652 12029 : serialized->nranges = range->nranges;
653 12029 : serialized->nvalues = range->nvalues;
654 12029 : serialized->maxvalues = range->target_maxvalues;
655 :
656 : /*
657 : * And now copy also the boundary values (like the length calculation this
658 : * depends on the particular data type).
659 : */
660 12029 : ptr = serialized->data; /* start of the serialized data */
661 :
662 59295 : for (int i = 0; i < nvalues; i++)
663 : {
664 47266 : if (typbyval) /* simple by-value data types */
665 : {
666 : Datum tmp;
667 :
668 : /*
669 : * For byval types, we need to copy just the significant bytes -
670 : * we can't use memcpy directly, as that assumes little-endian
671 : * behavior. store_att_byval does almost what we need, but it
672 : * requires a properly aligned buffer - the output buffer does not
673 : * guarantee that. So we simply use a local Datum variable (which
674 : * guarantees proper alignment), and then copy the value from it.
675 : */
676 29377 : store_att_byval(&tmp, range->values[i], typlen);
677 :
678 29377 : memcpy(ptr, &tmp, typlen);
679 29377 : ptr += typlen;
680 : }
681 17889 : else if (typlen > 0) /* fixed-length by-ref types */
682 : {
683 11658 : memcpy(ptr, DatumGetPointer(range->values[i]), typlen);
684 11658 : ptr += typlen;
685 : }
686 6231 : else if (typlen == -1) /* varlena */
687 : {
688 6231 : int tmp = VARSIZE_ANY(DatumGetPointer(range->values[i]));
689 :
690 6231 : memcpy(ptr, DatumGetPointer(range->values[i]), tmp);
691 6231 : ptr += tmp;
692 : }
693 0 : else if (typlen == -2) /* cstring */
694 : {
695 0 : int tmp = strlen(DatumGetCString(range->values[i])) + 1;
696 :
697 0 : memcpy(ptr, DatumGetCString(range->values[i]), tmp);
698 0 : ptr += tmp;
699 : }
700 :
701 : /* make sure we haven't overflown the buffer end */
702 : Assert(ptr <= ((char *) serialized + len));
703 : }
704 :
705 : /* exact size */
706 : Assert(ptr == ((char *) serialized + len));
707 :
708 12029 : return serialized;
709 : }
710 :
711 : /*
712 : * brin_range_deserialize
713 : * Deserialize a compact varlena value into the in-memory representation.
714 : *
715 : * Simply copy the header and then also the individual values, as stored
716 : * in the in-memory value array.
717 : */
718 : static Ranges *
719 30524 : brin_range_deserialize(int maxvalues, SerializedRanges *serialized)
720 : {
721 : int i,
722 : nvalues;
723 : char *ptr,
724 : *dataptr;
725 : bool typbyval;
726 : int typlen;
727 : Size datalen;
728 :
729 : Ranges *range;
730 :
731 : Assert(serialized->nranges >= 0);
732 : Assert(serialized->nvalues >= 0);
733 : Assert(serialized->maxvalues > 0);
734 :
735 30524 : nvalues = 2 * serialized->nranges + serialized->nvalues;
736 :
737 : Assert(nvalues <= serialized->maxvalues);
738 : Assert(serialized->maxvalues <= maxvalues);
739 :
740 30524 : range = minmax_multi_init(maxvalues);
741 :
742 : /* copy the header info */
743 30524 : range->nranges = serialized->nranges;
744 30524 : range->nvalues = serialized->nvalues;
745 30524 : range->nsorted = serialized->nvalues;
746 30524 : range->maxvalues = maxvalues;
747 30524 : range->target_maxvalues = serialized->maxvalues;
748 :
749 30524 : range->typid = serialized->typid;
750 :
751 30524 : typbyval = get_typbyval(serialized->typid);
752 30524 : typlen = get_typlen(serialized->typid);
753 :
754 : /*
755 : * And now deconstruct the values into Datum array. We have to copy the
756 : * data because the serialized representation ignores alignment, and we
757 : * don't want to rely on it being kept around anyway.
758 : */
759 30524 : ptr = serialized->data;
760 :
761 : /*
762 : * We don't want to allocate many pieces, so we just allocate everything
763 : * in one chunk. How much space will we need?
764 : *
765 : * XXX We don't need to copy simple by-value data types.
766 : */
767 30524 : datalen = 0;
768 30524 : dataptr = NULL;
769 70380 : for (i = 0; (i < nvalues) && (!typbyval); i++)
770 : {
771 39856 : if (typlen > 0) /* fixed-length by-ref types */
772 24104 : datalen += MAXALIGN(typlen);
773 15752 : else if (typlen == -1) /* varlena */
774 : {
775 15752 : datalen += MAXALIGN(VARSIZE_ANY(ptr));
776 15752 : ptr += VARSIZE_ANY(ptr);
777 : }
778 0 : else if (typlen == -2) /* cstring */
779 : {
780 0 : Size slen = strlen(ptr) + 1;
781 :
782 0 : datalen += MAXALIGN(slen);
783 0 : ptr += slen;
784 : }
785 : }
786 :
787 30524 : if (datalen > 0)
788 11596 : dataptr = palloc(datalen);
789 :
790 : /*
791 : * Restore the source pointer (might have been modified when calculating
792 : * the space we need to allocate).
793 : */
794 30524 : ptr = serialized->data;
795 :
796 157548 : for (i = 0; i < nvalues; i++)
797 : {
798 127024 : if (typbyval) /* simple by-value data types */
799 : {
800 87168 : Datum v = 0;
801 :
802 87168 : memcpy(&v, ptr, typlen);
803 :
804 87168 : range->values[i] = fetch_att(&v, true, typlen);
805 87168 : ptr += typlen;
806 : }
807 39856 : else if (typlen > 0) /* fixed-length by-ref types */
808 : {
809 24104 : range->values[i] = PointerGetDatum(dataptr);
810 :
811 24104 : memcpy(dataptr, ptr, typlen);
812 24104 : dataptr += MAXALIGN(typlen);
813 :
814 24104 : ptr += typlen;
815 : }
816 15752 : else if (typlen == -1) /* varlena */
817 : {
818 15752 : range->values[i] = PointerGetDatum(dataptr);
819 :
820 15752 : memcpy(dataptr, ptr, VARSIZE_ANY(ptr));
821 15752 : dataptr += MAXALIGN(VARSIZE_ANY(ptr));
822 15752 : ptr += VARSIZE_ANY(ptr);
823 : }
824 0 : else if (typlen == -2) /* cstring */
825 : {
826 0 : Size slen = strlen(ptr) + 1;
827 :
828 0 : range->values[i] = PointerGetDatum(dataptr);
829 :
830 0 : memcpy(dataptr, ptr, slen);
831 0 : dataptr += MAXALIGN(slen);
832 0 : ptr += slen;
833 : }
834 :
835 : /* make sure we haven't overflown the buffer end */
836 : Assert(ptr <= ((char *) serialized + VARSIZE_ANY(serialized)));
837 : }
838 :
839 : /* should have consumed the whole input value exactly */
840 : Assert(ptr == ((char *) serialized + VARSIZE_ANY(serialized)));
841 :
842 : /* return the deserialized value */
843 30524 : return range;
844 : }
845 :
846 : /*
847 : * compare_expanded_ranges
848 : * Compare the expanded ranges - first by minimum, then by maximum.
849 : *
850 : * We do guarantee that ranges in a single Ranges object do not overlap, so it
851 : * may seem strange that we don't order just by minimum. But when merging two
852 : * Ranges (which happens in the union function), the ranges may in fact
853 : * overlap. So we do compare both.
854 : */
855 : static int
856 544649 : compare_expanded_ranges(const void *a, const void *b, void *arg)
857 : {
858 544649 : const ExpandedRange *ra = a;
859 544649 : const ExpandedRange *rb = b;
860 : Datum r;
861 :
862 544649 : compare_context *cxt = (compare_context *) arg;
863 :
864 : /* first compare minvals */
865 544649 : r = FunctionCall2Coll(cxt->cmpFn, cxt->colloid, ra->minval, rb->minval);
866 :
867 544649 : if (DatumGetBool(r))
868 356473 : return -1;
869 :
870 188176 : r = FunctionCall2Coll(cxt->cmpFn, cxt->colloid, rb->minval, ra->minval);
871 :
872 188176 : if (DatumGetBool(r))
873 151366 : return 1;
874 :
875 : /* then compare maxvals */
876 36810 : r = FunctionCall2Coll(cxt->cmpFn, cxt->colloid, ra->maxval, rb->maxval);
877 :
878 36810 : if (DatumGetBool(r))
879 0 : return -1;
880 :
881 36810 : r = FunctionCall2Coll(cxt->cmpFn, cxt->colloid, rb->maxval, ra->maxval);
882 :
883 36810 : if (DatumGetBool(r))
884 0 : return 1;
885 :
886 36810 : return 0;
887 : }
888 :
889 : /*
890 : * compare_values
891 : * Compare the values.
892 : */
893 : static int
894 761536 : compare_values(const void *a, const void *b, void *arg)
895 : {
896 761536 : const Datum *da = a;
897 761536 : const Datum *db = b;
898 : Datum r;
899 :
900 761536 : compare_context *cxt = (compare_context *) arg;
901 :
902 761536 : r = FunctionCall2Coll(cxt->cmpFn, cxt->colloid, *da, *db);
903 :
904 761536 : if (DatumGetBool(r))
905 404896 : return -1;
906 :
907 356640 : r = FunctionCall2Coll(cxt->cmpFn, cxt->colloid, *db, *da);
908 :
909 356640 : if (DatumGetBool(r))
910 311820 : return 1;
911 :
912 44820 : return 0;
913 : }
914 :
915 : /*
916 : * Check if the new value matches one of the existing ranges.
917 : */
918 : static bool
919 90304 : has_matching_range(BrinDesc *bdesc, Oid colloid, Ranges *ranges,
920 : Datum newval, AttrNumber attno, Oid typid)
921 : {
922 : Datum compar;
923 :
924 : Datum minvalue;
925 : Datum maxvalue;
926 :
927 : FmgrInfo *cmpLessFn;
928 : FmgrInfo *cmpGreaterFn;
929 :
930 : /* binary search on ranges */
931 : int start,
932 : end;
933 :
934 90304 : if (ranges->nranges == 0)
935 51004 : return false;
936 :
937 39300 : minvalue = ranges->values[0];
938 39300 : maxvalue = ranges->values[2 * ranges->nranges - 1];
939 :
940 : /*
941 : * Otherwise, need to compare the new value with boundaries of all the
942 : * ranges. First check if it's less than the absolute minimum, which is
943 : * the first value in the array.
944 : */
945 39300 : cmpLessFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
946 : BTLessStrategyNumber);
947 39300 : compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
948 :
949 : /* smaller than the smallest value in the range list */
950 39300 : if (DatumGetBool(compar))
951 12 : return false;
952 :
953 : /*
954 : * And now compare it to the existing maximum (last value in the data
955 : * array). But only if we haven't already ruled out a possible match in
956 : * the minvalue check.
957 : */
958 39288 : cmpGreaterFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
959 : BTGreaterStrategyNumber);
960 39288 : compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
961 :
962 39288 : if (DatumGetBool(compar))
963 38804 : return false;
964 :
965 : /*
966 : * So we know it's in the general min/max, the question is whether it
967 : * falls in one of the ranges or gaps. We'll do a binary search on
968 : * individual ranges - for each range we check equality (value falls into
969 : * the range), and then check ranges either above or below the current
970 : * range.
971 : */
972 484 : start = 0; /* first range */
973 484 : end = (ranges->nranges - 1); /* last range */
974 : while (true)
975 1036 : {
976 1520 : int midpoint = (start + end) / 2;
977 :
978 : /* this means we ran out of ranges in the last step */
979 1520 : if (start > end)
980 192 : return false;
981 :
982 : /* copy the min/max values from the ranges */
983 1328 : minvalue = ranges->values[2 * midpoint];
984 1328 : maxvalue = ranges->values[2 * midpoint + 1];
985 :
986 : /*
987 : * Is the value smaller than the minval? If yes, we'll recurse to the
988 : * left side of range array.
989 : */
990 1328 : compar = FunctionCall2Coll(cmpLessFn, colloid, newval, minvalue);
991 :
992 : /* smaller than the smallest value in this range */
993 1328 : if (DatumGetBool(compar))
994 : {
995 388 : end = (midpoint - 1);
996 388 : continue;
997 : }
998 :
999 : /*
1000 : * Is the value greater than the minval? If yes, we'll recurse to the
1001 : * right side of range array.
1002 : */
1003 940 : compar = FunctionCall2Coll(cmpGreaterFn, colloid, newval, maxvalue);
1004 :
1005 : /* larger than the largest value in this range */
1006 940 : if (DatumGetBool(compar))
1007 : {
1008 648 : start = (midpoint + 1);
1009 648 : continue;
1010 : }
1011 :
1012 : /* hey, we found a matching range */
1013 292 : return true;
1014 : }
1015 :
1016 : return false;
1017 : }
1018 :
1019 :
1020 : /*
1021 : * range_contains_value
1022 : * See if the new value is already contained in the range list.
1023 : *
1024 : * We first inspect the list of intervals. We use a small trick - we check
1025 : * the value against min/max of the whole range (min of the first interval,
1026 : * max of the last one) first, and only inspect the individual intervals if
1027 : * this passes.
1028 : *
1029 : * If the value matches none of the intervals, we check the exact values.
1030 : * We simply loop through them and invoke equality operator on them.
1031 : *
1032 : * The last parameter (full) determines whether we need to search all the
1033 : * values, including the unsorted part. With full=false, the unsorted part
1034 : * is not searched, which may produce false negatives and duplicate values
1035 : * (in the unsorted part only), but when we're building the range that's
1036 : * fine - we'll deduplicate before serialization, and it can only happen
1037 : * if there already are unsorted values (so it was already modified).
1038 : *
1039 : * Serialized ranges don't have any unsorted values, so this can't cause
1040 : * false negatives during querying.
1041 : */
1042 : static bool
1043 90304 : range_contains_value(BrinDesc *bdesc, Oid colloid,
1044 : AttrNumber attno, Form_pg_attribute attr,
1045 : Ranges *ranges, Datum newval, bool full)
1046 : {
1047 : int i;
1048 : FmgrInfo *cmpEqualFn;
1049 90304 : Oid typid = attr->atttypid;
1050 :
1051 : /*
1052 : * First inspect the ranges, if there are any. We first check the whole
1053 : * range, and only when there's still a chance of getting a match we
1054 : * inspect the individual ranges.
1055 : */
1056 90304 : if (has_matching_range(bdesc, colloid, ranges, newval, attno, typid))
1057 292 : return true;
1058 :
1059 90012 : cmpEqualFn = minmax_multi_get_strategy_procinfo(bdesc, attno, typid,
1060 : BTEqualStrategyNumber);
1061 :
1062 : /*
1063 : * There is no matching range, so let's inspect the sorted values.
1064 : *
1065 : * We do a sequential search for small numbers of values, and binary
1066 : * search once we have more than 16 values. This threshold is somewhat
1067 : * arbitrary, as it depends on how expensive the comparison function is.
1068 : *
1069 : * XXX If we use the threshold here, maybe we should do the same thing in
1070 : * has_matching_range? Or maybe we should do the bin search all the time?
1071 : *
1072 : * XXX We could use the same optimization as for ranges, to check if the
1073 : * value is between min/max, to maybe rule out all sorted values without
1074 : * having to inspect all of them.
1075 : */
1076 90012 : if (ranges->nsorted >= 16)
1077 : {
1078 : compare_context cxt;
1079 :
1080 38728 : cxt.colloid = ranges->colloid;
1081 38728 : cxt.cmpFn = ranges->cmp;
1082 :
1083 38728 : if (bsearch_arg(&newval, &ranges->values[2 * ranges->nranges],
1084 38728 : ranges->nsorted, sizeof(Datum),
1085 : compare_values, &cxt) != NULL)
1086 0 : return true;
1087 : }
1088 : else
1089 : {
1090 102378 : for (i = 2 * ranges->nranges; i < 2 * ranges->nranges + ranges->nsorted; i++)
1091 : {
1092 : Datum compar;
1093 :
1094 61213 : compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
1095 :
1096 : /* found an exact match */
1097 61213 : if (DatumGetBool(compar))
1098 10119 : return true;
1099 : }
1100 : }
1101 :
1102 : /* If not asked to inspect the unsorted part, we're done. */
1103 79893 : if (!full)
1104 79893 : return false;
1105 :
1106 : /* Inspect the unsorted part. */
1107 0 : for (i = 2 * ranges->nranges + ranges->nsorted; i < 2 * ranges->nranges + ranges->nvalues; i++)
1108 : {
1109 : Datum compar;
1110 :
1111 0 : compar = FunctionCall2Coll(cmpEqualFn, colloid, newval, ranges->values[i]);
1112 :
1113 : /* found an exact match */
1114 0 : if (DatumGetBool(compar))
1115 0 : return true;
1116 : }
1117 :
1118 : /* the value is not covered by this BRIN tuple */
1119 0 : return false;
1120 : }
1121 :
1122 : /*
1123 : * Expand ranges from Ranges into ExpandedRange array. This expects the
1124 : * eranges to be pre-allocated and with the correct size - there needs to be
1125 : * (nranges + nvalues) elements.
1126 : *
1127 : * The order of expanded ranges is arbitrary. We do expand the ranges first,
1128 : * and this part is sorted. But then we expand the values, and this part may
1129 : * be unsorted.
1130 : */
1131 : static void
1132 4193 : fill_expanded_ranges(ExpandedRange *eranges, int neranges, Ranges *ranges)
1133 : {
1134 : int idx;
1135 : int i;
1136 :
1137 : /* Check that the output array has the right size. */
1138 : Assert(neranges == (ranges->nranges + ranges->nvalues));
1139 :
1140 4193 : idx = 0;
1141 5705 : for (i = 0; i < ranges->nranges; i++)
1142 : {
1143 1512 : eranges[idx].minval = ranges->values[2 * i];
1144 1512 : eranges[idx].maxval = ranges->values[2 * i + 1];
1145 1512 : eranges[idx].collapsed = false;
1146 1512 : idx++;
1147 :
1148 : Assert(idx <= neranges);
1149 : }
1150 :
1151 101730 : for (i = 0; i < ranges->nvalues; i++)
1152 : {
1153 97537 : eranges[idx].minval = ranges->values[2 * ranges->nranges + i];
1154 97537 : eranges[idx].maxval = ranges->values[2 * ranges->nranges + i];
1155 97537 : eranges[idx].collapsed = true;
1156 97537 : idx++;
1157 :
1158 : Assert(idx <= neranges);
1159 : }
1160 :
1161 : /* Did we produce the expected number of elements? */
1162 : Assert(idx == neranges);
1163 :
1164 4193 : return;
1165 : }
1166 :
1167 : /*
1168 : * Sort and deduplicate expanded ranges.
1169 : *
1170 : * The ranges may be deduplicated - we're simply appending values, without
1171 : * checking for duplicates etc. So maybe the deduplication will reduce the
1172 : * number of ranges enough, and we won't have to compute the distances etc.
1173 : *
1174 : * Returns the number of expanded ranges.
1175 : */
1176 : static int
1177 4193 : sort_expanded_ranges(FmgrInfo *cmp, Oid colloid,
1178 : ExpandedRange *eranges, int neranges)
1179 : {
1180 : int n;
1181 : int i;
1182 : compare_context cxt;
1183 :
1184 : Assert(neranges > 0);
1185 :
1186 : /* sort the values */
1187 4193 : cxt.colloid = colloid;
1188 4193 : cxt.cmpFn = cmp;
1189 :
1190 : /*
1191 : * XXX We do qsort on all the values, but we could also leverage the fact
1192 : * that some of the input data is already sorted (all the ranges and maybe
1193 : * some of the points) and do merge sort.
1194 : */
1195 4193 : qsort_arg(eranges, neranges, sizeof(ExpandedRange),
1196 : compare_expanded_ranges, &cxt);
1197 :
1198 : /*
1199 : * Deduplicate the ranges - simply compare each range to the preceding
1200 : * one, and skip the duplicate ones.
1201 : */
1202 4193 : n = 1;
1203 99049 : for (i = 1; i < neranges; i++)
1204 : {
1205 : /* if the current range is equal to the preceding one, do nothing */
1206 94856 : if (!compare_expanded_ranges(&eranges[i - 1], &eranges[i], &cxt))
1207 16887 : continue;
1208 :
1209 : /* otherwise, copy it to n-th place (if not already there) */
1210 77969 : if (i != n)
1211 5992 : memcpy(&eranges[n], &eranges[i], sizeof(ExpandedRange));
1212 :
1213 77969 : n++;
1214 : }
1215 :
1216 : Assert((n > 0) && (n <= neranges));
1217 :
1218 4193 : return n;
1219 : }
1220 :
1221 : /*
1222 : * When combining multiple Range values (in union function), some of the
1223 : * ranges may overlap. We simply merge the overlapping ranges to fix that.
1224 : *
1225 : * XXX This assumes the expanded ranges were previously sorted (by minval
1226 : * and then maxval). We leverage this when detecting overlap.
1227 : */
1228 : static int
1229 0 : merge_overlapping_ranges(FmgrInfo *cmp, Oid colloid,
1230 : ExpandedRange *eranges, int neranges)
1231 : {
1232 : int idx;
1233 :
1234 : /* Merge ranges (idx) and (idx+1) if they overlap. */
1235 0 : idx = 0;
1236 0 : while (idx < (neranges - 1))
1237 : {
1238 : Datum r;
1239 :
1240 : /*
1241 : * comparing [?,maxval] vs. [minval,?] - the ranges overlap if (minval
1242 : * < maxval)
1243 : */
1244 0 : r = FunctionCall2Coll(cmp, colloid,
1245 0 : eranges[idx].maxval,
1246 0 : eranges[idx + 1].minval);
1247 :
1248 : /*
1249 : * Nope, maxval < minval, so no overlap. And we know the ranges are
1250 : * ordered, so there are no more overlaps, because all the remaining
1251 : * ranges have greater or equal minval.
1252 : */
1253 0 : if (DatumGetBool(r))
1254 : {
1255 : /* proceed to the next range */
1256 0 : idx += 1;
1257 0 : continue;
1258 : }
1259 :
1260 : /*
1261 : * So ranges 'idx' and 'idx+1' do overlap, but we don't know if
1262 : * 'idx+1' is contained in 'idx', or if they overlap only partially.
1263 : * So compare the upper bounds and keep the larger one.
1264 : */
1265 0 : r = FunctionCall2Coll(cmp, colloid,
1266 0 : eranges[idx].maxval,
1267 0 : eranges[idx + 1].maxval);
1268 :
1269 0 : if (DatumGetBool(r))
1270 0 : eranges[idx].maxval = eranges[idx + 1].maxval;
1271 :
1272 : /*
1273 : * The range certainly is no longer collapsed (irrespectively of the
1274 : * previous state).
1275 : */
1276 0 : eranges[idx].collapsed = false;
1277 :
1278 : /*
1279 : * Now get rid of the (idx+1) range entirely by shifting the remaining
1280 : * ranges by 1. There are neranges elements, and we need to move
1281 : * elements from (idx+2). That means the number of elements to move is
1282 : * [ncranges - (idx+2)].
1283 : */
1284 0 : memmove(&eranges[idx + 1], &eranges[idx + 2],
1285 0 : (neranges - (idx + 2)) * sizeof(ExpandedRange));
1286 :
1287 : /*
1288 : * Decrease the number of ranges, and repeat (with the same range, as
1289 : * it might overlap with additional ranges thanks to the merge).
1290 : */
1291 0 : neranges--;
1292 : }
1293 :
1294 0 : return neranges;
1295 : }
1296 :
1297 : /*
1298 : * Simple comparator for distance values, comparing the double value.
1299 : * This is intentionally sorting the distances in descending order, i.e.
1300 : * the longer gaps will be at the front.
1301 : */
1302 : static int
1303 114819 : compare_distances(const void *a, const void *b)
1304 : {
1305 114819 : const DistanceValue *da = a;
1306 114819 : const DistanceValue *db = b;
1307 :
1308 114819 : if (da->value < db->value)
1309 27370 : return 1;
1310 87449 : else if (da->value > db->value)
1311 18517 : return -1;
1312 :
1313 68932 : return 0;
1314 : }
1315 :
1316 : /*
1317 : * Given an array of expanded ranges, compute size of the gaps between each
1318 : * range. For neranges there are (neranges-1) gaps.
1319 : *
1320 : * We simply call the "distance" function to compute the (max-min) for pairs
1321 : * of consecutive ranges. The function may be fairly expensive, so we do that
1322 : * just once (and then use it to pick as many ranges to merge as possible).
1323 : *
1324 : * See reduce_expanded_ranges for details.
1325 : */
1326 : static DistanceValue *
1327 4193 : build_distances(FmgrInfo *distanceFn, Oid colloid,
1328 : ExpandedRange *eranges, int neranges)
1329 : {
1330 : int i;
1331 : int ndistances;
1332 : DistanceValue *distances;
1333 :
1334 : Assert(neranges > 0);
1335 :
1336 : /* If there's only a single range, there's no distance to calculate. */
1337 4193 : if (neranges == 1)
1338 0 : return NULL;
1339 :
1340 4193 : ndistances = (neranges - 1);
1341 4193 : distances = palloc0_array(DistanceValue, ndistances);
1342 :
1343 : /*
1344 : * Walk through the ranges once and compute the distance between the
1345 : * ranges so that we can sort them once.
1346 : */
1347 82162 : for (i = 0; i < ndistances; i++)
1348 : {
1349 : Datum a1,
1350 : a2,
1351 : r;
1352 :
1353 77969 : a1 = eranges[i].maxval;
1354 77969 : a2 = eranges[i + 1].minval;
1355 :
1356 : /* compute length of the gap (between max/min) */
1357 77969 : r = FunctionCall2Coll(distanceFn, colloid, a1, a2);
1358 :
1359 : /* remember the index of the gap the distance is for */
1360 77969 : distances[i].index = i;
1361 77969 : distances[i].value = DatumGetFloat8(r);
1362 : }
1363 :
1364 : /*
1365 : * Sort the distances in descending order, so that the longest gaps are at
1366 : * the front.
1367 : */
1368 4193 : qsort(distances, ndistances, sizeof(DistanceValue), compare_distances);
1369 :
1370 4193 : return distances;
1371 : }
1372 :
1373 : /*
1374 : * Builds expanded ranges for the existing ranges (and single-point ranges),
1375 : * and also the new value (which did not fit into the array). This expanded
1376 : * representation makes the processing a bit easier, as it allows handling
1377 : * ranges and points the same way.
1378 : *
1379 : * We sort and deduplicate the expanded ranges - this is necessary, because
1380 : * the points may be unsorted. And moreover the two parts (ranges and
1381 : * points) are sorted on their own.
1382 : */
1383 : static ExpandedRange *
1384 4193 : build_expanded_ranges(FmgrInfo *cmp, Oid colloid, Ranges *ranges,
1385 : int *nranges)
1386 : {
1387 : int neranges;
1388 : ExpandedRange *eranges;
1389 :
1390 : /* both ranges and points are expanded into a separate element */
1391 4193 : neranges = ranges->nranges + ranges->nvalues;
1392 :
1393 4193 : eranges = (ExpandedRange *) palloc0(neranges * sizeof(ExpandedRange));
1394 :
1395 : /* fill the expanded ranges */
1396 4193 : fill_expanded_ranges(eranges, neranges, ranges);
1397 :
1398 : /* sort and deduplicate the expanded ranges */
1399 4193 : neranges = sort_expanded_ranges(cmp, colloid, eranges, neranges);
1400 :
1401 : /* remember how many ranges we built */
1402 4193 : *nranges = neranges;
1403 :
1404 4193 : return eranges;
1405 : }
1406 :
1407 : #ifdef USE_ASSERT_CHECKING
1408 : /*
1409 : * Counts boundary values needed to store the ranges. Each single-point
1410 : * range is stored using a single value, each regular range needs two.
1411 : */
1412 : static int
1413 : count_values(ExpandedRange *cranges, int ncranges)
1414 : {
1415 : int i;
1416 : int count;
1417 :
1418 : count = 0;
1419 : for (i = 0; i < ncranges; i++)
1420 : {
1421 : if (cranges[i].collapsed)
1422 : count += 1;
1423 : else
1424 : count += 2;
1425 : }
1426 :
1427 : return count;
1428 : }
1429 : #endif
1430 :
1431 : /*
1432 : * reduce_expanded_ranges
1433 : * reduce the ranges until the number of values is low enough
1434 : *
1435 : * Combines ranges until the number of boundary values drops below the
1436 : * threshold specified by max_values. This happens by merging enough
1437 : * ranges by the distance between them.
1438 : *
1439 : * Returns the number of result ranges.
1440 : *
1441 : * We simply use the global min/max and then add boundaries for enough
1442 : * largest gaps. Each gap adds 2 values, so we simply use (target/2-1)
1443 : * distances. Then we simply sort all the values - each two values are
1444 : * a boundary of a range (possibly collapsed).
1445 : *
1446 : * XXX Some of the ranges may be collapsed (i.e. the min/max values are
1447 : * equal), but we ignore that for now. We could repeat the process,
1448 : * adding a couple more gaps recursively.
1449 : *
1450 : * XXX The ranges to merge are selected solely using the distance. But
1451 : * that may not be the best strategy, for example when multiple gaps
1452 : * are of equal (or very similar) length.
1453 : *
1454 : * Consider for example points 1, 2, 3, .., 64, which have gaps of the
1455 : * same length 1 of course. In that case, we tend to pick the first
1456 : * gap of that length, which leads to this:
1457 : *
1458 : * step 1: [1, 2], 3, 4, 5, .., 64
1459 : * step 2: [1, 3], 4, 5, .., 64
1460 : * step 3: [1, 4], 5, .., 64
1461 : * ...
1462 : *
1463 : * So in the end we'll have one "large" range and multiple small points.
1464 : * That may be fine, but it seems a bit strange and non-optimal. Maybe
1465 : * we should consider other things when picking ranges to merge - e.g.
1466 : * length of the ranges? Or perhaps randomize the choice of ranges, with
1467 : * probability inversely proportional to the distance (the gap lengths
1468 : * may be very close, but not exactly the same).
1469 : *
1470 : * XXX Or maybe we could just handle this by using random value as a
1471 : * tie-break, or by adding random noise to the actual distance.
1472 : */
1473 : static int
1474 4193 : reduce_expanded_ranges(ExpandedRange *eranges, int neranges,
1475 : DistanceValue *distances, int max_values,
1476 : FmgrInfo *cmp, Oid colloid)
1477 : {
1478 : int i;
1479 : int nvalues;
1480 : Datum *values;
1481 :
1482 : compare_context cxt;
1483 :
1484 : /* total number of gaps between ranges */
1485 4193 : int ndistances = (neranges - 1);
1486 :
1487 : /* number of gaps to keep */
1488 4193 : int keep = (max_values / 2 - 1);
1489 :
1490 : /*
1491 : * Maybe we have a sufficiently low number of ranges already?
1492 : *
1493 : * XXX This should happen before we actually do the expensive stuff like
1494 : * sorting, so maybe this should be just an assert.
1495 : */
1496 4193 : if (keep >= ndistances)
1497 3629 : return neranges;
1498 :
1499 : /* sort the values */
1500 564 : cxt.colloid = colloid;
1501 564 : cxt.cmpFn = cmp;
1502 :
1503 : /* allocate space for the boundary values */
1504 564 : nvalues = 0;
1505 564 : values = palloc_array(Datum, max_values);
1506 :
1507 : /* add the global min/max values, from the first/last range */
1508 564 : values[nvalues++] = eranges[0].minval;
1509 564 : values[nvalues++] = eranges[neranges - 1].maxval;
1510 :
1511 : /* add boundary values for enough gaps */
1512 19520 : for (i = 0; i < keep; i++)
1513 : {
1514 : /* index of the gap between (index) and (index+1) ranges */
1515 18956 : int index = distances[i].index;
1516 :
1517 : Assert((index >= 0) && ((index + 1) < neranges));
1518 :
1519 : /* add max from the preceding range, minval from the next one */
1520 18956 : values[nvalues++] = eranges[index].maxval;
1521 18956 : values[nvalues++] = eranges[index + 1].minval;
1522 :
1523 : Assert(nvalues <= max_values);
1524 : }
1525 :
1526 : /* We should have an even number of range values. */
1527 : Assert(nvalues % 2 == 0);
1528 :
1529 : /*
1530 : * Sort the values using the comparator function, and form ranges from the
1531 : * sorted result.
1532 : */
1533 564 : qsort_arg(values, nvalues, sizeof(Datum),
1534 : compare_values, &cxt);
1535 :
1536 : /* We have nvalues boundary values, which means nvalues/2 ranges. */
1537 20084 : for (i = 0; i < (nvalues / 2); i++)
1538 : {
1539 19520 : eranges[i].minval = values[2 * i];
1540 19520 : eranges[i].maxval = values[2 * i + 1];
1541 :
1542 : /* if the boundary values are the same, it's a collapsed range */
1543 39040 : eranges[i].collapsed = (compare_values(&values[2 * i],
1544 19520 : &values[2 * i + 1],
1545 19520 : &cxt) == 0);
1546 : }
1547 :
1548 564 : return (nvalues / 2);
1549 : }
1550 :
1551 : /*
1552 : * Store the boundary values from ExpandedRanges back into 'ranges' (using
1553 : * only the minimal number of values needed).
1554 : */
1555 : static void
1556 4193 : store_expanded_ranges(Ranges *ranges, ExpandedRange *eranges, int neranges)
1557 : {
1558 : int i;
1559 4193 : int idx = 0;
1560 :
1561 : /* first copy in the regular ranges */
1562 4193 : ranges->nranges = 0;
1563 37131 : for (i = 0; i < neranges; i++)
1564 : {
1565 32938 : if (!eranges[i].collapsed)
1566 : {
1567 2944 : ranges->values[idx++] = eranges[i].minval;
1568 2944 : ranges->values[idx++] = eranges[i].maxval;
1569 2944 : ranges->nranges++;
1570 : }
1571 : }
1572 :
1573 : /* now copy in the collapsed ones */
1574 4193 : ranges->nvalues = 0;
1575 37131 : for (i = 0; i < neranges; i++)
1576 : {
1577 32938 : if (eranges[i].collapsed)
1578 : {
1579 29994 : ranges->values[idx++] = eranges[i].minval;
1580 29994 : ranges->nvalues++;
1581 : }
1582 : }
1583 :
1584 : /* all the values are sorted */
1585 4193 : ranges->nsorted = ranges->nvalues;
1586 :
1587 : Assert(count_values(eranges, neranges) == 2 * ranges->nranges + ranges->nvalues);
1588 : Assert(2 * ranges->nranges + ranges->nvalues <= ranges->maxvalues);
1589 4193 : }
1590 :
1591 :
1592 : /*
1593 : * Consider freeing space in the ranges. Checks if there's space for at least
1594 : * one new value, and performs compaction if needed.
1595 : *
1596 : * Returns true if the value was actually modified.
1597 : */
1598 : static bool
1599 90304 : ensure_free_space_in_buffer(BrinDesc *bdesc, Oid colloid,
1600 : AttrNumber attno, Form_pg_attribute attr,
1601 : Ranges *range)
1602 : {
1603 : MemoryContext ctx;
1604 : MemoryContext oldctx;
1605 :
1606 : FmgrInfo *cmpFn,
1607 : *distanceFn;
1608 :
1609 : /* expanded ranges */
1610 : ExpandedRange *eranges;
1611 : int neranges;
1612 : DistanceValue *distances;
1613 :
1614 : /*
1615 : * If there is free space in the buffer, we're done without having to
1616 : * modify anything.
1617 : */
1618 90304 : if (2 * range->nranges + range->nvalues < range->maxvalues)
1619 90140 : return false;
1620 :
1621 : /* we'll certainly need the comparator, so just look it up now */
1622 164 : cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
1623 : BTLessStrategyNumber);
1624 :
1625 : /* deduplicate values, if there's an unsorted part */
1626 164 : range_deduplicate_values(range);
1627 :
1628 : /*
1629 : * Did we reduce enough free space by just the deduplication?
1630 : *
1631 : * We don't simply check against range->maxvalues again. The deduplication
1632 : * might have freed very little space (e.g. just one value), forcing us to
1633 : * do deduplication very often. In that case, it's better to do the
1634 : * compaction and reduce more space.
1635 : */
1636 164 : if (2 * range->nranges + range->nvalues <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR)
1637 0 : return true;
1638 :
1639 : /*
1640 : * We need to combine some of the existing ranges, to reduce the number of
1641 : * values we have to store.
1642 : *
1643 : * The distanceFn calls (which may internally call e.g. numeric_le) may
1644 : * allocate quite a bit of memory, and we must not leak it (we might have
1645 : * to do this repeatedly, even for a single BRIN page range). Otherwise
1646 : * we'd have problems e.g. when building new indexes. So we use a memory
1647 : * context and make sure we free the memory at the end (so if we call the
1648 : * distance function many times, it might be an issue, but meh).
1649 : */
1650 164 : ctx = AllocSetContextCreate(CurrentMemoryContext,
1651 : "minmax-multi context",
1652 : ALLOCSET_DEFAULT_SIZES);
1653 :
1654 164 : oldctx = MemoryContextSwitchTo(ctx);
1655 :
1656 : /* build the expanded ranges */
1657 164 : eranges = build_expanded_ranges(cmpFn, colloid, range, &neranges);
1658 :
1659 : /* Is the expanded representation of ranges correct? */
1660 164 : AssertCheckExpandedRanges(bdesc, colloid, attno, attr, eranges, neranges);
1661 :
1662 : /* and we'll also need the 'distance' procedure */
1663 164 : distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
1664 :
1665 : /* build array of gap distances and sort them in ascending order */
1666 164 : distances = build_distances(distanceFn, colloid, eranges, neranges);
1667 :
1668 : /*
1669 : * Combine ranges until we release at least 50% of the space. This
1670 : * threshold is somewhat arbitrary, perhaps needs tuning. We must not use
1671 : * too low or high value.
1672 : */
1673 328 : neranges = reduce_expanded_ranges(eranges, neranges, distances,
1674 164 : range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR,
1675 : cmpFn, colloid);
1676 :
1677 : /* Is the result of reducing expanded ranges correct? */
1678 164 : AssertCheckExpandedRanges(bdesc, colloid, attno, attr, eranges, neranges);
1679 :
1680 : /* Make sure we've sufficiently reduced the number of ranges. */
1681 : Assert(count_values(eranges, neranges) <= range->maxvalues * MINMAX_BUFFER_LOAD_FACTOR);
1682 :
1683 : /* decompose the expanded ranges into regular ranges and single values */
1684 164 : store_expanded_ranges(range, eranges, neranges);
1685 :
1686 164 : MemoryContextSwitchTo(oldctx);
1687 164 : MemoryContextDelete(ctx);
1688 :
1689 : /* Did we break the ranges somehow? */
1690 164 : AssertCheckRanges(range, cmpFn, colloid);
1691 :
1692 164 : return true;
1693 : }
1694 :
1695 : /*
1696 : * range_add_value
1697 : * Add the new value to the minmax-multi range.
1698 : */
1699 : static bool
1700 90304 : range_add_value(BrinDesc *bdesc, Oid colloid,
1701 : AttrNumber attno, Form_pg_attribute attr,
1702 : Ranges *ranges, Datum newval)
1703 : {
1704 : FmgrInfo *cmpFn;
1705 90304 : bool modified = false;
1706 :
1707 : /* we'll certainly need the comparator, so just look it up now */
1708 90304 : cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
1709 : BTLessStrategyNumber);
1710 :
1711 : /* comprehensive checks of the input ranges */
1712 90304 : AssertCheckRanges(ranges, cmpFn, colloid);
1713 :
1714 : /*
1715 : * Make sure there's enough free space in the buffer. We only trigger this
1716 : * when the buffer is full, which means it had to be modified as we size
1717 : * it to be larger than what is stored on disk.
1718 : *
1719 : * This needs to happen before we check if the value is contained in the
1720 : * range, because the value might be in the unsorted part, and we don't
1721 : * check that in range_contains_value. The deduplication would then move
1722 : * it to the sorted part, and we'd add the value too, which violates the
1723 : * rule that we never have duplicates with the ranges or sorted values.
1724 : *
1725 : * We might also deduplicate and recheck if the value is contained, but
1726 : * that seems like overkill. We'd need to deduplicate anyway, so why not
1727 : * do it now.
1728 : */
1729 90304 : modified = ensure_free_space_in_buffer(bdesc, colloid,
1730 : attno, attr, ranges);
1731 :
1732 : /*
1733 : * Bail out if the value already is covered by the range.
1734 : *
1735 : * We could also add values until we hit values_per_range, and then do the
1736 : * deduplication in a batch, hoping for better efficiency. But that would
1737 : * mean we actually modify the range every time, which means having to
1738 : * serialize the value, which does palloc, walks the values, copies them,
1739 : * etc. Not exactly cheap.
1740 : *
1741 : * So instead we do the check, which should be fairly cheap - assuming the
1742 : * comparator function is not very expensive.
1743 : *
1744 : * This also implies the values array can't contain duplicate values.
1745 : */
1746 90304 : if (range_contains_value(bdesc, colloid, attno, attr, ranges, newval, false))
1747 10411 : return modified;
1748 :
1749 : /* Make a copy of the value, if needed. */
1750 79893 : newval = datumCopy(newval, attr->attbyval, attr->attlen);
1751 :
1752 : /*
1753 : * If there's space in the values array, copy it in and we're done.
1754 : *
1755 : * We do want to keep the values sorted (to speed up searches), so we do a
1756 : * simple insertion sort. We could do something more elaborate, e.g. by
1757 : * sorting the values only now and then, but for small counts (e.g. when
1758 : * maxvalues is 64) this should be fine.
1759 : */
1760 79893 : ranges->values[2 * ranges->nranges + ranges->nvalues] = newval;
1761 79893 : ranges->nvalues++;
1762 :
1763 : /* If we added the first value, we can consider it as sorted. */
1764 79893 : if (ranges->nvalues == 1)
1765 3265 : ranges->nsorted = 1;
1766 :
1767 : /*
1768 : * Check we haven't broken the ordering of boundary values (checks both
1769 : * parts, but that doesn't hurt).
1770 : */
1771 79893 : AssertCheckRanges(ranges, cmpFn, colloid);
1772 :
1773 : /* Check the range contains the value we just added. */
1774 : Assert(range_contains_value(bdesc, colloid, attno, attr, ranges, newval, true));
1775 :
1776 : /* yep, we've modified the range */
1777 79893 : return true;
1778 : }
1779 :
1780 : /*
1781 : * Generate range representation of data collected during "batch mode".
1782 : * This is similar to reduce_expanded_ranges, except that we can't assume
1783 : * the values are sorted and there may be duplicate values.
1784 : */
1785 : static void
1786 12029 : compactify_ranges(BrinDesc *bdesc, Ranges *ranges, int max_values)
1787 : {
1788 : FmgrInfo *cmpFn,
1789 : *distanceFn;
1790 :
1791 : /* expanded ranges */
1792 : ExpandedRange *eranges;
1793 : int neranges;
1794 : DistanceValue *distances;
1795 :
1796 : MemoryContext ctx;
1797 : MemoryContext oldctx;
1798 :
1799 : /*
1800 : * Do we need to actually compactify anything?
1801 : *
1802 : * There are two reasons why compaction may be needed - firstly, there may
1803 : * be too many values, or some of the values may be unsorted.
1804 : */
1805 12029 : if ((ranges->nranges * 2 + ranges->nvalues <= max_values) &&
1806 11735 : (ranges->nsorted == ranges->nvalues))
1807 8000 : return;
1808 :
1809 : /* we'll certainly need the comparator, so just look it up now */
1810 4029 : cmpFn = minmax_multi_get_strategy_procinfo(bdesc, ranges->attno, ranges->typid,
1811 : BTLessStrategyNumber);
1812 :
1813 : /* and we'll also need the 'distance' procedure */
1814 4029 : distanceFn = minmax_multi_get_procinfo(bdesc, ranges->attno, PROCNUM_DISTANCE);
1815 :
1816 : /*
1817 : * The distanceFn calls (which may internally call e.g. numeric_le) may
1818 : * allocate quite a bit of memory, and we must not leak it. Otherwise,
1819 : * we'd have problems e.g. when building indexes. So we create a local
1820 : * memory context and make sure we free the memory before leaving this
1821 : * function (not after every call).
1822 : */
1823 4029 : ctx = AllocSetContextCreate(CurrentMemoryContext,
1824 : "minmax-multi context",
1825 : ALLOCSET_DEFAULT_SIZES);
1826 :
1827 4029 : oldctx = MemoryContextSwitchTo(ctx);
1828 :
1829 : /* build the expanded ranges */
1830 4029 : eranges = build_expanded_ranges(cmpFn, ranges->colloid, ranges, &neranges);
1831 :
1832 : /* build array of gap distances and sort them in ascending order */
1833 4029 : distances = build_distances(distanceFn, ranges->colloid,
1834 : eranges, neranges);
1835 :
1836 : /*
1837 : * Combine ranges until we get below max_values. We don't use any scale
1838 : * factor, because this is used during serialization, and we don't expect
1839 : * more tuples to be inserted anytime soon.
1840 : */
1841 4029 : neranges = reduce_expanded_ranges(eranges, neranges, distances,
1842 : max_values, cmpFn, ranges->colloid);
1843 :
1844 : Assert(count_values(eranges, neranges) <= max_values);
1845 :
1846 : /* transform back into regular ranges and single values */
1847 4029 : store_expanded_ranges(ranges, eranges, neranges);
1848 :
1849 : /* check all the range invariants */
1850 4029 : AssertCheckRanges(ranges, cmpFn, ranges->colloid);
1851 :
1852 4029 : MemoryContextSwitchTo(oldctx);
1853 4029 : MemoryContextDelete(ctx);
1854 : }
1855 :
1856 : Datum
1857 12974 : brin_minmax_multi_opcinfo(PG_FUNCTION_ARGS)
1858 : {
1859 : BrinOpcInfo *result;
1860 :
1861 : /*
1862 : * opaque->strategy_procinfos is initialized lazily; here it is set to
1863 : * all-uninitialized by palloc0 which sets fn_oid to InvalidOid.
1864 : */
1865 :
1866 12974 : result = palloc0(MAXALIGN(SizeofBrinOpcInfo(1)) +
1867 : sizeof(MinmaxMultiOpaque));
1868 12974 : result->oi_nstored = 1;
1869 12974 : result->oi_regular_nulls = true;
1870 12974 : result->oi_opaque = (MinmaxMultiOpaque *)
1871 12974 : MAXALIGN((char *) result + SizeofBrinOpcInfo(1));
1872 12974 : result->oi_typcache[0] = lookup_type_cache(PG_BRIN_MINMAX_MULTI_SUMMARYOID, 0);
1873 :
1874 12974 : PG_RETURN_POINTER(result);
1875 : }
1876 :
1877 : /*
1878 : * Compute the distance between two float4 values (plain subtraction).
1879 : */
1880 : Datum
1881 470 : brin_minmax_multi_distance_float4(PG_FUNCTION_ARGS)
1882 : {
1883 470 : float a1 = PG_GETARG_FLOAT4(0);
1884 470 : float a2 = PG_GETARG_FLOAT4(1);
1885 :
1886 : /* if both values are NaN, then we consider them the same */
1887 470 : if (isnan(a1) && isnan(a2))
1888 0 : PG_RETURN_FLOAT8(0.0);
1889 :
1890 : /* if one value is NaN, use infinite distance */
1891 470 : if (isnan(a1) || isnan(a2))
1892 4 : PG_RETURN_FLOAT8(get_float8_infinity());
1893 :
1894 : /*
1895 : * We know the values are range boundaries, but the range may be collapsed
1896 : * (i.e. single points), with equal values.
1897 : */
1898 : Assert(a1 <= a2);
1899 :
1900 466 : PG_RETURN_FLOAT8((double) a2 - (double) a1);
1901 : }
1902 :
1903 : /*
1904 : * Compute the distance between two float8 values (plain subtraction).
1905 : */
1906 : Datum
1907 702 : brin_minmax_multi_distance_float8(PG_FUNCTION_ARGS)
1908 : {
1909 702 : double a1 = PG_GETARG_FLOAT8(0);
1910 702 : double a2 = PG_GETARG_FLOAT8(1);
1911 :
1912 : /* if both values are NaN, then we consider them the same */
1913 702 : if (isnan(a1) && isnan(a2))
1914 0 : PG_RETURN_FLOAT8(0.0);
1915 :
1916 : /* if one value is NaN, use infinite distance */
1917 702 : if (isnan(a1) || isnan(a2))
1918 4 : PG_RETURN_FLOAT8(get_float8_infinity());
1919 :
1920 : /*
1921 : * We know the values are range boundaries, but the range may be collapsed
1922 : * (i.e. single points), with equal values.
1923 : */
1924 : Assert(a1 <= a2);
1925 :
1926 698 : PG_RETURN_FLOAT8(a2 - a1);
1927 : }
1928 :
1929 : /*
1930 : * Compute the distance between two int2 values (plain subtraction).
1931 : */
1932 : Datum
1933 682 : brin_minmax_multi_distance_int2(PG_FUNCTION_ARGS)
1934 : {
1935 682 : int16 a1 = PG_GETARG_INT16(0);
1936 682 : int16 a2 = PG_GETARG_INT16(1);
1937 :
1938 : /*
1939 : * We know the values are range boundaries, but the range may be collapsed
1940 : * (i.e. single points), with equal values.
1941 : */
1942 : Assert(a1 <= a2);
1943 :
1944 682 : PG_RETURN_FLOAT8((double) a2 - (double) a1);
1945 : }
1946 :
1947 : /*
1948 : * Compute the distance between two int4 values (plain subtraction).
1949 : */
1950 : Datum
1951 57848 : brin_minmax_multi_distance_int4(PG_FUNCTION_ARGS)
1952 : {
1953 57848 : int32 a1 = PG_GETARG_INT32(0);
1954 57848 : int32 a2 = PG_GETARG_INT32(1);
1955 :
1956 : /*
1957 : * We know the values are range boundaries, but the range may be collapsed
1958 : * (i.e. single points), with equal values.
1959 : */
1960 : Assert(a1 <= a2);
1961 :
1962 57848 : PG_RETURN_FLOAT8((double) a2 - (double) a1);
1963 : }
1964 :
1965 : /*
1966 : * Compute the distance between two int8 values (plain subtraction).
1967 : */
1968 : Datum
1969 7187 : brin_minmax_multi_distance_int8(PG_FUNCTION_ARGS)
1970 : {
1971 7187 : int64 a1 = PG_GETARG_INT64(0);
1972 7187 : int64 a2 = PG_GETARG_INT64(1);
1973 :
1974 : /*
1975 : * We know the values are range boundaries, but the range may be collapsed
1976 : * (i.e. single points), with equal values.
1977 : */
1978 : Assert(a1 <= a2);
1979 :
1980 7187 : PG_RETURN_FLOAT8((double) a2 - (double) a1);
1981 : }
1982 :
1983 : /*
1984 : * Compute the distance between two tid values (by mapping them to float8 and
1985 : * then subtracting them).
1986 : */
1987 : Datum
1988 690 : brin_minmax_multi_distance_tid(PG_FUNCTION_ARGS)
1989 : {
1990 : double da1,
1991 : da2;
1992 :
1993 690 : ItemPointer pa1 = (ItemPointer) PG_GETARG_POINTER(0);
1994 690 : ItemPointer pa2 = (ItemPointer) PG_GETARG_POINTER(1);
1995 :
1996 : /*
1997 : * We know the values are range boundaries, but the range may be collapsed
1998 : * (i.e. single points), with equal values.
1999 : */
2000 : Assert(ItemPointerCompare(pa1, pa2) <= 0);
2001 :
2002 : /*
2003 : * We use the no-check variants here, because user-supplied values may
2004 : * have (ip_posid == 0). See ItemPointerCompare.
2005 : */
2006 690 : da1 = ItemPointerGetBlockNumberNoCheck(pa1) * MaxHeapTuplesPerPage +
2007 690 : ItemPointerGetOffsetNumberNoCheck(pa1);
2008 :
2009 690 : da2 = ItemPointerGetBlockNumberNoCheck(pa2) * MaxHeapTuplesPerPage +
2010 690 : ItemPointerGetOffsetNumberNoCheck(pa2);
2011 :
2012 690 : PG_RETURN_FLOAT8(da2 - da1);
2013 : }
2014 :
2015 : /*
2016 : * Compute the distance between two numeric values (plain subtraction).
2017 : */
2018 : Datum
2019 690 : brin_minmax_multi_distance_numeric(PG_FUNCTION_ARGS)
2020 : {
2021 : Datum d;
2022 690 : Datum a1 = PG_GETARG_DATUM(0);
2023 690 : Datum a2 = PG_GETARG_DATUM(1);
2024 :
2025 : /*
2026 : * We know the values are range boundaries, but the range may be collapsed
2027 : * (i.e. single points), with equal values.
2028 : */
2029 : Assert(DatumGetBool(DirectFunctionCall2(numeric_le, a1, a2)));
2030 :
2031 690 : d = DirectFunctionCall2(numeric_sub, a2, a1); /* a2 - a1 */
2032 :
2033 690 : PG_RETURN_DATUM(DirectFunctionCall1(numeric_float8, d));
2034 : }
2035 :
2036 : /*
2037 : * Compute the approximate distance between two UUID values.
2038 : *
2039 : * XXX We do not need a perfectly accurate value, so we approximate the
2040 : * deltas (which would have to be 128-bit integers) with a 64-bit float.
2041 : * The small inaccuracies do not matter in practice, in the worst case
2042 : * we'll decide to merge ranges that are not the closest ones.
2043 : */
2044 : Datum
2045 1154 : brin_minmax_multi_distance_uuid(PG_FUNCTION_ARGS)
2046 : {
2047 : int i;
2048 1154 : float8 delta = 0;
2049 :
2050 1154 : Datum a1 = PG_GETARG_DATUM(0);
2051 1154 : Datum a2 = PG_GETARG_DATUM(1);
2052 :
2053 1154 : pg_uuid_t *u1 = DatumGetUUIDP(a1);
2054 1154 : pg_uuid_t *u2 = DatumGetUUIDP(a2);
2055 :
2056 : /*
2057 : * We know the values are range boundaries, but the range may be collapsed
2058 : * (i.e. single points), with equal values.
2059 : */
2060 : Assert(DatumGetBool(DirectFunctionCall2(uuid_le, a1, a2)));
2061 :
2062 : /* compute approximate delta as a double precision value */
2063 19618 : for (i = UUID_LEN - 1; i >= 0; i--)
2064 : {
2065 18464 : delta += (int) u2->data[i] - (int) u1->data[i];
2066 18464 : delta /= 256;
2067 : }
2068 :
2069 : Assert(delta >= 0);
2070 :
2071 1154 : PG_RETURN_FLOAT8(delta);
2072 : }
2073 :
2074 : /*
2075 : * Compute the approximate distance between two dates.
2076 : */
2077 : Datum
2078 1090 : brin_minmax_multi_distance_date(PG_FUNCTION_ARGS)
2079 : {
2080 1090 : float8 delta = 0;
2081 1090 : DateADT dateVal1 = PG_GETARG_DATEADT(0);
2082 1090 : DateADT dateVal2 = PG_GETARG_DATEADT(1);
2083 :
2084 1090 : delta = (float8) dateVal2 - (float8) dateVal1;
2085 :
2086 : Assert(delta >= 0);
2087 :
2088 1090 : PG_RETURN_FLOAT8(delta);
2089 : }
2090 :
2091 : /*
2092 : * Compute the approximate distance between two time (without tz) values.
2093 : *
2094 : * TimeADT is just an int64, so we simply subtract the values directly.
2095 : */
2096 : Datum
2097 682 : brin_minmax_multi_distance_time(PG_FUNCTION_ARGS)
2098 : {
2099 682 : float8 delta = 0;
2100 :
2101 682 : TimeADT ta = PG_GETARG_TIMEADT(0);
2102 682 : TimeADT tb = PG_GETARG_TIMEADT(1);
2103 :
2104 682 : delta = (tb - ta);
2105 :
2106 : Assert(delta >= 0);
2107 :
2108 682 : PG_RETURN_FLOAT8(delta);
2109 : }
2110 :
2111 : /*
2112 : * Compute the approximate distance between two timetz values.
2113 : *
2114 : * Simply subtracts the TimeADT (int64) values embedded in TimeTzADT.
2115 : */
2116 : Datum
2117 530 : brin_minmax_multi_distance_timetz(PG_FUNCTION_ARGS)
2118 : {
2119 530 : float8 delta = 0;
2120 :
2121 530 : TimeTzADT *ta = PG_GETARG_TIMETZADT_P(0);
2122 530 : TimeTzADT *tb = PG_GETARG_TIMETZADT_P(1);
2123 :
2124 530 : delta = (tb->time - ta->time) + (tb->zone - ta->zone) * USECS_PER_SEC;
2125 :
2126 : Assert(delta >= 0);
2127 :
2128 530 : PG_RETURN_FLOAT8(delta);
2129 : }
2130 :
2131 : /*
2132 : * Compute the distance between two timestamp values.
2133 : */
2134 : Datum
2135 1772 : brin_minmax_multi_distance_timestamp(PG_FUNCTION_ARGS)
2136 : {
2137 1772 : float8 delta = 0;
2138 :
2139 1772 : Timestamp dt1 = PG_GETARG_TIMESTAMP(0);
2140 1772 : Timestamp dt2 = PG_GETARG_TIMESTAMP(1);
2141 :
2142 1772 : delta = (float8) dt2 - (float8) dt1;
2143 :
2144 : Assert(delta >= 0);
2145 :
2146 1772 : PG_RETURN_FLOAT8(delta);
2147 : }
2148 :
2149 : /*
2150 : * Compute the distance between two interval values.
2151 : */
2152 : Datum
2153 1022 : brin_minmax_multi_distance_interval(PG_FUNCTION_ARGS)
2154 : {
2155 1022 : float8 delta = 0;
2156 :
2157 1022 : Interval *ia = PG_GETARG_INTERVAL_P(0);
2158 1022 : Interval *ib = PG_GETARG_INTERVAL_P(1);
2159 :
2160 : int64 dayfraction;
2161 : int64 days;
2162 :
2163 : /*
2164 : * Delta is (fractional) number of days between the intervals. Assume
2165 : * months have 30 days for consistency with interval_cmp_internal. We
2166 : * don't need to be exact, in the worst case we'll build a bit less
2167 : * efficient ranges. But we should not contradict interval_cmp.
2168 : */
2169 1022 : dayfraction = (ib->time % USECS_PER_DAY) - (ia->time % USECS_PER_DAY);
2170 1022 : days = (ib->time / USECS_PER_DAY) - (ia->time / USECS_PER_DAY);
2171 1022 : days += (int64) ib->day - (int64) ia->day;
2172 1022 : days += ((int64) ib->month - (int64) ia->month) * INT64CONST(30);
2173 :
2174 : /* convert to double precision */
2175 1022 : delta = (double) days + dayfraction / (double) USECS_PER_DAY;
2176 :
2177 : Assert(delta >= 0);
2178 :
2179 1022 : PG_RETURN_FLOAT8(delta);
2180 : }
2181 :
2182 : /*
2183 : * Compute the distance between two pg_lsn values.
2184 : *
2185 : * LSN is just an int64 encoding position in the stream, so just subtract
2186 : * those int64 values directly.
2187 : */
2188 : Datum
2189 690 : brin_minmax_multi_distance_pg_lsn(PG_FUNCTION_ARGS)
2190 : {
2191 690 : float8 delta = 0;
2192 :
2193 690 : XLogRecPtr lsna = PG_GETARG_LSN(0);
2194 690 : XLogRecPtr lsnb = PG_GETARG_LSN(1);
2195 :
2196 690 : delta = (lsnb - lsna);
2197 :
2198 : Assert(delta >= 0);
2199 :
2200 690 : PG_RETURN_FLOAT8(delta);
2201 : }
2202 :
2203 : /*
2204 : * Compute the distance between two macaddr values.
2205 : *
2206 : * mac addresses are treated as 6 unsigned chars, so do the same thing we
2207 : * already do for UUID values.
2208 : */
2209 : Datum
2210 530 : brin_minmax_multi_distance_macaddr(PG_FUNCTION_ARGS)
2211 : {
2212 : float8 delta;
2213 :
2214 530 : macaddr *a = PG_GETARG_MACADDR_P(0);
2215 530 : macaddr *b = PG_GETARG_MACADDR_P(1);
2216 :
2217 530 : delta = ((float8) b->f - (float8) a->f);
2218 530 : delta /= 256;
2219 :
2220 530 : delta += ((float8) b->e - (float8) a->e);
2221 530 : delta /= 256;
2222 :
2223 530 : delta += ((float8) b->d - (float8) a->d);
2224 530 : delta /= 256;
2225 :
2226 530 : delta += ((float8) b->c - (float8) a->c);
2227 530 : delta /= 256;
2228 :
2229 530 : delta += ((float8) b->b - (float8) a->b);
2230 530 : delta /= 256;
2231 :
2232 530 : delta += ((float8) b->a - (float8) a->a);
2233 530 : delta /= 256;
2234 :
2235 : Assert(delta >= 0);
2236 :
2237 530 : PG_RETURN_FLOAT8(delta);
2238 : }
2239 :
2240 : /*
2241 : * Compute the distance between two macaddr8 values.
2242 : *
2243 : * macaddr8 addresses are 8 unsigned chars, so do the same thing we
2244 : * already do for UUID values.
2245 : */
2246 : Datum
2247 690 : brin_minmax_multi_distance_macaddr8(PG_FUNCTION_ARGS)
2248 : {
2249 : float8 delta;
2250 :
2251 690 : macaddr8 *a = PG_GETARG_MACADDR8_P(0);
2252 690 : macaddr8 *b = PG_GETARG_MACADDR8_P(1);
2253 :
2254 690 : delta = ((float8) b->h - (float8) a->h);
2255 690 : delta /= 256;
2256 :
2257 690 : delta += ((float8) b->g - (float8) a->g);
2258 690 : delta /= 256;
2259 :
2260 690 : delta += ((float8) b->f - (float8) a->f);
2261 690 : delta /= 256;
2262 :
2263 690 : delta += ((float8) b->e - (float8) a->e);
2264 690 : delta /= 256;
2265 :
2266 690 : delta += ((float8) b->d - (float8) a->d);
2267 690 : delta /= 256;
2268 :
2269 690 : delta += ((float8) b->c - (float8) a->c);
2270 690 : delta /= 256;
2271 :
2272 690 : delta += ((float8) b->b - (float8) a->b);
2273 690 : delta /= 256;
2274 :
2275 690 : delta += ((float8) b->a - (float8) a->a);
2276 690 : delta /= 256;
2277 :
2278 : Assert(delta >= 0);
2279 :
2280 690 : PG_RETURN_FLOAT8(delta);
2281 : }
2282 :
2283 : /*
2284 : * Compute the distance between two inet values.
2285 : *
2286 : * The distance is defined as the difference between 32-bit/128-bit values,
2287 : * depending on the IP version. The distance is computed by subtracting
2288 : * the bytes and normalizing it to [0,1] range for each IP family.
2289 : * Addresses from different families are considered to be in maximum
2290 : * distance, which is 1.0.
2291 : *
2292 : * XXX Does this need to consider the mask (bits)? For now, it's ignored.
2293 : */
2294 : Datum
2295 1540 : brin_minmax_multi_distance_inet(PG_FUNCTION_ARGS)
2296 : {
2297 : float8 delta;
2298 : int i;
2299 : int len;
2300 : unsigned char *addra,
2301 : *addrb;
2302 :
2303 1540 : inet *ipa = PG_GETARG_INET_PP(0);
2304 1540 : inet *ipb = PG_GETARG_INET_PP(1);
2305 :
2306 : int lena,
2307 : lenb;
2308 :
2309 : /*
2310 : * If the addresses are from different families, consider them to be in
2311 : * maximal possible distance (which is 1.0).
2312 : */
2313 1540 : if (ip_family(ipa) != ip_family(ipb))
2314 126 : PG_RETURN_FLOAT8(1.0);
2315 :
2316 1414 : addra = (unsigned char *) palloc(ip_addrsize(ipa));
2317 1414 : memcpy(addra, ip_addr(ipa), ip_addrsize(ipa));
2318 :
2319 1414 : addrb = (unsigned char *) palloc(ip_addrsize(ipb));
2320 1414 : memcpy(addrb, ip_addr(ipb), ip_addrsize(ipb));
2321 :
2322 : /*
2323 : * The length is calculated from the mask length, because we sort the
2324 : * addresses by first address in the range, so A.B.C.D/24 < A.B.C.1 (the
2325 : * first range starts at A.B.C.0, which is before A.B.C.1). We don't want
2326 : * to produce a negative delta in this case, so we just cut the extra
2327 : * bytes.
2328 : *
2329 : * XXX Maybe this should be a bit more careful and cut the bits, not just
2330 : * whole bytes.
2331 : */
2332 1414 : lena = ip_bits(ipa);
2333 1414 : lenb = ip_bits(ipb);
2334 :
2335 1414 : len = ip_addrsize(ipa);
2336 :
2337 : /* apply the network mask to both addresses */
2338 10694 : for (i = 0; i < len; i++)
2339 : {
2340 : unsigned char mask;
2341 : int nbits;
2342 :
2343 9280 : nbits = Max(0, lena - (i * 8));
2344 9280 : if (nbits < 8)
2345 : {
2346 1112 : mask = (0xFF << (8 - nbits));
2347 1112 : addra[i] = (addra[i] & mask);
2348 : }
2349 :
2350 9280 : nbits = Max(0, lenb - (i * 8));
2351 9280 : if (nbits < 8)
2352 : {
2353 1108 : mask = (0xFF << (8 - nbits));
2354 1108 : addrb[i] = (addrb[i] & mask);
2355 : }
2356 : }
2357 :
2358 : /* Calculate the difference between the addresses. */
2359 1414 : delta = 0;
2360 10694 : for (i = len - 1; i >= 0; i--)
2361 : {
2362 9280 : unsigned char a = addra[i];
2363 9280 : unsigned char b = addrb[i];
2364 :
2365 9280 : delta += (float8) b - (float8) a;
2366 9280 : delta /= 256;
2367 : }
2368 :
2369 : Assert((delta >= 0) && (delta <= 1));
2370 :
2371 1414 : pfree(addra);
2372 1414 : pfree(addrb);
2373 :
2374 1414 : PG_RETURN_FLOAT8(delta);
2375 : }
2376 :
2377 : static void
2378 12029 : brin_minmax_multi_serialize(BrinDesc *bdesc, Datum src, Datum *dst)
2379 : {
2380 12029 : Ranges *ranges = (Ranges *) DatumGetPointer(src);
2381 : SerializedRanges *s;
2382 :
2383 : /*
2384 : * In batch mode, we need to compress the accumulated values to the
2385 : * actually requested number of values/ranges.
2386 : */
2387 12029 : compactify_ranges(bdesc, ranges, ranges->target_maxvalues);
2388 :
2389 : /* At this point everything has to be fully sorted. */
2390 : Assert(ranges->nsorted == ranges->nvalues);
2391 :
2392 12029 : s = brin_range_serialize(ranges);
2393 12029 : dst[0] = PointerGetDatum(s);
2394 12029 : }
2395 :
2396 : static int
2397 3265 : brin_minmax_multi_get_values(BrinDesc *bdesc, MinMaxMultiOptions *opts)
2398 : {
2399 3265 : return MinMaxMultiGetValuesPerRange(opts);
2400 : }
2401 :
2402 : /*
2403 : * Examine the given index tuple (which contains the partial status of a
2404 : * certain page range) by comparing it to the given value that comes from
2405 : * another heap tuple. If the new value is outside the min/max range
2406 : * specified by the existing tuple values, update the index tuple and return
2407 : * true. Otherwise, return false and do not modify in this case.
2408 : */
2409 : Datum
2410 90304 : brin_minmax_multi_add_value(PG_FUNCTION_ARGS)
2411 : {
2412 90304 : BrinDesc *bdesc = (BrinDesc *) PG_GETARG_POINTER(0);
2413 90304 : BrinValues *column = (BrinValues *) PG_GETARG_POINTER(1);
2414 90304 : Datum newval = PG_GETARG_DATUM(2);
2415 90304 : bool isnull PG_USED_FOR_ASSERTS_ONLY = PG_GETARG_BOOL(3);
2416 90304 : MinMaxMultiOptions *opts = (MinMaxMultiOptions *) PG_GET_OPCLASS_OPTIONS();
2417 90304 : Oid colloid = PG_GET_COLLATION();
2418 90304 : bool modified = false;
2419 : Form_pg_attribute attr;
2420 : AttrNumber attno;
2421 : Ranges *ranges;
2422 90304 : SerializedRanges *serialized = NULL;
2423 :
2424 : Assert(!isnull);
2425 :
2426 90304 : attno = column->bv_attno;
2427 90304 : attr = TupleDescAttr(bdesc->bd_tupdesc, attno - 1);
2428 :
2429 : /* use the already deserialized value, if possible */
2430 90304 : ranges = (Ranges *) DatumGetPointer(column->bv_mem_value);
2431 :
2432 : /*
2433 : * If this is the first non-null value, we need to initialize the range
2434 : * list. Otherwise, just extract the existing range list from BrinValues.
2435 : *
2436 : * When starting with an empty range, we assume this is a batch mode and
2437 : * we use a larger buffer. The buffer size is derived from the BRIN range
2438 : * size, number of rows per page, with some sensible min/max values. A
2439 : * small buffer would be bad for performance, but a large buffer might
2440 : * require a lot of memory (because of keeping all the values).
2441 : */
2442 90304 : if (column->bv_allnulls)
2443 : {
2444 : MemoryContext oldctx;
2445 :
2446 : int target_maxvalues;
2447 : int maxvalues;
2448 3265 : BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
2449 :
2450 : /* what was specified as a reloption? */
2451 3265 : target_maxvalues = brin_minmax_multi_get_values(bdesc, opts);
2452 :
2453 : /*
2454 : * Determine the insert buffer size - we use 10x the target, capped to
2455 : * the maximum number of values in the heap range. This is more than
2456 : * enough, considering the actual number of rows per page is likely
2457 : * much lower, but meh.
2458 : */
2459 3265 : maxvalues = Min(target_maxvalues * MINMAX_BUFFER_FACTOR,
2460 : MaxHeapTuplesPerPage * pagesPerRange);
2461 :
2462 : /* but always at least the original value */
2463 3265 : maxvalues = Max(maxvalues, target_maxvalues);
2464 :
2465 : /* always cap by MIN/MAX */
2466 3265 : maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
2467 3265 : maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
2468 :
2469 3265 : oldctx = MemoryContextSwitchTo(column->bv_context);
2470 3265 : ranges = minmax_multi_init(maxvalues);
2471 3265 : ranges->attno = attno;
2472 3265 : ranges->colloid = colloid;
2473 3265 : ranges->typid = attr->atttypid;
2474 3265 : ranges->target_maxvalues = target_maxvalues;
2475 :
2476 : /* we'll certainly need the comparator, so just look it up now */
2477 3265 : ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
2478 : BTLessStrategyNumber);
2479 :
2480 3265 : MemoryContextSwitchTo(oldctx);
2481 :
2482 3265 : column->bv_allnulls = false;
2483 3265 : modified = true;
2484 :
2485 3265 : column->bv_mem_value = PointerGetDatum(ranges);
2486 3265 : column->bv_serialize = brin_minmax_multi_serialize;
2487 : }
2488 87039 : else if (!ranges)
2489 : {
2490 : MemoryContext oldctx;
2491 :
2492 : int maxvalues;
2493 9480 : BlockNumber pagesPerRange = BrinGetPagesPerRange(bdesc->bd_index);
2494 :
2495 9480 : oldctx = MemoryContextSwitchTo(column->bv_context);
2496 :
2497 9480 : serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
2498 :
2499 : /*
2500 : * Determine the insert buffer size - we use 10x the target, capped to
2501 : * the maximum number of values in the heap range. This is more than
2502 : * enough, considering the actual number of rows per page is likely
2503 : * much lower, but meh.
2504 : */
2505 9480 : maxvalues = Min(serialized->maxvalues * MINMAX_BUFFER_FACTOR,
2506 : MaxHeapTuplesPerPage * pagesPerRange);
2507 :
2508 : /* but always at least the original value */
2509 9480 : maxvalues = Max(maxvalues, serialized->maxvalues);
2510 :
2511 : /* always cap by MIN/MAX */
2512 9480 : maxvalues = Max(maxvalues, MINMAX_BUFFER_MIN);
2513 9480 : maxvalues = Min(maxvalues, MINMAX_BUFFER_MAX);
2514 :
2515 9480 : ranges = brin_range_deserialize(maxvalues, serialized);
2516 :
2517 9480 : ranges->attno = attno;
2518 9480 : ranges->colloid = colloid;
2519 9480 : ranges->typid = attr->atttypid;
2520 :
2521 : /* we'll certainly need the comparator, so just look it up now */
2522 9480 : ranges->cmp = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
2523 : BTLessStrategyNumber);
2524 :
2525 9480 : column->bv_mem_value = PointerGetDatum(ranges);
2526 9480 : column->bv_serialize = brin_minmax_multi_serialize;
2527 :
2528 9480 : MemoryContextSwitchTo(oldctx);
2529 : }
2530 :
2531 : /*
2532 : * Try to add the new value to the range. We need to update the modified
2533 : * flag, so that we serialize the updated summary later.
2534 : */
2535 90304 : modified |= range_add_value(bdesc, colloid, attno, attr, ranges, newval);
2536 :
2537 :
2538 90304 : PG_RETURN_BOOL(modified);
2539 : }
2540 :
2541 : /*
2542 : * Given an index tuple corresponding to a certain page range and a scan key,
2543 : * return whether the scan key is consistent with the index tuple's min/max
2544 : * values. Return true if so, false otherwise.
2545 : */
2546 : Datum
2547 20924 : brin_minmax_multi_consistent(PG_FUNCTION_ARGS)
2548 : {
2549 20924 : BrinDesc *bdesc = (BrinDesc *) PG_GETARG_POINTER(0);
2550 20924 : BrinValues *column = (BrinValues *) PG_GETARG_POINTER(1);
2551 20924 : ScanKey *keys = (ScanKey *) PG_GETARG_POINTER(2);
2552 20924 : int nkeys = PG_GETARG_INT32(3);
2553 :
2554 20924 : Oid colloid = PG_GET_COLLATION(),
2555 : subtype;
2556 : AttrNumber attno;
2557 : Datum value;
2558 : FmgrInfo *finfo;
2559 : SerializedRanges *serialized;
2560 : Ranges *ranges;
2561 : int keyno;
2562 : int rangeno;
2563 : int i;
2564 :
2565 20924 : attno = column->bv_attno;
2566 :
2567 20924 : serialized = (SerializedRanges *) PG_DETOAST_DATUM(column->bv_values[0]);
2568 20924 : ranges = brin_range_deserialize(serialized->maxvalues, serialized);
2569 :
2570 : /* inspect the ranges, and for each one evaluate the scan keys */
2571 22672 : for (rangeno = 0; rangeno < ranges->nranges; rangeno++)
2572 : {
2573 2208 : Datum minval = ranges->values[2 * rangeno];
2574 2208 : Datum maxval = ranges->values[2 * rangeno + 1];
2575 :
2576 : /* assume the range is matching, and we'll try to prove otherwise */
2577 2208 : bool matching = true;
2578 :
2579 2668 : for (keyno = 0; keyno < nkeys; keyno++)
2580 : {
2581 : bool matches;
2582 2208 : ScanKey key = keys[keyno];
2583 :
2584 : /* NULL keys are handled and filtered-out in bringetbitmap */
2585 : Assert(!(key->sk_flags & SK_ISNULL));
2586 :
2587 2208 : attno = key->sk_attno;
2588 2208 : subtype = key->sk_subtype;
2589 2208 : value = key->sk_argument;
2590 2208 : switch (key->sk_strategy)
2591 : {
2592 612 : case BTLessStrategyNumber:
2593 : case BTLessEqualStrategyNumber:
2594 612 : finfo = minmax_multi_get_strategy_procinfo(bdesc, attno, subtype,
2595 612 : key->sk_strategy);
2596 : /* first value from the array */
2597 612 : matches = DatumGetBool(FunctionCall2Coll(finfo, colloid, minval, value));
2598 612 : break;
2599 :
2600 660 : case BTEqualStrategyNumber:
2601 : {
2602 : Datum compar;
2603 : FmgrInfo *cmpFn;
2604 :
2605 : /* by default this range does not match */
2606 660 : matches = false;
2607 :
2608 : /*
2609 : * Otherwise, need to compare the new value with
2610 : * boundaries of all the ranges. First check if it's
2611 : * less than the absolute minimum, which is the first
2612 : * value in the array.
2613 : */
2614 660 : cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, subtype,
2615 : BTGreaterStrategyNumber);
2616 660 : compar = FunctionCall2Coll(cmpFn, colloid, minval, value);
2617 :
2618 : /* smaller than the smallest value in this range */
2619 660 : if (DatumGetBool(compar))
2620 80 : break;
2621 :
2622 580 : cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, subtype,
2623 : BTLessStrategyNumber);
2624 580 : compar = FunctionCall2Coll(cmpFn, colloid, maxval, value);
2625 :
2626 : /* larger than the largest value in this range */
2627 580 : if (DatumGetBool(compar))
2628 548 : break;
2629 :
2630 : /*
2631 : * We haven't managed to eliminate this range, so
2632 : * consider it matching.
2633 : */
2634 32 : matches = true;
2635 :
2636 32 : break;
2637 : }
2638 936 : case BTGreaterEqualStrategyNumber:
2639 : case BTGreaterStrategyNumber:
2640 936 : finfo = minmax_multi_get_strategy_procinfo(bdesc, attno, subtype,
2641 936 : key->sk_strategy);
2642 : /* last value from the array */
2643 936 : matches = DatumGetBool(FunctionCall2Coll(finfo, colloid, maxval, value));
2644 936 : break;
2645 :
2646 0 : default:
2647 : /* shouldn't happen */
2648 0 : elog(ERROR, "invalid strategy number %d", key->sk_strategy);
2649 : matches = false;
2650 : break;
2651 : }
2652 :
2653 : /* the range has to match all the scan keys */
2654 2208 : matching &= matches;
2655 :
2656 : /* once we find a non-matching key, we're done */
2657 2208 : if (!matching)
2658 1748 : break;
2659 : }
2660 :
2661 : /*
2662 : * have we found a range matching all scan keys? if yes, we're done
2663 : */
2664 2208 : if (matching)
2665 460 : PG_RETURN_BOOL(true);
2666 : }
2667 :
2668 : /*
2669 : * And now inspect the values. We don't bother with doing a binary search
2670 : * here, because we're dealing with serialized / fully compacted ranges,
2671 : * so there should be only very few values.
2672 : */
2673 41525 : for (i = 0; i < ranges->nvalues; i++)
2674 : {
2675 36925 : Datum val = ranges->values[2 * ranges->nranges + i];
2676 :
2677 : /* assume the range is matching, and we'll try to prove otherwise */
2678 36925 : bool matching = true;
2679 :
2680 52789 : for (keyno = 0; keyno < nkeys; keyno++)
2681 : {
2682 : bool matches;
2683 36925 : ScanKey key = keys[keyno];
2684 :
2685 : /* we've already dealt with NULL keys at the beginning */
2686 36925 : if (key->sk_flags & SK_ISNULL)
2687 0 : continue;
2688 :
2689 36925 : attno = key->sk_attno;
2690 36925 : subtype = key->sk_subtype;
2691 36925 : value = key->sk_argument;
2692 36925 : switch (key->sk_strategy)
2693 : {
2694 36925 : case BTLessStrategyNumber:
2695 : case BTLessEqualStrategyNumber:
2696 : case BTEqualStrategyNumber:
2697 : case BTGreaterEqualStrategyNumber:
2698 : case BTGreaterStrategyNumber:
2699 :
2700 36925 : finfo = minmax_multi_get_strategy_procinfo(bdesc, attno, subtype,
2701 36925 : key->sk_strategy);
2702 36925 : matches = DatumGetBool(FunctionCall2Coll(finfo, colloid, val, value));
2703 36925 : break;
2704 :
2705 0 : default:
2706 : /* shouldn't happen */
2707 0 : elog(ERROR, "invalid strategy number %d", key->sk_strategy);
2708 : matches = false;
2709 : break;
2710 : }
2711 :
2712 : /* the range has to match all the scan keys */
2713 36925 : matching &= matches;
2714 :
2715 : /* once we find a non-matching key, we're done */
2716 36925 : if (!matching)
2717 21061 : break;
2718 : }
2719 :
2720 : /* have we found a range matching all scan keys? if yes, we're done */
2721 36925 : if (matching)
2722 15864 : PG_RETURN_BOOL(true);
2723 : }
2724 :
2725 4600 : PG_RETURN_BOOL(false);
2726 : }
2727 :
2728 : /*
2729 : * Given two BrinValues, update the first of them as a union of the summary
2730 : * values contained in both. The second one is untouched.
2731 : */
2732 : Datum
2733 0 : brin_minmax_multi_union(PG_FUNCTION_ARGS)
2734 : {
2735 0 : BrinDesc *bdesc = (BrinDesc *) PG_GETARG_POINTER(0);
2736 0 : BrinValues *col_a = (BrinValues *) PG_GETARG_POINTER(1);
2737 0 : BrinValues *col_b = (BrinValues *) PG_GETARG_POINTER(2);
2738 :
2739 0 : Oid colloid = PG_GET_COLLATION();
2740 : SerializedRanges *serialized_a;
2741 : SerializedRanges *serialized_b;
2742 : Ranges *ranges_a;
2743 : Ranges *ranges_b;
2744 : AttrNumber attno;
2745 : Form_pg_attribute attr;
2746 : ExpandedRange *eranges;
2747 : int neranges;
2748 : FmgrInfo *cmpFn,
2749 : *distanceFn;
2750 : DistanceValue *distances;
2751 : MemoryContext ctx;
2752 : MemoryContext oldctx;
2753 :
2754 : Assert(col_a->bv_attno == col_b->bv_attno);
2755 : Assert(!col_a->bv_allnulls && !col_b->bv_allnulls);
2756 :
2757 0 : attno = col_a->bv_attno;
2758 0 : attr = TupleDescAttr(bdesc->bd_tupdesc, attno - 1);
2759 :
2760 0 : serialized_a = (SerializedRanges *) PG_DETOAST_DATUM(col_a->bv_values[0]);
2761 0 : serialized_b = (SerializedRanges *) PG_DETOAST_DATUM(col_b->bv_values[0]);
2762 :
2763 0 : ranges_a = brin_range_deserialize(serialized_a->maxvalues, serialized_a);
2764 0 : ranges_b = brin_range_deserialize(serialized_b->maxvalues, serialized_b);
2765 :
2766 : /* make sure neither of the ranges is NULL */
2767 : Assert(ranges_a && ranges_b);
2768 :
2769 0 : neranges = (ranges_a->nranges + ranges_a->nvalues) +
2770 0 : (ranges_b->nranges + ranges_b->nvalues);
2771 :
2772 : /*
2773 : * The distanceFn calls (which may internally call e.g. numeric_le) may
2774 : * allocate quite a bit of memory, and we must not leak it. Otherwise,
2775 : * we'd have problems e.g. when building indexes. So we create a local
2776 : * memory context and make sure we free the memory before leaving this
2777 : * function (not after every call).
2778 : */
2779 0 : ctx = AllocSetContextCreate(CurrentMemoryContext,
2780 : "minmax-multi context",
2781 : ALLOCSET_DEFAULT_SIZES);
2782 :
2783 0 : oldctx = MemoryContextSwitchTo(ctx);
2784 :
2785 : /* allocate and fill */
2786 0 : eranges = (ExpandedRange *) palloc0(neranges * sizeof(ExpandedRange));
2787 :
2788 : /* fill the expanded ranges with entries for the first range */
2789 0 : fill_expanded_ranges(eranges, ranges_a->nranges + ranges_a->nvalues,
2790 : ranges_a);
2791 :
2792 : /* and now add combine ranges for the second range */
2793 0 : fill_expanded_ranges(&eranges[ranges_a->nranges + ranges_a->nvalues],
2794 0 : ranges_b->nranges + ranges_b->nvalues,
2795 : ranges_b);
2796 :
2797 0 : cmpFn = minmax_multi_get_strategy_procinfo(bdesc, attno, attr->atttypid,
2798 : BTLessStrategyNumber);
2799 :
2800 : /* sort the expanded ranges */
2801 0 : neranges = sort_expanded_ranges(cmpFn, colloid, eranges, neranges);
2802 :
2803 : /*
2804 : * We've loaded two different lists of expanded ranges, so some of them
2805 : * may be overlapping. So walk through them and merge them.
2806 : */
2807 0 : neranges = merge_overlapping_ranges(cmpFn, colloid, eranges, neranges);
2808 :
2809 : /* check that the combine ranges are correct (no overlaps, ordering) */
2810 0 : AssertCheckExpandedRanges(bdesc, colloid, attno, attr, eranges, neranges);
2811 :
2812 : /*
2813 : * If needed, reduce some of the ranges.
2814 : *
2815 : * XXX This may be fairly expensive, so maybe we should do it only when
2816 : * it's actually needed (when we have too many ranges).
2817 : */
2818 :
2819 : /* build array of gap distances and sort them in ascending order */
2820 0 : distanceFn = minmax_multi_get_procinfo(bdesc, attno, PROCNUM_DISTANCE);
2821 0 : distances = build_distances(distanceFn, colloid, eranges, neranges);
2822 :
2823 : /*
2824 : * See how many values would be needed to store the current ranges, and if
2825 : * needed combine as many of them to get below the threshold. The
2826 : * collapsed ranges will be stored as a single value.
2827 : *
2828 : * XXX This does not apply the load factor, as we don't expect to add more
2829 : * values to the range, so we prefer to keep as many ranges as possible.
2830 : *
2831 : * XXX Can the maxvalues be different in the two ranges? Perhaps we should
2832 : * use maximum of those?
2833 : */
2834 0 : neranges = reduce_expanded_ranges(eranges, neranges, distances,
2835 : ranges_a->maxvalues,
2836 : cmpFn, colloid);
2837 :
2838 : /* Is the result of reducing expanded ranges correct? */
2839 0 : AssertCheckExpandedRanges(bdesc, colloid, attno, attr, eranges, neranges);
2840 :
2841 : /* update the first range summary */
2842 0 : store_expanded_ranges(ranges_a, eranges, neranges);
2843 :
2844 0 : MemoryContextSwitchTo(oldctx);
2845 0 : MemoryContextDelete(ctx);
2846 :
2847 : /* cleanup and update the serialized value */
2848 0 : pfree(serialized_a);
2849 0 : col_a->bv_values[0] = PointerGetDatum(brin_range_serialize(ranges_a));
2850 :
2851 0 : PG_RETURN_VOID();
2852 : }
2853 :
2854 : /*
2855 : * Cache and return minmax multi opclass support procedure
2856 : *
2857 : * Return the procedure corresponding to the given function support number
2858 : * or null if it does not exist.
2859 : */
2860 : static FmgrInfo *
2861 4193 : minmax_multi_get_procinfo(BrinDesc *bdesc, uint16 attno, uint16 procnum)
2862 : {
2863 : MinmaxMultiOpaque *opaque;
2864 4193 : uint16 basenum = procnum - PROCNUM_BASE;
2865 :
2866 : /*
2867 : * We cache these in the opaque struct, to avoid repetitive syscache
2868 : * lookups.
2869 : */
2870 4193 : opaque = (MinmaxMultiOpaque *) bdesc->bd_info[attno - 1]->oi_opaque;
2871 :
2872 4193 : if (opaque->extra_procinfos[basenum].fn_oid == InvalidOid)
2873 : {
2874 371 : if (RegProcedureIsValid(index_getprocid(bdesc->bd_index, attno,
2875 : procnum)))
2876 371 : fmgr_info_copy(&opaque->extra_procinfos[basenum],
2877 : index_getprocinfo(bdesc->bd_index, attno, procnum),
2878 : bdesc->bd_context);
2879 : else
2880 0 : ereport(ERROR,
2881 : errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
2882 : errmsg_internal("invalid opclass definition"),
2883 : errdetail_internal("The operator class is missing support function %d for column %d.",
2884 : procnum, attno));
2885 : }
2886 :
2887 4193 : return &opaque->extra_procinfos[basenum];
2888 : }
2889 :
2890 : /*
2891 : * Cache and return the procedure for the given strategy.
2892 : *
2893 : * Note: this function mirrors minmax_multi_get_strategy_procinfo; see notes
2894 : * there. If changes are made here, see that function too.
2895 : */
2896 : static FmgrInfo *
2897 315555 : minmax_multi_get_strategy_procinfo(BrinDesc *bdesc, uint16 attno, Oid subtype,
2898 : uint16 strategynum)
2899 : {
2900 : MinmaxMultiOpaque *opaque;
2901 :
2902 : Assert(strategynum >= 1 &&
2903 : strategynum <= BTMaxStrategyNumber);
2904 :
2905 315555 : opaque = (MinmaxMultiOpaque *) bdesc->bd_info[attno - 1]->oi_opaque;
2906 :
2907 : /*
2908 : * We cache the procedures for the previous subtype in the opaque struct,
2909 : * to avoid repetitive syscache lookups. If the subtype changed,
2910 : * invalidate all the cached entries.
2911 : */
2912 315555 : if (opaque->cached_subtype != subtype)
2913 : {
2914 : uint16 i;
2915 :
2916 7722 : for (i = 1; i <= BTMaxStrategyNumber; i++)
2917 6435 : opaque->strategy_procinfos[i - 1].fn_oid = InvalidOid;
2918 1287 : opaque->cached_subtype = subtype;
2919 : }
2920 :
2921 315555 : if (opaque->strategy_procinfos[strategynum - 1].fn_oid == InvalidOid)
2922 : {
2923 : Form_pg_attribute attr;
2924 : HeapTuple tuple;
2925 : Oid opfamily,
2926 : oprid;
2927 :
2928 1874 : opfamily = bdesc->bd_index->rd_opfamily[attno - 1];
2929 1874 : attr = TupleDescAttr(bdesc->bd_tupdesc, attno - 1);
2930 1874 : tuple = SearchSysCache4(AMOPSTRATEGY, ObjectIdGetDatum(opfamily),
2931 : ObjectIdGetDatum(attr->atttypid),
2932 : ObjectIdGetDatum(subtype),
2933 : UInt16GetDatum(strategynum));
2934 1874 : if (!HeapTupleIsValid(tuple))
2935 0 : elog(ERROR, "missing operator %d(%u,%u) in opfamily %u",
2936 : strategynum, attr->atttypid, subtype, opfamily);
2937 :
2938 1874 : oprid = DatumGetObjectId(SysCacheGetAttrNotNull(AMOPSTRATEGY, tuple,
2939 : Anum_pg_amop_amopopr));
2940 1874 : ReleaseSysCache(tuple);
2941 : Assert(RegProcedureIsValid(oprid));
2942 :
2943 1874 : fmgr_info_cxt(get_opcode(oprid),
2944 1874 : &opaque->strategy_procinfos[strategynum - 1],
2945 : bdesc->bd_context);
2946 : }
2947 :
2948 315555 : return &opaque->strategy_procinfos[strategynum - 1];
2949 : }
2950 :
2951 : Datum
2952 892 : brin_minmax_multi_options(PG_FUNCTION_ARGS)
2953 : {
2954 892 : local_relopts *relopts = (local_relopts *) PG_GETARG_POINTER(0);
2955 :
2956 892 : init_local_reloptions(relopts, sizeof(MinMaxMultiOptions));
2957 :
2958 892 : add_local_int_reloption(relopts, "values_per_range", "desc",
2959 : MINMAX_MULTI_DEFAULT_VALUES_PER_PAGE, 8, 256,
2960 : offsetof(MinMaxMultiOptions, valuesPerRange));
2961 :
2962 892 : PG_RETURN_VOID();
2963 : }
2964 :
2965 : /*
2966 : * brin_minmax_multi_summary_in
2967 : * - input routine for type brin_minmax_multi_summary.
2968 : *
2969 : * brin_minmax_multi_summary is only used internally to represent summaries
2970 : * in BRIN minmax-multi indexes, so it has no operations of its own, and we
2971 : * disallow input too.
2972 : */
2973 : Datum
2974 0 : brin_minmax_multi_summary_in(PG_FUNCTION_ARGS)
2975 : {
2976 : /*
2977 : * brin_minmax_multi_summary stores the data in binary form and parsing
2978 : * text input is not needed, so disallow this.
2979 : */
2980 0 : ereport(ERROR,
2981 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
2982 : errmsg("cannot accept a value of type %s", "brin_minmax_multi_summary")));
2983 :
2984 : PG_RETURN_VOID(); /* keep compiler quiet */
2985 : }
2986 :
2987 :
2988 : /*
2989 : * brin_minmax_multi_summary_out
2990 : * - output routine for type brin_minmax_multi_summary.
2991 : *
2992 : * BRIN minmax-multi summaries are serialized into a bytea value, but we
2993 : * want to output something nicer humans can understand.
2994 : */
2995 : Datum
2996 120 : brin_minmax_multi_summary_out(PG_FUNCTION_ARGS)
2997 : {
2998 : int i;
2999 : int idx;
3000 : SerializedRanges *ranges;
3001 : Ranges *ranges_deserialized;
3002 : StringInfoData str;
3003 : bool isvarlena;
3004 : Oid outfunc;
3005 : FmgrInfo fmgrinfo;
3006 120 : ArrayBuildState *astate_values = NULL;
3007 :
3008 120 : initStringInfo(&str);
3009 120 : appendStringInfoChar(&str, '{');
3010 :
3011 : /*
3012 : * Detoast to get value with full 4B header (can't be stored in a toast
3013 : * table, but can use 1B header).
3014 : */
3015 120 : ranges = (SerializedRanges *) PG_DETOAST_DATUM(PG_GETARG_DATUM(0));
3016 :
3017 : /* lookup output func for the type */
3018 120 : getTypeOutputInfo(ranges->typid, &outfunc, &isvarlena);
3019 120 : fmgr_info(outfunc, &fmgrinfo);
3020 :
3021 : /* deserialize the range info easy-to-process pieces */
3022 120 : ranges_deserialized = brin_range_deserialize(ranges->maxvalues, ranges);
3023 :
3024 120 : appendStringInfo(&str, "nranges: %d nvalues: %d maxvalues: %d",
3025 : ranges_deserialized->nranges,
3026 : ranges_deserialized->nvalues,
3027 : ranges_deserialized->maxvalues);
3028 :
3029 : /* serialize ranges */
3030 120 : idx = 0;
3031 120 : for (i = 0; i < ranges_deserialized->nranges; i++)
3032 : {
3033 : char *a,
3034 : *b;
3035 : text *c;
3036 : StringInfoData buf;
3037 :
3038 0 : initStringInfo(&buf);
3039 :
3040 0 : a = OutputFunctionCall(&fmgrinfo, ranges_deserialized->values[idx++]);
3041 0 : b = OutputFunctionCall(&fmgrinfo, ranges_deserialized->values[idx++]);
3042 :
3043 0 : appendStringInfo(&buf, "%s ... %s", a, b);
3044 :
3045 0 : c = cstring_to_text_with_len(buf.data, buf.len);
3046 :
3047 0 : astate_values = accumArrayResult(astate_values,
3048 : PointerGetDatum(c),
3049 : false,
3050 : TEXTOID,
3051 : CurrentMemoryContext);
3052 : }
3053 :
3054 120 : if (ranges_deserialized->nranges > 0)
3055 : {
3056 : Oid typoutput;
3057 : bool typIsVarlena;
3058 : Datum val;
3059 : char *extval;
3060 :
3061 0 : getTypeOutputInfo(ANYARRAYOID, &typoutput, &typIsVarlena);
3062 :
3063 0 : val = makeArrayResult(astate_values, CurrentMemoryContext);
3064 :
3065 0 : extval = OidOutputFunctionCall(typoutput, val);
3066 :
3067 0 : appendStringInfo(&str, " ranges: %s", extval);
3068 : }
3069 :
3070 : /* serialize individual values */
3071 120 : astate_values = NULL;
3072 :
3073 1296 : for (i = 0; i < ranges_deserialized->nvalues; i++)
3074 : {
3075 : Datum a;
3076 : text *b;
3077 :
3078 1176 : a = FunctionCall1(&fmgrinfo, ranges_deserialized->values[idx++]);
3079 1176 : b = cstring_to_text(DatumGetCString(a));
3080 :
3081 1176 : astate_values = accumArrayResult(astate_values,
3082 : PointerGetDatum(b),
3083 : false,
3084 : TEXTOID,
3085 : CurrentMemoryContext);
3086 : }
3087 :
3088 120 : if (ranges_deserialized->nvalues > 0)
3089 : {
3090 : Oid typoutput;
3091 : bool typIsVarlena;
3092 : Datum val;
3093 : char *extval;
3094 :
3095 120 : getTypeOutputInfo(ANYARRAYOID, &typoutput, &typIsVarlena);
3096 :
3097 120 : val = makeArrayResult(astate_values, CurrentMemoryContext);
3098 :
3099 120 : extval = OidOutputFunctionCall(typoutput, val);
3100 :
3101 120 : appendStringInfo(&str, " values: %s", extval);
3102 : }
3103 :
3104 :
3105 120 : appendStringInfoChar(&str, '}');
3106 :
3107 120 : PG_RETURN_CSTRING(str.data);
3108 : }
3109 :
3110 : /*
3111 : * brin_minmax_multi_summary_recv
3112 : * - binary input routine for type brin_minmax_multi_summary.
3113 : */
3114 : Datum
3115 0 : brin_minmax_multi_summary_recv(PG_FUNCTION_ARGS)
3116 : {
3117 0 : ereport(ERROR,
3118 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
3119 : errmsg("cannot accept a value of type %s", "brin_minmax_multi_summary")));
3120 :
3121 : PG_RETURN_VOID(); /* keep compiler quiet */
3122 : }
3123 :
3124 : /*
3125 : * brin_minmax_multi_summary_send
3126 : * - binary output routine for type brin_minmax_multi_summary.
3127 : *
3128 : * BRIN minmax-multi summaries are serialized in a bytea value (although
3129 : * the type is named differently), so let's just send that.
3130 : */
3131 : Datum
3132 0 : brin_minmax_multi_summary_send(PG_FUNCTION_ARGS)
3133 : {
3134 0 : return byteasend(fcinfo);
3135 : }
|
