dbsp 0.287.0

Continuous streaming analytics engine
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
//! A binary heap implementation optimized for DBSP.
//!
//! This implementation was shamelessly stole from the standard library and extended with
//! several operations necessary to efficiently implement DBSP-style merge operations. In DBSP,
//! when merging sorted batches, we need to simultaneously peek all cursors the the max key or
//! value. The standard heap implementation only allows peeking a single max element at a time;
//! the next max can only be observed by removing or updating the current max first.
//!
//! We add:
//! - `peek_all` operation, which peeks all max elements at once by performing
//!   `num_max * 2` comparisons.
//! - `remove` operation, which removes an element from the heap by index.
//! - `update_pos_sift_down` operation, which updates an element at a given
//!   index to a smaller value and sifting it down.

// SPDX-License-Identifier: MIT OR Apache-2.0
// Copyright (c) 2026 Feldera Inc.
// Copyright (c) The Rust Project Developers

use std::cmp::Ordering;
use std::iter::FusedIterator;
use std::mem::{self, ManuallyDrop, swap};
use std::num::NonZero;
use std::ops::{Deref, DerefMut};
use std::{fmt, ptr};

use std::collections::TryReserveError;
use std::slice;
use std::vec::{self, Vec};

pub struct BinaryHeap<T, C> {
    data: Vec<T>,
    cmp: C,
}

/// Structure wrapping a mutable reference to the greatest item on a
/// `BinaryHeap`.
///
/// This `struct` is created by the [`peek_mut`] method on [`BinaryHeap`]. See
/// its documentation for more.
///
/// [`peek_mut`]: BinaryHeap::peek_mut
pub struct PeekMut<'a, T: 'a, C>
where
    C: Fn(&T, &T) -> Ordering,
{
    heap: &'a mut BinaryHeap<T, C>,
    // If a set_len + sift_down are required, this is Some. If a &mut T has not
    // yet been exposed to peek_mut()'s caller, it's None.
    original_len: Option<NonZero<usize>>,
}

impl<T: fmt::Debug, C> fmt::Debug for PeekMut<'_, T, C>
where
    C: Fn(&T, &T) -> Ordering,
{
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.debug_tuple("PeekMut").field(&self.heap.data[0]).finish()
    }
}

impl<T, C> Drop for PeekMut<'_, T, C>
where
    C: Fn(&T, &T) -> Ordering,
{
    fn drop(&mut self) {
        if let Some(original_len) = self.original_len {
            // SAFETY: That's how many elements were in the Vec at the time of
            // the PeekMut::deref_mut call, and therefore also at the time of
            // the BinaryHeap::peek_mut call. Since the PeekMut did not end up
            // getting leaked, we are now undoing the leak amplification that
            // the DerefMut prepared for.
            unsafe { self.heap.data.set_len(original_len.get()) };

            // SAFETY: PeekMut is only instantiated for non-empty heaps.
            unsafe { self.heap.sift_down(0) };
        }
    }
}

impl<T, C> Deref for PeekMut<'_, T, C>
where
    C: Fn(&T, &T) -> Ordering,
{
    type Target = T;
    fn deref(&self) -> &T {
        debug_assert!(!self.heap.is_empty());
        // SAFE: PeekMut is only instantiated for non-empty heaps
        unsafe { self.heap.data.get_unchecked(0) }
    }
}

impl<T, C> DerefMut for PeekMut<'_, T, C>
where
    C: Fn(&T, &T) -> Ordering,
{
    fn deref_mut(&mut self) -> &mut T {
        debug_assert!(!self.heap.is_empty());

        let len = self.heap.len();
        if len > 1 {
            // Here we preemptively leak all the rest of the underlying vector
            // after the currently max element. If the caller mutates the &mut T
            // we're about to give them, and then leaks the PeekMut, all these
            // elements will remain leaked. If they don't leak the PeekMut, then
            // either Drop or PeekMut::pop will un-leak the vector elements.
            //
            // This is technique is described throughout several other places in
            // the standard library as "leak amplification".
            unsafe {
                // SAFETY: len > 1 so len != 0.
                self.original_len = Some(NonZero::new_unchecked(len));
                // SAFETY: len > 1 so all this does for now is leak elements,
                // which is safe.
                self.heap.data.set_len(1);
            }
        }

        // SAFE: PeekMut is only instantiated for non-empty heaps
        unsafe { self.heap.data.get_unchecked_mut(0) }
    }
}

impl<'a, T, C> PeekMut<'a, T, C>
where
    C: Fn(&T, &T) -> Ordering,
{
    /// Sifts the current element to its new position.
    ///
    /// Afterwards refers to the new element. Returns if the element changed.
    #[must_use = "is equivalent to dropping and getting a new PeekMut except for return information"]
    pub fn refresh(&mut self) -> bool {
        // The length of the underlying heap is unchanged by sifting down. The value stored for leak
        // amplification thus remains accurate. We erase the leak amplification firstly because the
        // operation is then equivalent to constructing a new PeekMut and secondly this avoids any
        // future complication where original_len being non-empty would be interpreted as the heap
        // having been leak amplified instead of checking the heap itself.
        if let Some(original_len) = self.original_len.take() {
            // SAFETY: This is how many elements were in the Vec at the time of
            // the BinaryHeap::peek_mut call.
            unsafe { self.heap.data.set_len(original_len.get()) };

            // The length of the heap did not change by sifting, upholding our own invariants.

            // SAFETY: PeekMut is only instantiated for non-empty heaps.
            (unsafe { self.heap.sift_down(0) }) != 0
        } else {
            // The element was not modified.
            false
        }
    }

    /// Removes the peeked value from the heap and returns it.
    pub fn pop(mut this: PeekMut<'a, T, C>) -> T {
        if let Some(original_len) = this.original_len.take() {
            // SAFETY: This is how many elements were in the Vec at the time of
            // the BinaryHeap::peek_mut call.
            unsafe { this.heap.data.set_len(original_len.get()) };

            // Unlike in Drop, here we don't also need to do a sift_down even if
            // the caller could've mutated the element. It is removed from the
            // heap on the next line and pop() is not sensitive to its value.
        }

        // SAFETY: Have a `PeekMut` element proves that the associated binary heap being non-empty,
        // so the `pop` operation will not fail.
        unsafe { this.heap.pop().unwrap_unchecked() }
    }
}

impl<T: Clone, C: Clone> Clone for BinaryHeap<T, C> {
    fn clone(&self) -> Self {
        BinaryHeap {
            data: self.data.clone(),
            cmp: self.cmp.clone(),
        }
    }

    /// Overwrites the contents of `self` with a clone of the contents of `source`.
    ///
    /// This method is preferred over simply assigning `source.clone()` to `self`,
    /// as it avoids reallocation if possible.
    ///
    /// See [`Vec::clone_from()`] for more details.
    fn clone_from(&mut self, source: &Self) {
        self.data.clone_from(&source.data);
    }
}

impl<T: fmt::Debug, C> fmt::Debug for BinaryHeap<T, C> {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.debug_list().entries(self.iter()).finish()
    }
}

struct RebuildOnDrop<'a, T, C>
where
    C: Fn(&T, &T) -> Ordering,
{
    heap: &'a mut BinaryHeap<T, C>,
    rebuild_from: usize,
}

impl<T, C> Drop for RebuildOnDrop<'_, T, C>
where
    C: Fn(&T, &T) -> Ordering,
{
    fn drop(&mut self) {
        self.heap.rebuild_tail(self.rebuild_from);
    }
}

impl<T, C> BinaryHeap<T, C>
where
    C: Fn(&T, &T) -> Ordering,
{
    pub const fn new(cmp: C) -> BinaryHeap<T, C> {
        BinaryHeap { data: vec![], cmp }
    }

    pub fn from_vec(vec: Vec<T>, cmp: C) -> BinaryHeap<T, C> {
        let mut heap = BinaryHeap { data: vec, cmp };
        heap.rebuild();
        heap
    }

    /// Constructs a binary heap from a vector without checking if it's a valid heap.
    ///
    /// # Safety
    ///
    /// `vec` must be a valid binary heap built using `cmp`.
    pub unsafe fn from_vec_unchecked(vec: Vec<T>, cmp: C) -> BinaryHeap<T, C> {
        BinaryHeap { data: vec, cmp }
    }

    #[must_use]
    pub fn with_capacity(capacity: usize, cmp: C) -> BinaryHeap<T, C> {
        BinaryHeap {
            data: Vec::with_capacity(capacity),
            cmp,
        }
    }
}

impl<T, C> BinaryHeap<T, C>
where
    C: Fn(&T, &T) -> Ordering,
{
    /// Returns a mutable reference to the greatest item in the binary heap, or
    /// `None` if it is empty.
    ///
    /// Note: If the `PeekMut` value is leaked, some heap elements might get
    /// leaked along with it, but the remaining elements will remain a valid
    /// heap.
    ///
    /// # Time complexity
    ///
    /// If the item is modified then the worst case time complexity is *O*(log(*n*)),
    /// otherwise it's *O*(1).
    pub fn peek_mut(&mut self) -> Option<PeekMut<'_, T, C>> {
        if self.is_empty() {
            None
        } else {
            Some(PeekMut {
                heap: self,
                original_len: None,
            })
        }
    }

    /// Removes the greatest item from the binary heap and returns it, or `None` if it
    /// is empty.
    ///
    /// # Time complexity
    ///
    /// The worst case cost of `pop` on a heap containing *n* elements is *O*(log(*n*)).
    pub fn pop(&mut self) -> Option<T> {
        self.data.pop().map(|mut item| {
            if !self.is_empty() {
                swap(&mut item, &mut self.data[0]);
                // SAFETY: !self.is_empty() means that self.len() > 0
                unsafe { self.sift_down_to_bottom(0) };
            }
            item
        })
    }

    /// Replaces the element at position `pos` with `item` and sifts it down.
    ///
    /// # Safety
    ///
    /// `item` must be <= the element at position `pos`.
    pub unsafe fn update_pos_sift_down(&mut self, pos: usize, item: T) {
        debug_assert!((self.cmp)(&item, &self.data[pos]) != Ordering::Greater);
        self.data[pos] = item;
        unsafe { self.sift_down(pos) };
    }

    /// Peek all max elements.
    ///
    /// Invokes `cb` for every element equal to the current maximum value.
    /// The first argument to `cb` is the index of the element in the heap's internal vector.
    /// The second argument is a reference to the element.
    ///
    /// The traversal prunes subtrees as soon as a node is smaller than the
    /// maximum, so work is proportional to the number of maximal elements.
    ///
    /// Elements are visited in breadth-first order: the root is visited first,
    /// followed by its children, then its grandchildren, etc.
    ///
    /// The `queue` argument is used as scratch space for the traversal
    /// to avoid a dynamic allocation.
    pub fn peek_all(&self, mut cb: impl FnMut(usize, &T), queue: &mut Vec<usize>) {
        let Some(max) = self.peek() else {
            return;
        };

        // Root is known to be equal to `max`; avoid a redundant self-compare.
        cb(0, max);

        let mut head = 0usize;
        queue.clear();

        let len = self.data.len();

        if len > 1 {
            queue.push(1);
        }
        if len > 2 {
            queue.push(2);
        }

        while let Some(&index) = queue.get(head) {
            head += 1;

            // SAFETY: values written to `queue` are valid indices into `self.data`.
            let node = unsafe { self.data.get_unchecked(index) };
            if (self.cmp)(node, max) != Ordering::Equal {
                continue;
            }

            cb(index, node);

            let left = index * 2 + 1;
            if left < len {
                queue.push(left);
            }

            let right = left + 1;
            if right < len {
                queue.push(right);
            }
        }
    }

    /// Removes the element at position `index` from the binary heap.
    ///
    /// # Panics
    ///
    /// Panics if `index` is out of bounds.
    pub fn remove(&mut self, index: usize) {
        let last = self.data.pop().expect("index is in bounds");

        // Removing the last element needs no reheapification.
        if index == self.data.len() {
            return;
        }

        self.data[index] = last;

        // SAFETY: `index < self.len()` after the previous line.
        unsafe {
            self.sift_down_to_bottom(index);
        }
    }

    /// Pushes an item onto the binary heap.
    ///
    /// # Time complexity
    ///
    /// The expected cost of `push`, averaged over every possible ordering of
    /// the elements being pushed, and over a sufficiently large number of
    /// pushes, is *O*(1). This is the most meaningful cost metric when pushing
    /// elements that are *not* already in any sorted pattern.
    ///
    /// The time complexity degrades if elements are pushed in predominantly
    /// ascending order. In the worst case, elements are pushed in ascending
    /// sorted order and the amortized cost per push is *O*(log(*n*)) against a heap
    /// containing *n* elements.
    ///
    /// The worst case cost of a *single* call to `push` is *O*(*n*). The worst case
    /// occurs when capacity is exhausted and needs a resize. The resize cost
    /// has been amortized in the previous figures.
    pub fn push(&mut self, item: T) {
        let old_len = self.len();
        self.data.push(item);
        // SAFETY: Since we pushed a new item it means that
        //  old_len = self.len() - 1 < self.len()
        unsafe { self.sift_up(0, old_len) };
    }

    /// Consumes the `BinaryHeap` and returns a vector in sorted
    /// (ascending) order.
    ///
    pub fn into_sorted_vec(mut self) -> Vec<T> {
        let mut end = self.len();
        while end > 1 {
            end -= 1;
            // SAFETY: `end` goes from `self.len() - 1` to 1 (both included),
            //  so it's always a valid index to access.
            //  It is safe to access index 0 (i.e. `ptr`), because
            //  1 <= end < self.len(), which means self.len() >= 2.
            unsafe {
                let ptr = self.data.as_mut_ptr();
                ptr::swap(ptr, ptr.add(end));
            }
            // SAFETY: `end` goes from `self.len() - 1` to 1 (both included) so:
            //  0 < 1 <= end <= self.len() - 1 < self.len()
            //  Which means 0 < end and end < self.len().
            unsafe { self.sift_down_range(0, end) };
        }
        self.into_vec()
    }

    // The implementations of sift_up and sift_down use unsafe blocks in
    // order to move an element out of the vector (leaving behind a
    // hole), shift along the others and move the removed element back into the
    // vector at the final location of the hole.
    // The `Hole` type is used to represent this, and make sure
    // the hole is filled back at the end of its scope, even on panic.
    // Using a hole reduces the constant factor compared to using swaps,
    // which involves twice as many moves.

    /// # Safety
    ///
    /// The caller must guarantee that `pos < self.len()`.
    ///
    /// Returns the new position of the element.
    unsafe fn sift_up(&mut self, start: usize, pos: usize) -> usize {
        // Take out the value at `pos` and create a hole.
        // SAFETY: The caller guarantees that pos < self.len()
        let mut hole = unsafe { Hole::new(&mut self.data, pos) };

        while hole.pos() > start {
            let parent = (hole.pos() - 1) / 2;

            // SAFETY: hole.pos() > start >= 0, which means hole.pos() > 0
            //  and so hole.pos() - 1 can't underflow.
            //  This guarantees that parent < hole.pos() so
            //  it's a valid index and also != hole.pos().
            if (self.cmp)(hole.element(), unsafe { hole.get(parent) }) != Ordering::Greater {
                break;
            }

            // SAFETY: Same as above
            unsafe { hole.move_to(parent) };
        }

        hole.pos()
    }

    /// Take an element at `pos` and move it down the heap,
    /// while its children are larger.
    ///
    /// Returns the new position of the element.
    ///
    /// # Safety
    ///
    /// The caller must guarantee that `pos < end <= self.len()`.
    unsafe fn sift_down_range(&mut self, pos: usize, end: usize) -> usize {
        // SAFETY: The caller guarantees that pos < end <= self.len().
        let mut hole = unsafe { Hole::new(&mut self.data, pos) };
        let mut child = 2 * hole.pos() + 1;

        // Loop invariant: child == 2 * hole.pos() + 1.
        while child <= end.saturating_sub(2) {
            // compare with the greater of the two children
            // SAFETY: child < end - 1 < self.len() and
            //  child + 1 < end <= self.len(), so they're valid indexes.
            //  child == 2 * hole.pos() + 1 != hole.pos() and
            //  child + 1 == 2 * hole.pos() + 2 != hole.pos().
            // FIXME: 2 * hole.pos() + 1 or 2 * hole.pos() + 2 could overflow
            //  if T is a ZST
            child += ((self.cmp)(unsafe { hole.get(child) }, unsafe { hole.get(child + 1) })
                != Ordering::Greater) as usize;

            // if we are already in order, stop.
            // SAFETY: child is now either the old child or the old child+1
            //  We already proven that both are < self.len() and != hole.pos()
            if (self.cmp)(hole.element(), unsafe { hole.get(child) }) != Ordering::Less {
                return hole.pos();
            }

            // SAFETY: same as above.
            unsafe { hole.move_to(child) };
            child = 2 * hole.pos() + 1;
        }

        // SAFETY: && short circuit, which means that in the
        //  second condition it's already true that child == end - 1 < self.len().
        if child == end - 1
            && (self.cmp)(hole.element(), unsafe { hole.get(child) }) == Ordering::Less
        {
            // SAFETY: child is already proven to be a valid index and
            //  child == 2 * hole.pos() + 1 != hole.pos().
            unsafe { hole.move_to(child) };
        }

        hole.pos()
    }

    /// # Safety
    ///
    /// The caller must guarantee that `pos < self.len()`.
    pub unsafe fn sift_down(&mut self, pos: usize) -> usize {
        let len = self.len();
        // SAFETY: pos < len is guaranteed by the caller and
        //  obviously len = self.len() <= self.len().
        unsafe { self.sift_down_range(pos, len) }
    }

    /// Take an element at `pos` and move it all the way down the heap,
    /// then sift it up to its position.
    ///
    /// Note: This is faster when the element is known to be large / should
    /// be closer to the bottom.
    ///
    /// # Safety
    ///
    /// The caller must guarantee that `pos < self.len()`.
    pub unsafe fn sift_down_to_bottom(&mut self, mut pos: usize) {
        let end = self.len();
        let start = pos;

        // SAFETY: The caller guarantees that pos < self.len().
        let mut hole = unsafe { Hole::new(&mut self.data, pos) };
        let mut child = 2 * hole.pos() + 1;

        // Loop invariant: child == 2 * hole.pos() + 1.
        while child <= end.saturating_sub(2) {
            // SAFETY: child < end - 1 < self.len() and
            //  child + 1 < end <= self.len(), so they're valid indexes.
            //  child == 2 * hole.pos() + 1 != hole.pos() and
            //  child + 1 == 2 * hole.pos() + 2 != hole.pos().
            // FIXME: 2 * hole.pos() + 1 or 2 * hole.pos() + 2 could overflow
            //  if T is a ZST
            child += ((self.cmp)(unsafe { hole.get(child) }, unsafe { hole.get(child + 1) })
                != Ordering::Greater) as usize;

            // SAFETY: Same as above
            unsafe { hole.move_to(child) };
            child = 2 * hole.pos() + 1;
        }

        if child == end - 1 {
            // SAFETY: child == end - 1 < self.len(), so it's a valid index
            //  and child == 2 * hole.pos() + 1 != hole.pos().
            unsafe { hole.move_to(child) };
        }
        pos = hole.pos();
        drop(hole);

        // SAFETY: pos is the position in the hole and was already proven
        //  to be a valid index.
        unsafe { self.sift_up(start, pos) };
    }

    /// Rebuild assuming data[0..start] is still a proper heap.
    fn rebuild_tail(&mut self, start: usize) {
        if start == self.len() {
            return;
        }

        let tail_len = self.len() - start;

        #[inline(always)]
        fn log2_fast(x: usize) -> usize {
            (usize::BITS - x.leading_zeros() - 1) as usize
        }

        // `rebuild` takes O(self.len()) operations
        // and about 2 * self.len() comparisons in the worst case
        // while repeating `sift_up` takes O(tail_len * log(start)) operations
        // and about 1 * tail_len * log_2(start) comparisons in the worst case,
        // assuming start >= tail_len. For larger heaps, the crossover point
        // no longer follows this reasoning and was determined empirically.
        let better_to_rebuild = if start < tail_len {
            true
        } else if self.len() <= 2048 {
            2 * self.len() < tail_len * log2_fast(start)
        } else {
            2 * self.len() < tail_len * 11
        };

        if better_to_rebuild {
            self.rebuild();
        } else {
            for i in start..self.len() {
                // SAFETY: The index `i` is always less than self.len().
                unsafe { self.sift_up(0, i) };
            }
        }
    }

    fn rebuild(&mut self) {
        let mut n = self.len() / 2;
        while n > 0 {
            n -= 1;
            // SAFETY: n starts from self.len() / 2 and goes down to 0.
            //  The only case when !(n < self.len()) is if
            //  self.len() == 0, but it's ruled out by the loop condition.
            unsafe { self.sift_down(n) };
        }
    }

    /// Moves all the elements of `other` into `self`, leaving `other` empty.
    pub fn append(&mut self, other: &mut Self) {
        if self.len() < other.len() {
            swap(self, other);
        }

        let start = self.data.len();

        self.data.append(&mut other.data);

        self.rebuild_tail(start);
    }

    /// Clears the binary heap, returning an iterator over the removed elements
    /// in heap order. If the iterator is dropped before being fully consumed,
    /// it drops the remaining elements in heap order.
    ///
    /// The returned iterator keeps a mutable borrow on the heap to optimize
    /// its implementation.
    ///
    /// Note:
    /// * `.drain_sorted()` is *O*(*n* \* log(*n*)); much slower than `.drain()`.
    ///   You should use the latter for most cases.
    #[inline]
    pub fn drain_sorted(&mut self) -> DrainSorted<'_, T, C> {
        DrainSorted { inner: self }
    }

    /// Retains only the elements specified by the predicate.
    ///
    /// In other words, remove all elements `e` for which `f(&e)` returns
    /// `false`. The elements are visited in unsorted (and unspecified) order.
    pub fn retain<F>(&mut self, mut f: F)
    where
        F: FnMut(&T) -> bool,
    {
        // rebuild_start will be updated to the first touched element below, and the rebuild will
        // only be done for the tail.
        let mut guard = RebuildOnDrop {
            rebuild_from: self.len(),
            heap: self,
        };
        let mut i = 0;

        guard.heap.data.retain(|e| {
            let keep = f(e);
            if !keep && i < guard.rebuild_from {
                guard.rebuild_from = i;
            }
            i += 1;
            keep
        });
    }
}

impl<T, C> BinaryHeap<T, C> {
    /// Returns an iterator visiting all values in the underlying vector, in
    /// arbitrary order.
    pub fn iter(&self) -> Iter<'_, T> {
        Iter {
            iter: self.data.iter(),
        }
    }

    /// Returns an iterator which retrieves elements in heap order.
    ///
    /// This method consumes the original heap.
    pub fn into_iter_sorted(self) -> IntoIterSorted<T, C> {
        IntoIterSorted { inner: self }
    }

    /// Returns the greatest item in the binary heap, or `None` if it is empty.
    ///
    /// # Time complexity
    ///
    /// Cost is *O*(1) in the worst case.
    #[must_use]
    pub fn peek(&self) -> Option<&T> {
        self.data.first()
    }

    /// Returns the number of elements the binary heap can hold without reallocating.
    #[must_use]
    pub fn capacity(&self) -> usize {
        self.data.capacity()
    }

    /// Reserves the minimum capacity for at least `additional` elements more than
    /// the current length. Unlike [`reserve`], this will not
    /// deliberately over-allocate to speculatively avoid frequent allocations.
    /// After calling `reserve_exact`, capacity will be greater than or equal to
    /// `self.len() + additional`. Does nothing if the capacity is already
    /// sufficient.
    ///
    /// [`reserve`]: BinaryHeap::reserve
    ///
    /// # Panics
    ///
    /// Panics if the new capacity overflows [`usize`].
    pub fn reserve_exact(&mut self, additional: usize) {
        self.data.reserve_exact(additional);
    }

    /// Reserves capacity for at least `additional` elements more than the
    /// current length. The allocator may reserve more space to speculatively
    /// avoid frequent allocations. After calling `reserve`,
    /// capacity will be greater than or equal to `self.len() + additional`.
    /// Does nothing if capacity is already sufficient.
    ///
    /// # Panics
    ///
    /// Panics if the new capacity overflows [`usize`].
    pub fn reserve(&mut self, additional: usize) {
        self.data.reserve(additional);
    }

    /// Tries to reserve the minimum capacity for at least `additional` elements
    /// more than the current length. Unlike [`try_reserve`], this will not
    /// deliberately over-allocate to speculatively avoid frequent allocations.
    /// After calling `try_reserve_exact`, capacity will be greater than or
    /// equal to `self.len() + additional` if it returns `Ok(())`.
    /// Does nothing if the capacity is already sufficient.
    ///
    /// Note that the allocator may give the collection more space than it
    /// requests. Therefore, capacity can not be relied upon to be precisely
    /// minimal. Prefer [`try_reserve`] if future insertions are expected.
    ///
    /// [`try_reserve`]: BinaryHeap::try_reserve
    ///
    /// # Errors
    ///
    /// If the capacity overflows, or the allocator reports a failure, then an error
    /// is returned.
    pub fn try_reserve_exact(&mut self, additional: usize) -> Result<(), TryReserveError> {
        self.data.try_reserve_exact(additional)
    }

    /// Tries to reserve capacity for at least `additional` elements more than the
    /// current length. The allocator may reserve more space to speculatively
    /// avoid frequent allocations. After calling `try_reserve`, capacity will be
    /// greater than or equal to `self.len() + additional` if it returns
    /// `Ok(())`. Does nothing if capacity is already sufficient. This method
    /// preserves the contents even if an error occurs.
    ///
    /// # Errors
    ///
    /// If the capacity overflows, or the allocator reports a failure, then an error
    /// is returned.
    pub fn try_reserve(&mut self, additional: usize) -> Result<(), TryReserveError> {
        self.data.try_reserve(additional)
    }

    /// Discards as much additional capacity as possible.
    pub fn shrink_to_fit(&mut self) {
        self.data.shrink_to_fit();
    }

    /// Discards capacity with a lower bound.
    ///
    /// The capacity will remain at least as large as both the length
    /// and the supplied value.
    ///
    /// If the current capacity is less than the lower limit, this is a no-op.
    #[inline]
    pub fn shrink_to(&mut self, min_capacity: usize) {
        self.data.shrink_to(min_capacity)
    }

    /// Returns a slice of all values in the underlying vector, in arbitrary
    /// order.
    #[must_use]
    pub fn as_slice(&self) -> &[T] {
        self.data.as_slice()
    }

    /// Consumes the `BinaryHeap` and returns the underlying vector
    /// in arbitrary order.
    #[must_use = "`self` will be dropped if the result is not used"]
    pub fn into_vec(self) -> Vec<T> {
        self.into()
    }

    /// Returns the length of the binary heap.
    #[must_use]
    pub fn len(&self) -> usize {
        self.data.len()
    }

    /// Checks if the binary heap is empty.
    #[must_use]
    pub fn is_empty(&self) -> bool {
        self.len() == 0
    }

    /// Clears the binary heap, returning an iterator over the removed elements
    /// in arbitrary order. If the iterator is dropped before being fully
    /// consumed, it drops the remaining elements in arbitrary order.
    ///
    /// The returned iterator keeps a mutable borrow on the heap to optimize
    /// its implementation.
    #[inline]
    pub fn drain(&mut self) -> Drain<'_, T> {
        Drain {
            iter: self.data.drain(..),
        }
    }

    /// Drops all items from the binary heap.
    pub fn clear(&mut self) {
        self.drain();
    }
}

/// Hole represents a hole in a slice i.e., an index without valid value
/// (because it was moved from or duplicated).
/// In drop, `Hole` will restore the slice by filling the hole
/// position with the value that was originally removed.
struct Hole<'a, T: 'a> {
    data: &'a mut [T],
    elt: ManuallyDrop<T>,
    pos: usize,
}

impl<'a, T> Hole<'a, T> {
    /// Creates a new `Hole` at index `pos`.
    ///
    /// Unsafe because pos must be within the data slice.
    #[inline]
    unsafe fn new(data: &'a mut [T], pos: usize) -> Self {
        debug_assert!(pos < data.len());
        // SAFE: pos should be inside the slice
        let elt = unsafe { ptr::read(data.get_unchecked(pos)) };
        Hole {
            data,
            elt: ManuallyDrop::new(elt),
            pos,
        }
    }

    #[inline]
    fn pos(&self) -> usize {
        self.pos
    }

    /// Returns a reference to the element removed.
    #[inline]
    fn element(&self) -> &T {
        &self.elt
    }

    /// Returns a reference to the element at `index`.
    ///
    /// Unsafe because index must be within the data slice and not equal to pos.
    #[inline]
    unsafe fn get(&self, index: usize) -> &T {
        debug_assert!(index != self.pos);
        debug_assert!(index < self.data.len());
        unsafe { self.data.get_unchecked(index) }
    }

    /// Move hole to new location
    ///
    /// Unsafe because index must be within the data slice and not equal to pos.
    #[inline]
    unsafe fn move_to(&mut self, index: usize) {
        debug_assert!(index != self.pos);
        debug_assert!(index < self.data.len());
        unsafe {
            let ptr = self.data.as_mut_ptr();
            let index_ptr: *const _ = ptr.add(index);
            let hole_ptr = ptr.add(self.pos);
            ptr::copy_nonoverlapping(index_ptr, hole_ptr, 1);
        }
        self.pos = index;
    }
}

impl<T> Drop for Hole<'_, T> {
    #[inline]
    fn drop(&mut self) {
        // fill the hole again
        unsafe {
            let pos = self.pos;
            ptr::copy_nonoverlapping(&*self.elt, self.data.get_unchecked_mut(pos), 1);
        }
    }
}

/// An iterator over the elements of a `BinaryHeap`.
///
/// This `struct` is created by [`BinaryHeap::iter()`]. See its
/// documentation for more.
///
/// [`iter`]: BinaryHeap::iter
#[must_use = "iterators are lazy and do nothing unless consumed"]
pub struct Iter<'a, T: 'a> {
    iter: slice::Iter<'a, T>,
}

impl<T> Default for Iter<'_, T> {
    /// Creates an empty `binary_heap::Iter`.
    ///
    /// ```
    /// # use std::collections::binary_heap;
    /// let iter: binary_heap::Iter<'_, u8> = Default::default();
    /// assert_eq!(iter.len(), 0);
    /// ```
    fn default() -> Self {
        Iter {
            iter: Default::default(),
        }
    }
}

impl<T: fmt::Debug> fmt::Debug for Iter<'_, T> {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.debug_tuple("Iter").field(&self.iter.as_slice()).finish()
    }
}

// FIXME(#26925) Remove in favor of `#[derive(Clone)]`
impl<T> Clone for Iter<'_, T> {
    fn clone(&self) -> Self {
        Iter {
            iter: self.iter.clone(),
        }
    }
}

impl<'a, T> Iterator for Iter<'a, T> {
    type Item = &'a T;

    #[inline]
    fn next(&mut self) -> Option<&'a T> {
        self.iter.next()
    }

    #[inline]
    fn size_hint(&self) -> (usize, Option<usize>) {
        self.iter.size_hint()
    }

    #[inline]
    fn last(self) -> Option<&'a T> {
        self.iter.last()
    }
}

impl<'a, T> DoubleEndedIterator for Iter<'a, T> {
    #[inline]
    fn next_back(&mut self) -> Option<&'a T> {
        self.iter.next_back()
    }
}

impl<T> FusedIterator for Iter<'_, T> {}

/// An owning iterator over the elements of a `BinaryHeap`.
///
/// This `struct` is created by [`BinaryHeap::into_iter()`]
/// (provided by the [`IntoIterator`] trait). See its documentation for more.
///
/// [`into_iter`]: BinaryHeap::into_iter
#[derive(Clone)]
pub struct IntoIter<T> {
    iter: vec::IntoIter<T>,
}

impl<T: fmt::Debug> fmt::Debug for IntoIter<T> {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.debug_tuple("IntoIter")
            .field(&self.iter.as_slice())
            .finish()
    }
}

impl<T> Iterator for IntoIter<T> {
    type Item = T;

    #[inline]
    fn next(&mut self) -> Option<T> {
        self.iter.next()
    }

    #[inline]
    fn size_hint(&self) -> (usize, Option<usize>) {
        self.iter.size_hint()
    }
}

impl<T> DoubleEndedIterator for IntoIter<T> {
    #[inline]
    fn next_back(&mut self) -> Option<T> {
        self.iter.next_back()
    }
}

impl<T> FusedIterator for IntoIter<T> {}

impl<T> Default for IntoIter<T> {
    /// Creates an empty `binary_heap::IntoIter`.
    ///
    /// ```
    /// # use std::collections::binary_heap;
    /// let iter: binary_heap::IntoIter<u8> = Default::default();
    /// assert_eq!(iter.len(), 0);
    /// ```
    fn default() -> Self {
        IntoIter {
            iter: Default::default(),
        }
    }
}

#[must_use = "iterators are lazy and do nothing unless consumed"]
#[derive(Clone, Debug)]
pub struct IntoIterSorted<T, C> {
    inner: BinaryHeap<T, C>,
}

impl<T, C> Iterator for IntoIterSorted<T, C>
where
    C: Fn(&T, &T) -> Ordering,
{
    type Item = T;

    #[inline]
    fn next(&mut self) -> Option<T> {
        self.inner.pop()
    }

    #[inline]
    fn size_hint(&self) -> (usize, Option<usize>) {
        let exact = self.inner.len();
        (exact, Some(exact))
    }
}

/// A draining iterator over the elements of a `BinaryHeap`.
///
/// This `struct` is created by [`BinaryHeap::drain()`]. See its
/// documentation for more.
///
/// [`drain`]: BinaryHeap::drain
#[derive(Debug)]
pub struct Drain<'a, T: 'a> {
    iter: vec::Drain<'a, T>,
}

impl<T> Iterator for Drain<'_, T> {
    type Item = T;

    #[inline]
    fn next(&mut self) -> Option<T> {
        self.iter.next()
    }

    #[inline]
    fn size_hint(&self) -> (usize, Option<usize>) {
        self.iter.size_hint()
    }
}

impl<T> DoubleEndedIterator for Drain<'_, T> {
    #[inline]
    fn next_back(&mut self) -> Option<T> {
        self.iter.next_back()
    }
}

impl<T> ExactSizeIterator for Drain<'_, T> {}

impl<T> FusedIterator for Drain<'_, T> {}

/// A draining iterator over the elements of a `BinaryHeap`.
///
/// This `struct` is created by [`BinaryHeap::drain_sorted()`]. See its
/// documentation for more.
///
/// [`drain_sorted`]: BinaryHeap::drain_sorted
#[derive(Debug)]
pub struct DrainSorted<'a, T, C>
where
    C: Fn(&T, &T) -> Ordering,
{
    inner: &'a mut BinaryHeap<T, C>,
}

impl<'a, T, C> Drop for DrainSorted<'a, T, C>
where
    C: Fn(&T, &T) -> Ordering,
{
    /// Removes heap elements in heap order.
    fn drop(&mut self) {
        struct DropGuard<'r, 'a, T, C>(&'r mut DrainSorted<'a, T, C>)
        where
            C: Fn(&T, &T) -> Ordering;

        impl<'r, 'a, T, C> Drop for DropGuard<'r, 'a, T, C>
        where
            C: Fn(&T, &T) -> Ordering,
        {
            fn drop(&mut self) {
                while self.0.inner.pop().is_some() {}
            }
        }

        while let Some(item) = self.inner.pop() {
            let guard = DropGuard(self);
            drop(item);
            mem::forget(guard);
        }
    }
}

impl<T, C> Iterator for DrainSorted<'_, T, C>
where
    C: Fn(&T, &T) -> Ordering,
{
    type Item = T;

    #[inline]
    fn next(&mut self) -> Option<T> {
        self.inner.pop()
    }

    #[inline]
    fn size_hint(&self) -> (usize, Option<usize>) {
        let exact = self.inner.len();
        (exact, Some(exact))
    }
}

impl<T, C> ExactSizeIterator for DrainSorted<'_, T, C> where C: Fn(&T, &T) -> Ordering {}

impl<T, C> FusedIterator for DrainSorted<'_, T, C> where C: Fn(&T, &T) -> Ordering {}

impl<T, C> From<BinaryHeap<T, C>> for Vec<T> {
    /// Converts a `BinaryHeap<T>` into a `Vec<T>`.
    ///
    /// This conversion requires no data movement or allocation, and has
    /// constant time complexity.
    fn from(heap: BinaryHeap<T, C>) -> Vec<T> {
        heap.data
    }
}

impl<T, C> IntoIterator for BinaryHeap<T, C> {
    type Item = T;
    type IntoIter = IntoIter<T>;

    /// Creates a consuming iterator, that is, one that moves each value out of
    /// the binary heap in arbitrary order. The binary heap cannot be used
    /// after calling this.
    fn into_iter(self) -> IntoIter<T> {
        IntoIter {
            iter: self.data.into_iter(),
        }
    }
}

impl<'a, T, C> IntoIterator for &'a BinaryHeap<T, C> {
    type Item = &'a T;
    type IntoIter = Iter<'a, T>;

    fn into_iter(self) -> Iter<'a, T> {
        self.iter()
    }
}

impl<T, C> Extend<T> for BinaryHeap<T, C>
where
    C: Fn(&T, &T) -> Ordering,
{
    #[inline]
    fn extend<I: IntoIterator<Item = T>>(&mut self, iter: I) {
        let guard = RebuildOnDrop {
            rebuild_from: self.len(),
            heap: self,
        };
        guard.heap.data.extend(iter);
    }
}

impl<'a, T, C> Extend<&'a T> for BinaryHeap<T, C>
where
    T: Clone,
    C: Fn(&T, &T) -> Ordering,
{
    fn extend<I: IntoIterator<Item = &'a T>>(&mut self, iter: I) {
        self.extend(iter.into_iter().cloned());
    }
}

#[cfg(test)]
mod tests {
    use super::BinaryHeap;

    #[test]
    fn peek_all_returns_all_max_elements() {
        let heap = BinaryHeap::from_vec(vec![3, 5, 5, 2, 5, 4], |a, b| a.cmp(b));

        let mut values = Vec::new();
        heap.peek_all(|_, x| values.push(*x), &mut Vec::new());

        assert_eq!(values.len(), 3);
        assert!(values.into_iter().all(|x| x == 5));
    }

    #[test]
    fn peek_all_on_empty_heap_invokes_nothing() {
        let heap = BinaryHeap::<i32, _>::new(|a, b| a.cmp(b));
        let mut called = false;

        heap.peek_all(|__, _| called = true, &mut Vec::new());

        assert!(!called);
    }

    #[test]
    fn peek_all_on_single_element_heap_returns_that_element() {
        let heap = BinaryHeap::<i32, _>::from_vec(vec![42], |a, b| a.cmp(b));
        let mut values = Vec::new();

        heap.peek_all(|_, x| values.push(*x), &mut Vec::new());

        assert_eq!(values, vec![42]);
    }

    #[test]
    fn peek_all_returns_every_element_when_all_are_equal() {
        let heap = BinaryHeap::<i32, _>::from_vec(vec![7, 7, 7, 7, 7, 7, 7, 7], |a, b| a.cmp(b));
        let mut values = Vec::new();

        heap.peek_all(|_, x| values.push(*x), &mut Vec::new());

        assert_eq!(values.len(), heap.len());
        assert!(values.into_iter().all(|x| x == 7));
    }

    #[test]
    fn peek_all_returns_only_one_when_maximum_is_unique() {
        let heap = BinaryHeap::<i32, _>::from_vec(vec![9, 8, 7, 6, 5, 4, 3, 2, 1], |a, b| a.cmp(b));
        let mut values = Vec::new();

        heap.peek_all(|_, x| values.push(*x), &mut Vec::new());

        assert_eq!(values, vec![9]);
    }
}