Repository navigation
Expand file tree
/
Copy pathbyte_view.rs
More file actions
1343 lines (1211 loc) · 52.2 KB
/
Copy pathbyte_view.rs
File metadata and controls
1343 lines (1211 loc) · 52.2 KB
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
// SPDX-License-Identifier: Apache-2.0
// SPDX-FileCopyrightText: See the NOTICE file distributed with this work for additional information regarding copyright ownership
// SPDX-FileContributor: Licensed to the Apache Software Foundation (ASF) under one or more contributor license agreements
use crate::coalesce::{InProgressArray, append_gathered_nulls, check_filter_len};
use crate::filter::{FilterPredicate, FilterSelection};
use arrow_array::cast::AsArray;
use arrow_array::types::ByteViewType;
use arrow_array::{Array, ArrayRef, GenericByteViewArray, UInt32Array};
use arrow_buffer::{Buffer, NullBuffer, NullBufferBuilder};
use arrow_data::{ByteView, MAX_INLINE_VIEW_LEN};
use arrow_schema::ArrowError;
use std::marker::PhantomData;
use std::sync::{Arc, Weak};
/// InProgressArray for [`StringViewArray`] and [`BinaryViewArray`]
///
/// This structure buffers the views and data buffers as they are copied from
/// the source array, and then produces a new array when `finish` is called. It
/// also handles "garbage collection" by copying strings to a new buffer when
/// the source buffer is sparse (i.e. uses at least 2x more than the memory it
/// needs).
///
/// [`StringViewArray`]: arrow_array::StringViewArray
/// [`BinaryViewArray`]: arrow_array::BinaryViewArray
pub(crate) struct InProgressByteViewArray<B: ByteViewType> {
/// The source array and information
source: Option<Source>,
/// the target batch size (and thus size for views allocation)
batch_size: usize,
/// The in progress views (guaranteed to be valid)
views: Vec<u128>,
/// In progress nulls
nulls: NullBufferBuilder,
/// current buffer
current: Option<Vec<u8>>,
/// completed buffers (corresponding to content of views)
completed: Vec<Buffer>,
/// Allocates new buffers of increasing size as needed
buffer_source: BufferSource,
/// Phantom so we can use the same struct for both StringViewArray and
/// BinaryViewArray
_phantom: PhantomData<B>,
/// The size in bytes the [`Buffer`]s in [`Self::completed`] is taking
completed_buffers_size: usize,
/// The size in bytes from [`Self::source`] that it is being used in [`Self::completed`]
size_of_completed_buffers_from_current_source: usize,
/// Weak identity of the last indexed source. Repeated indexed copies from
/// this source reuse the data-buffer references already in `completed`.
/// Keeping only a weak reference avoids retaining its views and null bitmap.
last_indexed_source: Option<Weak<dyn Array>>,
/// Position in `completed` where `last_indexed_source`'s buffers begin.
last_indexed_starting_buffer: u32,
}
struct Source {
/// The array to copy form
array: ArrayRef,
/// Should the strings from the source array be copied into new buffers?
need_gc: bool,
/// How many bytes were actually used in the source array's buffers?
ideal_buffer_size: usize,
}
// manually implement Debug because ByteViewType doesn't implement Debug
impl<B: ByteViewType> std::fmt::Debug for InProgressByteViewArray<B> {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_struct("InProgressByteViewArray")
.field("batch_size", &self.batch_size)
.field("views", &self.views.len())
.field("nulls", &self.nulls)
.field("current", &self.current.as_ref().map(|_| "Some(...)"))
.field("completed", &self.completed.len())
.finish()
}
}
impl<B: ByteViewType> InProgressByteViewArray<B> {
pub(crate) fn new(batch_size: usize) -> Self {
let buffer_source = BufferSource::new();
Self {
batch_size,
source: None,
views: Vec::new(), // allocate in push
nulls: NullBufferBuilder::new(batch_size), // no allocation
current: None,
completed: vec![],
completed_buffers_size: 0,
size_of_completed_buffers_from_current_source: 0,
last_indexed_source: None,
last_indexed_starting_buffer: 0,
buffer_source,
_phantom: PhantomData,
}
}
/// Allocate space for output views and nulls if needed
///
/// This is done on write (when we know it is necessary) rather than
/// eagerly to avoid allocations that are not used.
fn ensure_capacity(&mut self) {
if self.views.capacity() == 0 {
self.views.reserve(self.batch_size);
}
}
/// Finishes in progress buffer, if any
fn finish_current(&mut self) {
let Some(next_buffer) = self.current.take() else {
return;
};
let buffer: Buffer = next_buffer.into();
self.completed_buffers_size += buffer.capacity();
self.completed.push(buffer);
}
/// # Safety
///
/// `filter.filter_len()` must not exceed `views.len()`.
unsafe fn append_views_by_filter(&mut self, views: &[u128], filter: &FilterPredicate) {
let selected_count = filter.count();
let current_len = self.views.len();
self.views.reserve(selected_count);
let mut written = 0;
// SAFETY: the caller guarantees that selected rows are in bounds,
// `FilterPredicate` yields `selected_count` rows, and capacity was
// reserved above.
unsafe {
let mut out = self.views.spare_capacity_mut().as_mut_ptr().cast::<u128>();
match filter.selection() {
FilterSelection::None => {}
FilterSelection::All { .. } => {
std::ptr::copy_nonoverlapping(views.as_ptr(), out, selected_count);
written = selected_count;
}
FilterSelection::Slices(slices) => {
slices.for_each(|(start, end)| {
let len = end - start;
std::ptr::copy_nonoverlapping(views.as_ptr().add(start), out, len);
out = out.add(len);
written += len;
});
}
FilterSelection::Indices(indices) => {
indices.for_each(|idx| {
out.write(*views.get_unchecked(idx));
out = out.add(1);
written += 1;
});
}
}
self.views.set_len(current_len + written);
}
debug_assert_eq!(written, selected_count);
}
fn append_nulls_by_filter(
&mut self,
filter: &FilterPredicate,
source_nulls: Option<&NullBuffer>,
) {
if let Some(nulls) = filter.filter_nulls(source_nulls) {
self.nulls.append_buffer(&nulls);
} else {
self.nulls.append_n_non_nulls(filter.count());
}
}
/// Append views to self.views, updating the buffer index if necessary
///
/// # Safety
///
/// [`B::validate`](ByteViewType::validate) must succeed for `views` and
/// `buffers`.
///
/// Note: this function performs no unsafe operations itself, but it assumes
/// `views` are valid views for self.views and `buffers` for self.buffers
#[inline(never)]
unsafe fn append_views_and_update_buffer_index(
&mut self,
views: &[u128],
buffers: &[Buffer],
is_reused: bool,
) {
if let Some(buffer) = self.current.take() {
let buffer: Buffer = buffer.into();
self.completed_buffers_size += buffer.capacity();
self.completed.push(buffer);
}
let buffers_size = buffers.iter().map(|b| b.capacity()).sum::<usize>();
if !is_reused {
self.completed_buffers_size += buffers_size;
} else if self.size_of_completed_buffers_from_current_source == 0 {
// Don't double count buffers size if already counted that
self.size_of_completed_buffers_from_current_source += buffers_size;
}
let starting_buffer = self.completed.len() as u32;
self.completed.extend_from_slice(buffers);
assert!(
u32::try_from(self.completed.len()).is_ok(),
"too many buffers"
);
if starting_buffer == 0 {
// If there are no buffers, we can just use the views as is
self.views.extend_from_slice(views);
} else {
// If there are buffers, we need to update the buffer index
let updated_views = views.iter().map(|v| {
// A view of a string of 12 bytes or fewer stores the string
// inline and has no buffer index, so adding 0 leaves it
// unchanged.
let offset = if *v as u32 > MAX_INLINE_VIEW_LEN {
starting_buffer
} else {
0
};
// The buffer index is bits 64 to 95 of a view. Adding
// `starting_buffer` gives the string's buffer index in the
// output, which fits in 32 bits, so the sum never carries into
// bit 96.
v + ((offset as u128) << 64)
});
self.views.extend(updated_views);
}
}
/// Append views to self.views, copying data from the buffers into
/// self.buffers and updating the buffer index as necessary.
///
/// # Arguments
/// - `views` - the views to append
/// - `view_buffer_size` - the total number of bytes pointed to by all
/// views (used to allocate new buffers if needed)
/// - `buffers` - the buffers the views point to
///
/// # Safety
///
/// [`B::validate`](ByteViewType::validate) must succeed for `views` and
/// `buffers`.
#[inline(never)]
unsafe fn append_views_and_copy_strings(
&mut self,
views: &[u128],
view_buffer_size: usize,
buffers: &[Buffer],
) {
// Note: the calculations below are designed to avoid any reallocations
// of the current buffer, and to only allocate new buffers when
// necessary, which is critical for performance.
// If there is no current buffer, allocate a new one
let Some(current) = self.current.take() else {
let new_buffer = self.buffer_source.next_buffer(view_buffer_size);
// SAFETY: the caller guarantees the views and buffers are valid.
unsafe { self.append_views_and_copy_strings_inner(views, new_buffer, buffers) };
return;
};
// If there is a current buffer with enough space, append the views and
// copy the strings into the existing buffer.
let mut remaining_capacity = current.capacity() - current.len();
if view_buffer_size <= remaining_capacity {
// SAFETY: the caller guarantees the views and buffers are valid.
unsafe { self.append_views_and_copy_strings_inner(views, current, buffers) };
return;
}
// Here there is a current buffer, but it doesn't have enough space to
// hold all the strings. Copy as many views as we can into the current
// buffer and then allocate a new buffer for the remaining views
//
// TODO: should we copy the strings too at the same time?
let mut num_view_to_current = 0;
for view in views {
let b = ByteView::from(*view);
let str_len = b.length;
if remaining_capacity < str_len as usize {
break;
}
if str_len > MAX_INLINE_VIEW_LEN {
remaining_capacity -= str_len as usize;
}
num_view_to_current += 1;
}
let first_views = &views[0..num_view_to_current];
let string_bytes_to_copy = current.capacity() - current.len() - remaining_capacity;
let remaining_view_buffer_size = view_buffer_size - string_bytes_to_copy;
// SAFETY: `first_views` is a subset of the caller-validated views.
unsafe { self.append_views_and_copy_strings_inner(first_views, current, buffers) };
let completed: Buffer = self.current.take().expect("completed").into();
self.completed_buffers_size += completed.capacity();
self.completed.push(completed);
// Copy any remaining views into a new buffer
let remaining_views = &views[num_view_to_current..];
let new_buffer = self.buffer_source.next_buffer(remaining_view_buffer_size);
// SAFETY: `remaining_views` is a subset of the caller-validated views.
unsafe { self.append_views_and_copy_strings_inner(remaining_views, new_buffer, buffers) };
}
/// Append views to self.views, copying data from the buffers into
/// dst_buffer, which is then set as self.current
///
/// # Panics:
/// If `self.current` is `Some`
///
/// # Safety
///
/// [`B::validate`](ByteViewType::validate) must succeed for `views` and
/// `buffers`.
///
/// See `append_views_and_copy_strings` for more details
#[inline(never)]
unsafe fn append_views_and_copy_strings_inner(
&mut self,
views: &[u128],
mut dst_buffer: Vec<u8>,
buffers: &[Buffer],
) {
assert!(self.current.is_none(), "current buffer should be None");
if views.is_empty() {
self.current = Some(dst_buffer);
return;
}
let new_buffer_index: u32 = self.completed.len().try_into().expect("too many buffers");
// In debug builds, check that the vector has enough capacity to copy
// the views into it without reallocating.
#[cfg(debug_assertions)]
{
let total_length: usize = views
.iter()
.filter_map(|v| {
let b = ByteView::from(*v);
if b.length > MAX_INLINE_VIEW_LEN {
Some(b.length as usize)
} else {
None
}
})
.sum();
debug_assert!(
dst_buffer.capacity() >= total_length,
"dst_buffer capacity {} is less than total length {}",
dst_buffer.capacity(),
total_length
);
}
// Copy the views, updating the buffer index and copying the data as needed
let new_views = views.iter().map(|v| {
let mut b: ByteView = ByteView::from(*v);
if b.length > MAX_INLINE_VIEW_LEN {
let buffer_index = b.buffer_index as usize;
let buffer_offset = b.offset as usize;
let str_len = b.length as usize;
// Update view to location in current
b.offset = dst_buffer.len() as u32;
b.buffer_index = new_buffer_index;
// safety: input views are validly constructed
let src = unsafe {
buffers
.get_unchecked(buffer_index)
.get_unchecked(buffer_offset..buffer_offset + str_len)
};
dst_buffer.extend_from_slice(src);
}
b.as_u128()
});
self.views.extend(new_views);
self.current = Some(dst_buffer);
}
/// # Safety
///
/// Every value in `indices` must be less than `src_views.len()`, and every
/// selected non-inline view must reference a valid range in `src_buffers`.
/// `selected_bytes` must equal the total length of the selected non-inline
/// values, and every selected value must fit within `i32::MAX` bytes.
#[inline(never)]
unsafe fn append_indexed_views_and_copy_strings(
&mut self,
src_views: &[u128],
indices: &[u32],
src_buffers: &[Buffer],
selected_bytes: usize,
) {
// SAFETY: the caller guarantees the same index and ByteView buffer
// invariants required by the delegated operation.
unsafe {
self.append_indexed_views_and_copy_strings_with_max_buffer_size(
src_views,
indices,
src_buffers,
selected_bytes,
i32::MAX as usize,
);
}
}
/// # Safety
///
/// Every value in `indices` must be less than `src_views.len()`, and every
/// selected non-inline view must reference a valid range in `src_buffers`.
/// `selected_bytes` must equal the total length of the selected non-inline
/// values, and every selected value must fit within `max_buffer_size`.
unsafe fn append_indexed_views_and_copy_strings_with_max_buffer_size(
&mut self,
src_views: &[u128],
indices: &[u32],
src_buffers: &[Buffer],
selected_bytes: usize,
max_buffer_size: usize,
) {
debug_assert!(
indices.iter().all(|&idx| (idx as usize) < src_views.len()),
"append_indexed_views_and_copy_strings_with_max_buffer_size requires caller-validated in-bounds indices",
);
if selected_bytes == 0 {
self.views.reserve(indices.len());
for &idx in indices {
// SAFETY: caller validated idx < src_views.len().
let raw = unsafe { *src_views.get_unchecked(idx as usize) };
self.views.push(raw);
}
return;
}
let mut dst_buffer = match self.current.take() {
Some(current)
if current.capacity() - current.len() >= selected_bytes
&& current
.len()
.checked_add(selected_bytes)
.is_some_and(|len| len <= max_buffer_size) =>
{
current
}
Some(current) => {
let completed_buffer: Buffer = current.into();
self.completed_buffers_size += completed_buffer.capacity();
self.completed.push(completed_buffer);
self.buffer_source
.next_buffer(selected_bytes.min(max_buffer_size))
}
None => self
.buffer_source
.next_buffer(selected_bytes.min(max_buffer_size)),
};
if selected_bytes > max_buffer_size {
// SAFETY: the caller guarantees the index, view, byte-count, and
// per-value size invariants required by the delegated operation.
unsafe {
self.append_indexed_views_and_copy_strings_split(
src_views,
indices,
src_buffers,
selected_bytes,
max_buffer_size,
dst_buffer,
)
};
return;
}
let current_len = self.views.len();
self.views.reserve(indices.len());
let output = self.views.spare_capacity_mut().as_mut_ptr().cast::<u128>();
let buffer_index = self.completed.len().try_into().expect("too many buffers");
for (i, &idx) in indices.iter().enumerate() {
// SAFETY: caller validated idx < src_views.len().
let raw = unsafe { *src_views.get_unchecked(idx as usize) };
let mut view = ByteView::from(raw);
if view.length > MAX_INLINE_VIEW_LEN {
let source_buffer_index = view.buffer_index as usize;
let source_buffer_offset = view.offset as usize;
let length = view.length as usize;
// SAFETY: source views carry valid buffer_index/offset/length.
let src = unsafe {
src_buffers
.get_unchecked(source_buffer_index)
.get_unchecked(source_buffer_offset..source_buffer_offset + length)
};
view.offset = dst_buffer
.len()
.try_into()
.expect("byte view offset exceeds u32::MAX");
view.buffer_index = buffer_index;
debug_assert!(dst_buffer.capacity() - dst_buffer.len() >= length);
dst_buffer.extend_from_slice(src);
}
// SAFETY: `reserve` above allocated room for every selected view.
unsafe { output.add(i).write(view.as_u128()) };
}
// SAFETY: the loop above initialized exactly `indices.len()` views.
unsafe { self.views.set_len(current_len + indices.len()) };
self.current = Some(dst_buffer);
}
/// Copies indexed views into bounded output buffers.
///
/// # Safety
///
/// Every value in `indices` must be less than `src_views.len()`, every
/// selected non-inline view must reference a valid range in `src_buffers`,
/// every selected value must fit within `max_buffer_size`, and
/// `selected_bytes` must equal their total length.
#[cold]
#[inline(never)]
unsafe fn append_indexed_views_and_copy_strings_split(
&mut self,
src_views: &[u128],
indices: &[u32],
src_buffers: &[Buffer],
selected_bytes: usize,
max_buffer_size: usize,
mut dst_buffer: Vec<u8>,
) {
let current_len = self.views.len();
self.views.reserve(indices.len());
let output = self.views.spare_capacity_mut().as_mut_ptr().cast::<u128>();
let mut copied_bytes = 0usize;
for (i, &idx) in indices.iter().enumerate() {
// SAFETY: caller validated idx < src_views.len().
let raw = unsafe { *src_views.get_unchecked(idx as usize) };
let mut view = ByteView::from(raw);
if view.length > MAX_INLINE_VIEW_LEN {
let buffer_index = view.buffer_index as usize;
let buffer_offset = view.offset as usize;
let str_len = view.length as usize;
assert!(str_len <= max_buffer_size);
let dst_end = dst_buffer
.len()
.checked_add(str_len)
.expect("byte view buffer size overflow");
if dst_end > max_buffer_size {
let completed_buffer: Buffer = dst_buffer.into();
self.completed_buffers_size += completed_buffer.capacity();
self.completed.push(completed_buffer);
let remaining = selected_bytes
.checked_sub(copied_bytes)
.expect("selected byte count mismatch");
dst_buffer = self
.buffer_source
.next_buffer(remaining.min(max_buffer_size));
}
// SAFETY: source views carry valid buffer_index/offset/length.
let src = unsafe {
src_buffers
.get_unchecked(buffer_index)
.get_unchecked(buffer_offset..buffer_offset + str_len)
};
view.offset = dst_buffer
.len()
.try_into()
.expect("byte view offset exceeds u32::MAX");
view.buffer_index = self.completed.len().try_into().expect("too many buffers");
debug_assert!(dst_buffer.capacity() - dst_buffer.len() >= str_len);
dst_buffer.extend_from_slice(src);
copied_bytes = copied_bytes
.checked_add(str_len)
.expect("selected byte count overflow");
}
// SAFETY: `reserve` above allocated room for every selected view.
unsafe { output.add(i).write(view.as_u128()) };
}
// SAFETY: the loop above initialized exactly `indices.len()` views.
unsafe { self.views.set_len(current_len + indices.len()) };
self.current = Some(dst_buffer);
}
}
/// Returns the total size of the selected non-inline values, or `None` if a
/// value exceeds `max_buffer_size` or the total overflows `usize`.
///
/// # Safety
///
/// Every value in `indices` must be less than `src_views.len()`.
unsafe fn selected_non_inline_bytes(
src_views: &[u128],
indices: &[u32],
max_buffer_size: usize,
) -> Option<usize> {
debug_assert!(
indices.iter().all(|&idx| (idx as usize) < src_views.len()),
"selected_non_inline_bytes requires caller-validated in-bounds indices",
);
let mut total = 0usize;
for &idx in indices {
// SAFETY: caller validated idx < src_views.len().
let raw = unsafe { *src_views.get_unchecked(idx as usize) };
let length = raw as u32;
if length > MAX_INLINE_VIEW_LEN {
let length = length as usize;
if length > max_buffer_size {
return None;
}
total = total.checked_add(length)?;
}
}
Some(total)
}
impl<B: ByteViewType> InProgressArray for InProgressByteViewArray<B> {
fn set_source(&mut self, source: Option<ArrayRef>) {
// If used values from source, add only the size that was used
self.completed_buffers_size += self.size_of_completed_buffers_from_current_source;
self.size_of_completed_buffers_from_current_source = 0;
self.last_indexed_source = None;
self.source = source.map(|array| {
let s = array.as_byte_view::<B>();
let (need_gc, ideal_buffer_size) = if s.data_buffers().is_empty() {
(false, 0)
} else {
let ideal_buffer_size = s.total_buffer_bytes_used();
// copying strings is expensive, so only do it if the array is
// sparse (uses at least 2x the memory it needs)
let need_gc = if ideal_buffer_size != 0 {
let mut is_sparse = false;
let mut curr_buffer_size = 0;
for buffer in s.data_buffers().iter() {
curr_buffer_size += buffer.capacity();
if curr_buffer_size > ideal_buffer_size * 2 {
is_sparse = true;
break;
}
}
is_sparse
} else {
false
};
(need_gc, ideal_buffer_size)
};
Source {
array,
need_gc,
ideal_buffer_size,
}
});
}
fn copy_rows(&mut self, offset: usize, len: usize) -> Result<(), ArrowError> {
self.ensure_capacity();
let source = self.source.take().ok_or_else(|| {
ArrowError::InvalidArgumentError(
"Internal Error: InProgressByteViewArray: source not set".to_string(),
)
})?;
// If creating StringViewArray output, ensure input was valid utf8 too
let s = source.array.as_byte_view::<B>();
// add any nulls, as necessary
if let Some(nulls) = s.nulls().as_ref() {
let nulls = nulls.slice(offset, len);
self.nulls.append_buffer(&nulls);
} else {
self.nulls.append_n_non_nulls(len);
}
let buffers = s.data_buffers();
let views = &s.views().as_ref()[offset..offset + len];
// If there are no data buffers in s (all inlined views), can append the
// views/nulls and done
if source.ideal_buffer_size == 0 {
self.views.extend_from_slice(views);
self.source = Some(source);
return Ok(());
}
// Copying the strings into a buffer can be time-consuming so
// only do it if the array is sparse
if source.need_gc {
// SAFETY: `views` and `buffers` come from the same valid byte-view array.
unsafe { self.append_views_and_copy_strings(views, source.ideal_buffer_size, buffers) };
} else {
// SAFETY: `views` and `buffers` come from the same valid byte-view array.
unsafe { self.append_views_and_update_buffer_index(views, buffers, true) };
}
self.source = Some(source);
Ok(())
}
fn copy_rows_by_filter(&mut self, filter: &FilterPredicate) -> Result<(), ArrowError> {
self.ensure_capacity();
let source = self.source.take().ok_or_else(|| {
ArrowError::InvalidArgumentError(
"Internal Error: InProgressByteViewArray: source not set".to_string(),
)
})?;
let s = source.array.as_byte_view::<B>();
if !s.data_buffers().is_empty() {
// Restore the source taken above before returning the guard error.
self.source = Some(source);
return Err(ArrowError::InvalidArgumentError(
"Internal Error: InProgressByteViewArray::copy_rows_by_filter requires inline views"
.to_string(),
));
}
if let Err(e) = check_filter_len(filter, s.len()) {
// Restore the source taken above before returning the error.
self.source = Some(source);
return Err(e);
}
self.append_nulls_by_filter(filter, s.nulls());
// SAFETY: `check_filter_len` verified the filter is no longer than
// `s.views()`.
unsafe { self.append_views_by_filter(s.views(), filter) };
self.source = Some(source);
Ok(())
}
fn copy_rows_by_filter_from(
&mut self,
source: ArrayRef,
filter: &FilterPredicate,
) -> Result<(), ArrowError> {
let s = source.as_byte_view::<B>();
// The source views reference no external buffers, so they must all be
// inline and we can copy just the nulls and views.
if s.data_buffers().is_empty() {
check_filter_len(filter, s.len())?;
self.ensure_capacity();
self.append_nulls_by_filter(filter, s.nulls());
// SAFETY: `check_filter_len` verified the filter is no longer than
// `s.views()`.
unsafe { self.append_views_by_filter(s.views(), filter) };
return Ok(());
}
// The views reference external buffers, so match the filter kernel:
// filter the views/nulls, but reuse the source's data buffers rather
// than copying the referenced string data.
let filtered = filter.filter(source.as_ref())?;
let filtered = filtered.as_byte_view::<B>();
self.ensure_capacity();
if let Some(nulls) = filtered.nulls().as_ref() {
self.nulls.append_buffer(nulls);
} else {
self.nulls.append_n_non_nulls(filter.count());
}
// SAFETY: the views and buffers come from the same valid filtered array.
unsafe {
self.append_views_and_update_buffer_index(
filtered.views(),
filtered.data_buffers(),
false,
)
};
Ok(())
}
/// # Safety
///
/// `indices` must not contain nulls, and every value must be less than
/// `source.len()`.
unsafe fn copy_rows_by_indices(
&mut self,
source: &ArrayRef,
indices: &UInt32Array,
) -> Result<(), ArrowError> {
if indices.is_empty() {
return Ok(());
}
let typed = source.as_byte_view::<B>();
self.ensure_capacity();
if let Some(src_nulls) = typed.nulls() {
// SAFETY: the caller validated every index against `typed.len()`, which
// equals the length of its null buffer.
unsafe { append_gathered_nulls(&mut self.nulls, src_nulls, indices) };
} else {
self.nulls.append_n_non_nulls(indices.len());
}
let index_values = indices.values();
let src_views = typed.views();
let src_buffers = typed.data_buffers();
debug_assert!(
index_values
.iter()
.all(|&idx| (idx as usize) < src_views.len()),
"copy_rows_by_indices requires caller-validated in-bounds indices",
);
if src_buffers.is_empty() {
let current_len = self.views.len();
self.views.reserve(indices.len());
// SAFETY: caller validated indices < typed.len() == src_views.len().
unsafe {
let base = self.views.spare_capacity_mut().as_mut_ptr().cast::<u128>();
let src = src_views.as_ptr();
for (i, &idx) in index_values.iter().enumerate() {
base.add(i).write(*src.add(idx as usize));
}
self.views.set_len(current_len + indices.len());
}
return Ok(());
}
let same_source = self
.last_indexed_source
.as_ref()
.is_some_and(|cached| std::ptr::addr_eq(cached.as_ptr(), Arc::as_ptr(source)));
if !same_source
&& let Some(selected_bytes) = unsafe {
// SAFETY: the caller validated every index against `typed.len()`,
// which equals `src_views.len()`.
selected_non_inline_bytes(src_views, index_values, i32::MAX as usize)
}
{
let source_capacity = src_buffers
.iter()
.try_fold(0usize, |total, buffer| total.checked_add(buffer.capacity()));
if let (Some(source_capacity), Some(compaction_threshold)) =
(source_capacity, selected_bytes.checked_mul(2))
&& source_capacity > compaction_threshold
{
// SAFETY: the caller validated every index against `typed.len()`,
// which equals `src_views.len()`, and both the views and buffers
// come from the same valid ByteView array.
unsafe {
self.append_indexed_views_and_copy_strings(
src_views,
index_values,
src_buffers,
selected_bytes,
);
}
return Ok(());
}
}
let starting_buffer: u32 = if same_source {
self.last_indexed_starting_buffer
} else {
if let Some(buffer) = self.current.take() {
let buffer: Buffer = buffer.into();
self.completed_buffers_size += buffer.capacity();
self.completed.push(buffer);
}
let start: u32 = self.completed.len().try_into().expect("too many buffers");
let buffers_size = src_buffers.iter().map(|b| b.capacity()).sum::<usize>();
self.completed_buffers_size += buffers_size;
self.completed.extend_from_slice(src_buffers);
self.last_indexed_source = Some(Arc::downgrade(source));
self.last_indexed_starting_buffer = start;
start
};
let current_len = self.views.len();
self.views.reserve(indices.len());
debug_assert!(
index_values
.iter()
.all(|&idx| (idx as usize) < src_views.len()),
"copy_rows_by_indices requires caller-validated in-bounds indices",
);
// SAFETY: caller validated indices < typed.len() == src_views.len().
unsafe {
let base = self.views.spare_capacity_mut().as_mut_ptr().cast::<u128>();
let src = src_views.as_ptr();
if starting_buffer == 0 {
for (i, &idx) in index_values.iter().enumerate() {
base.add(i).write(*src.add(idx as usize));
}
} else {
// Non-inline views (length > MAX_INLINE_VIEW_LEN) carry buffer_index
// in bits [64, 96); bump via raw u128 add to skip the ByteView round-trip.
let bump = (starting_buffer as u128) << 64;
for (i, &idx) in index_values.iter().enumerate() {
let raw = *src.add(idx as usize);
let len = raw as u32;
let adjusted = if len > MAX_INLINE_VIEW_LEN {
raw + bump
} else {
raw
};
base.add(i).write(adjusted);
}
}
self.views.set_len(current_len + indices.len());
}
Ok(())
}
fn finish(&mut self) -> Result<ArrayRef, ArrowError> {
self.finish_current();
assert!(self.current.is_none());
let buffers = std::mem::take(&mut self.completed);
let views = std::mem::take(&mut self.views);
let nulls = self.nulls.finish();
self.nulls = NullBufferBuilder::new(self.batch_size);
// Not reusing anything since we took all complete
self.size_of_completed_buffers_from_current_source = 0;
self.completed_buffers_size = 0;
self.last_indexed_source = None;
self.last_indexed_starting_buffer = 0;
// Safety: we created valid views and buffers above and the
// input arrays had value data and nulls
let new_array = unsafe {
GenericByteViewArray::<B>::new_unchecked(views.into(), buffers.into(), nulls)
};
Ok(Arc::new(new_array))
}
fn size(&self) -> usize {
self.completed_buffers_size
+ self.current.as_ref().map_or(0, |c| c.capacity())
+ self.nulls.allocated_size()
+ self.views.capacity() * size_of::<u128>()
+ self
.source
.as_ref()
.map_or(0, |s| s.array.get_array_memory_size())
}
}
const STARTING_BLOCK_SIZE: usize = 4 * 1024; // (note the first size used is actually 8KiB)
const MAX_BLOCK_SIZE: usize = 1024 * 1024; // 1MiB
/// Manages allocating new buffers for `StringViewArray` in increasing sizes
#[derive(Debug)]
struct BufferSource {
current_size: usize,
}
impl BufferSource {
fn new() -> Self {
Self {
current_size: STARTING_BLOCK_SIZE,
}
}
/// Return a new buffer, with a capacity of at least `min_size`
fn next_buffer(&mut self, min_size: usize) -> Vec<u8> {
let size = self.next_size(min_size);
Vec::with_capacity(size)
}
fn next_size(&mut self, min_size: usize) -> usize {
if self.current_size < MAX_BLOCK_SIZE {
// If the current size is less than the max size, we can double it
// we have fixed start/end block sizes, so we can't overflow
self.current_size = self.current_size.saturating_mul(2);
}
if self.current_size >= min_size {
self.current_size
} else {
// increase next size until we hit min_size or max size
while self.current_size <= min_size && self.current_size < MAX_BLOCK_SIZE {
self.current_size = self.current_size.saturating_mul(2);
}
self.current_size.max(min_size)
}
}
}
#[cfg(test)]
mod tests {
use super::*;
use crate::concat::concat;
use crate::filter::FilterBuilder;