pdf2 0.2.0-alpha.2

A PDF 2.0 (ISO 32000-2) reader, writer and text extractor.
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
//! Opening a PDF 2.0 file through its cross-reference data (§7.5).
//!
//! [`Document::open`] sniffs the header, follows `startxref`, and reads every
//! cross-reference section in the `/Prev` chain — classic tables, `/Type /XRef`
//! cross-reference streams, and the hybrid `/XRefStm` form — keeping the newest
//! entry for each object number. Objects are fetched lazily: the file bytes
//! stay put, and an object is parsed (and cached) the first time somebody asks
//! for it. Objects stored in `/ObjStm` streams are extracted on first use.

use alloc::borrow::ToOwned;
use alloc::string::String;
use alloc::vec;
use alloc::vec::Vec;

use alloc::collections::{BTreeMap, BTreeSet};
use core::cell::{Cell, RefCell};

use crate::crypto::Encryption;
use crate::error::{Error, Result};
use crate::filter::{self, FilterSpec};
use crate::lexer::Token;
use crate::object::{Dictionary, Object, ObjectId, Stream};
use crate::parser::{LengthResolver, Parser};

/// The header every file of this crate must start with.
pub const HEADER: &[u8] = b"%PDF-";

/// How far into the file the `%PDF-` header may sit (§7.5.2).
const HEADER_SEARCH_LIMIT: usize = 1024;

/// How many `/Prev` sections are followed before giving up on a loop.
const MAX_XREF_SECTIONS: usize = 128;

/// How deep an object may be resolved while parsing another one, which bounds
/// `/Length` and `/Filter` chains that point at each other.
const MAX_RESOLVE_DEPTH: usize = 32;

/// Where one object lives according to the cross-reference data.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
enum XrefEntry {
    /// A free entry (`f`, type 0): the object number is unused.
    Free { generation: u16 },
    /// An in-use entry (`n`, type 1): the object is at a byte offset.
    InUse { offset: usize, generation: u16 },
    /// A compressed entry (type 2): the object is inside an `/ObjStm` stream.
    Compressed { stream: u32, index: u32 },
}

/// A parsed PDF file.
///
/// Opening is cheap: only the header, the cross-reference sections and the
/// trailer are read. Pages, streams and content are parsed on demand by
/// [`Document::get_object`], which caches what it reads.
#[derive(Debug)]
pub struct Document {
    data: Vec<u8>,
    version: String,
    trailer: Dictionary,
    xref: BTreeMap<u32, XrefEntry>,
    cache: RefCell<BTreeMap<ObjectId, Object>>,
    /// Object numbers held by each object stream, in index order.
    object_streams: RefCell<BTreeMap<u32, Vec<u32>>>,
    /// The open security handler, when the document is encrypted.
    encryption: Option<Encryption>,
    /// The number of the `/Encrypt` dictionary, which is never decrypted.
    encrypt_number: Option<u32>,
    depth: Cell<usize>,
}

impl Document {
    /// Open a PDF from its bytes.
    ///
    /// Accepts `%PDF-2.x` files whose cross-reference is a classic table, an
    /// xref stream, or a hybrid of both, with any number of `/Prev` sections.
    ///
    /// An encrypted file is opened with the empty password, which covers the
    /// common "no password needed, but encrypted" case; use
    /// [`Document::open_with_password`] otherwise.
    pub fn open(data: impl Into<Vec<u8>>) -> Result<Self> {
        Self::open_with_password(data, b"")
    }

    /// Open a PDF with a password for its `/V 5` (AES-256) security handler.
    pub fn open_with_password(data: impl Into<Vec<u8>>, password: &[u8]) -> Result<Self> {
        let data = data.into();
        let version = header_version(&data)?;
        let (xref, trailer) = read_xref_chain(&data)?;
        match trailer.get(b"Root") {
            Some(Object::Reference(_)) => {}
            _ => {
                return Err(Error::Malformed {
                    what: "trailer /Root must be an indirect reference",
                    offset: None,
                })
            }
        }
        let mut document = Self {
            data,
            version,
            trailer,
            xref,
            cache: RefCell::new(BTreeMap::new()),
            object_streams: RefCell::new(BTreeMap::new()),
            encryption: None,
            encrypt_number: None,
            depth: Cell::new(0),
        };
        if let Some(encrypt) = document.trailer.get(b"Encrypt").cloned() {
            // The `/Encrypt` dictionary is plain even in an encrypted file, so
            // it is fetched before any handler is installed.
            let (dictionary, number) = match &encrypt {
                Object::Reference(id) => match document.get_object(*id)? {
                    Object::Dictionary(dict) => (dict, Some(id.number)),
                    _ => {
                        return Err(Error::Malformed {
                            what: "trailer /Encrypt is not a dictionary",
                            offset: None,
                        })
                    }
                },
                Object::Dictionary(dict) => (dict.clone(), None),
                _ => {
                    return Err(Error::Malformed {
                        what: "trailer /Encrypt must be a reference or a dictionary",
                        offset: None,
                    })
                }
            };
            document.encryption = Some(Encryption::new(&dictionary, password)?);
            document.encrypt_number = number;
        }
        Ok(document)
    }

    /// Open a PDF from a file path.
    #[cfg(feature = "std")]
    pub fn open_path(path: impl AsRef<std::path::Path>) -> Result<Self> {
        Self::open(std::fs::read(path)?)
    }

    /// Open a PDF from a file path with a password.
    #[cfg(feature = "std")]
    pub fn open_path_with_password(
        path: impl AsRef<std::path::Path>,
        password: &[u8],
    ) -> Result<Self> {
        Self::open_with_password(std::fs::read(path)?, password)
    }

    /// Whether the document was opened with a security handler.
    pub fn is_encrypted(&self) -> bool {
        self.encryption.is_some()
    }

    /// The version from the header, e.g. `"2.0"`.
    pub fn version(&self) -> &str {
        &self.version
    }

    /// The newest trailer dictionary.
    pub fn trailer(&self) -> &Dictionary {
        &self.trailer
    }

    /// The document catalog, reached through the trailer's `/Root`.
    ///
    /// The catalogue is the entry point of the whole object graph (§7.7.2).
    pub fn catalog(&self) -> Result<Object> {
        let root = match self.trailer.get(b"Root") {
            Some(Object::Reference(id)) => *id,
            _ => {
                return Err(Error::Malformed {
                    what: "trailer /Root must be an indirect reference",
                    offset: None,
                })
            }
        };
        self.get_object(root)
    }

    /// Fetch the object with this id, following the cross-reference.
    ///
    /// Only the object number selects the entry, so a stale generation still
    /// resolves — that is what real readers do and what incremental updates
    /// rely on. Use [`Document::get_object_by_number`] to fetch with the
    /// generation the cross-reference records.
    ///
    /// Results are cached, so fetching the same object twice parses it once.
    pub fn get_object(&self, id: ObjectId) -> Result<Object> {
        if let Some(cached) = self.cache.borrow().get(&id) {
            return Ok(cached.clone());
        }
        let entry = *self.xref.get(&id.number).ok_or(Error::Malformed {
            what: "object number is not in the cross-reference",
            offset: None,
        })?;
        let object = match entry {
            XrefEntry::Free { .. } => {
                return Err(Error::Malformed {
                    what: "object is free in the cross-reference",
                    offset: None,
                })
            }
            XrefEntry::InUse { offset, .. } => self.parse_at(id.number, offset)?,
            XrefEntry::Compressed { stream, index } => {
                self.compressed_object(id.number, stream, index)?
            }
        };
        let object = match &self.encryption {
            // The `/Encrypt` dictionary holds `/U`, `/O` and `/Perms` verbatim.
            Some(encryption) if Some(id.number) != self.encrypt_number => {
                encryption.decrypt_object(&object)?
            }
            _ => object,
        };
        self.cache.borrow_mut().insert(id, object.clone());
        Ok(object)
    }

    /// Fetch an object by number, using the generation the cross-reference has.
    pub fn get_object_by_number(&self, number: u32) -> Result<Object> {
        match self.xref.get(&number) {
            Some(XrefEntry::InUse { generation, .. }) => {
                self.get_object(ObjectId::new(number, *generation))
            }
            Some(XrefEntry::Compressed { .. }) => self.get_object(ObjectId::new(number, 0)),
            Some(XrefEntry::Free { .. }) => Err(Error::Malformed {
                what: "object is free in the cross-reference",
                offset: None,
            }),
            None => Err(Error::Malformed {
                what: "object number is not in the cross-reference",
                offset: None,
            }),
        }
    }

    /// Whether the cross-reference knows `number` as a live object.
    pub fn contains_object(&self, number: u32) -> bool {
        matches!(
            self.xref.get(&number),
            Some(XrefEntry::InUse { .. } | XrefEntry::Compressed { .. })
        )
    }

    /// Fetch a stream object by id.
    pub fn stream(&self, id: ObjectId) -> Result<Stream> {
        match self.get_object(id)? {
            Object::Stream(stream) => Ok(stream),
            _ => Err(Error::Malformed {
                what: "object is not a stream",
                offset: None,
            }),
        }
    }

    /// Decode a stream's payload through its filter chain.
    ///
    /// Indirect `/Filter` and `/DecodeParms` entries are resolved through the
    /// object table first, which is what the plain [`Stream::decode`] cannot
    /// do on its own.
    pub fn decode_stream(&self, stream: &Stream) -> Result<Vec<u8>> {
        let dict = stream.dict();
        let mut resolved = Dictionary::new();
        if let Some(filter) = self.resolve(dict.get(b"Filter"))? {
            resolved.insert(b"Filter", filter);
        }
        if let Some(parms) = self.resolve(dict.get(b"DecodeParms"))? {
            resolved.insert(b"DecodeParms", parms);
        }
        let chain = FilterSpec::from_dict(&resolved)?;
        // Decryption happens before decompression (§7.6.1).
        let data = match &self.encryption {
            Some(encryption) => encryption.decrypt_stream(stream)?,
            None => stream.data().to_vec(),
        };
        filter::decode_chain(&data, &chain)
    }

    /// Fetch a stream by id and decode its payload in one step.
    pub fn stream_data(&self, id: ObjectId) -> Result<Vec<u8>> {
        let stream = self.stream(id)?;
        self.decode_stream(&stream)
    }

    /// Replace indirect references with the objects they point at.
    pub(crate) fn resolve(&self, object: Option<&Object>) -> Result<Option<Object>> {
        match object {
            None => Ok(None),
            Some(Object::Reference(id)) => Ok(Some(self.get_object(*id)?)),
            Some(Object::Array(items)) => {
                let mut resolved = Vec::with_capacity(items.len());
                for item in items {
                    resolved.push(
                        self.resolve(Some(item))?
                            .ok_or(Error::malformed("array entry"))?,
                    );
                }
                Ok(Some(Object::Array(resolved)))
            }
            Some(other) => Ok(Some(other.clone())),
        }
    }

    fn parse_at(&self, number: u32, offset: usize) -> Result<Object> {
        let depth = self.depth.get() + 1;
        if depth > MAX_RESOLVE_DEPTH {
            return Err(Error::Malformed {
                what: "object resolution nested too deeply",
                offset: Some(offset),
            });
        }
        self.depth.set(depth);
        let result = self.parse_at_inner(number, offset);
        self.depth.set(depth - 1);
        result
    }

    fn parse_at_inner(&self, number: u32, offset: usize) -> Result<Object> {
        let slice = self.data.get(offset..).ok_or(Error::Malformed {
            what: "cross-reference offset is past the end of the file",
            offset: Some(offset),
        })?;
        let mut parser = Parser::with_resolver(slice, self);
        let indirect = parser.parse_indirect_object()?;
        if indirect.id.number != number {
            return Err(Error::Malformed {
                what: "cross-reference offset points at a different object",
                offset: Some(offset),
            });
        }
        Ok(indirect.object)
    }
}

impl Document {
    /// Fetch an object stored inside an object stream, extracting the whole
    /// stream once and caching every object it holds.
    fn compressed_object(&self, number: u32, stream: u32, index: u32) -> Result<Object> {
        self.load_object_stream(stream)?;
        {
            let object_streams = self.object_streams.borrow();
            let numbers = object_streams
                .get(&stream)
                .ok_or_else(|| Error::malformed("object stream was not indexed"))?;
            match numbers.get(usize::try_from(index).unwrap_or(usize::MAX)) {
                Some(found) if *found == number => {}
                Some(_) => {
                    return Err(Error::Malformed {
                        what: "object stream index points at a different object",
                        offset: None,
                    })
                }
                None => {
                    return Err(Error::Malformed {
                        what: "object stream index is out of range",
                        offset: None,
                    })
                }
            }
        }
        self.cache
            .borrow()
            .get(&ObjectId::new(number, 0))
            .cloned()
            .ok_or_else(|| Error::malformed("object missing from its object stream"))
    }

    /// Decode an `/ObjStm` stream and cache the objects it contains.
    fn load_object_stream(&self, stream_number: u32) -> Result<()> {
        if self.object_streams.borrow().contains_key(&stream_number) {
            return Ok(());
        }
        let stream = match self.get_object_by_number(stream_number)? {
            Object::Stream(stream) => stream,
            _ => {
                return Err(Error::Malformed {
                    what: "object stream entry points at an object that is not a stream",
                    offset: None,
                })
            }
        };
        let count = stream_integer(&stream, b"N")?;
        let first = stream_integer(&stream, b"First")?;
        let data = self.decode_stream(&stream)?;
        if first > data.len() {
            return Err(Error::Malformed {
                what: "object stream /First is past the end of its data",
                offset: None,
            });
        }
        // The header is `count` pairs of object number and relative offset.
        let mut parser = Parser::new(&data);
        let mut entries = Vec::with_capacity(count.min(1024));
        for _ in 0..count {
            let number = parser
                .parse_object()?
                .as_integer()
                .and_then(|value| u32::try_from(value).ok())
                .ok_or_else(|| Error::malformed("object stream header object number"))?;
            let offset = parser
                .parse_object()?
                .as_integer()
                .and_then(|value| usize::try_from(value).ok())
                .ok_or_else(|| Error::malformed("object stream header offset"))?;
            entries.push((number, offset));
        }
        let mut numbers = Vec::with_capacity(entries.len());
        let mut objects = Vec::with_capacity(entries.len());
        for (number, offset) in entries {
            let start = first.checked_add(offset).ok_or_else(|| {
                Error::malformed("object stream offset overflows the file length")
            })?;
            let slice = data.get(start..).ok_or(Error::Malformed {
                what: "object stream offset is past the end of its data",
                offset: Some(start),
            })?;
            let mut parser = Parser::new(slice);
            objects.push((number, parser.parse_object()?));
            numbers.push(number);
        }
        let mut cache = self.cache.borrow_mut();
        for (number, object) in objects {
            cache.insert(ObjectId::new(number, 0), object);
        }
        drop(cache);
        self.object_streams
            .borrow_mut()
            .insert(stream_number, numbers);
        Ok(())
    }
}

/// A required integer entry of a stream dictionary, as a `usize`.
fn stream_integer(stream: &Stream, key: &[u8]) -> Result<usize> {
    match stream.dict().get(key) {
        Some(Object::Integer(value)) => usize::try_from(*value).map_err(|_| Error::Malformed {
            what: "stream dictionary entry must not be negative",
            offset: None,
        }),
        _ => Err(Error::Malformed {
            what: "object stream needs integer /N and /First entries",
            offset: None,
        }),
    }
}

impl LengthResolver for Document {
    fn resolve_length(&self, id: ObjectId) -> Result<usize> {
        match self.get_object(id)? {
            Object::Integer(length) => usize::try_from(length).map_err(|_| Error::Malformed {
                what: "stream /Length must not be negative",
                offset: None,
            }),
            _ => Err(Error::Malformed {
                what: "stream /Length must be an integer",
                offset: None,
            }),
        }
    }
}

/// Read the version out of the `%PDF-` header.
fn header_version(data: &[u8]) -> Result<String> {
    let limit = data.len().min(HEADER_SEARCH_LIMIT);
    let start = data[..limit]
        .windows(HEADER.len())
        .position(|window| window == HEADER)
        .ok_or(Error::Malformed {
            what: "missing %PDF- header",
            offset: None,
        })?;
    let rest = &data[start + HEADER.len()..];
    let end = rest
        .iter()
        .position(|byte| *byte == b'\r' || *byte == b'\n')
        .unwrap_or(rest.len())
        .min(8);
    let version = core::str::from_utf8(&rest[..end]).map_err(|_| Error::Malformed {
        what: "header version is not ASCII",
        offset: Some(start),
    })?;
    let version = version.trim();
    if !version.starts_with("2.") {
        return Err(Error::Malformed {
            what: "only PDF 2.x is supported",
            offset: Some(start),
        });
    }
    Ok(version.to_owned())
}

/// Offset of the last occurrence of `needle`, if any.
fn find_last(data: &[u8], needle: &[u8]) -> Option<usize> {
    if needle.is_empty() || data.len() < needle.len() {
        return None;
    }
    data.windows(needle.len())
        .rposition(|window| window == needle)
}

/// Find the last `startxref` and the offset it points at.
fn startxref_offset(data: &[u8]) -> Result<usize> {
    let at = find_last(data, b"startxref").ok_or(Error::Malformed {
        what: "missing startxref",
        offset: None,
    })?;
    if find_last(data, b"%%EOF").is_none() {
        return Err(Error::Malformed {
            what: "missing %%EOF",
            offset: None,
        });
    }
    let rest = &data[at + b"startxref".len()..];
    let digits_start =
        rest.iter()
            .position(|byte| byte.is_ascii_digit())
            .ok_or(Error::Malformed {
                what: "startxref without an offset",
                offset: Some(at),
            })?;
    let digits_end = digits_start
        + rest[digits_start..]
            .iter()
            .position(|byte| !byte.is_ascii_digit())
            .unwrap_or(rest.len() - digits_start);
    let text =
        core::str::from_utf8(&rest[digits_start..digits_end]).map_err(|_| Error::Malformed {
            what: "startxref offset is not a number",
            offset: Some(at),
        })?;
    let offset: usize = text.parse().map_err(|_| Error::Malformed {
        what: "startxref offset is not a number",
        offset: Some(at),
    })?;
    if offset >= data.len() {
        return Err(Error::Malformed {
            what: "startxref offset is past the end of the file",
            offset: Some(at),
        });
    }
    Ok(offset)
}

/// Read the newest cross-reference section and every `/Prev` before it.
///
/// The newest entry for an object number wins: an incremental update re-points
/// objects and the older sections only fill in the numbers it does not mention.
fn read_xref_chain(data: &[u8]) -> Result<(BTreeMap<u32, XrefEntry>, Dictionary)> {
    let mut offset = startxref_offset(data)?;
    let mut merged = BTreeMap::new();
    let mut newest_trailer: Option<Dictionary> = None;
    let mut visited = BTreeSet::new();
    for _ in 0..MAX_XREF_SECTIONS {
        if !visited.insert(offset) {
            return Err(Error::Malformed {
                what: "cross-reference /Prev chain loops",
                offset: Some(offset),
            });
        }
        let (section, trailer) = parse_xref_section(data, offset)?;
        for (number, entry) in section {
            merged.entry(number).or_insert(entry);
        }
        // Hybrid files keep a classic table for old readers and hide the
        // compressed objects in a cross-reference stream named by `/XRefStm`.
        // The classic entries win; the stream only fills the gaps.
        if let Some(xref_stream) = trailer.get(b"XRefStm") {
            let stream_offset = xref_stream
                .as_integer()
                .and_then(|value| usize::try_from(value).ok())
                .ok_or(Error::Malformed {
                    what: "trailer /XRefStm must be a non-negative integer",
                    offset: Some(offset),
                })?;
            if !visited.insert(stream_offset) {
                return Err(Error::Malformed {
                    what: "cross-reference /XRefStm loops",
                    offset: Some(stream_offset),
                });
            }
            let (stream_entries, _) = parse_xref_stream(data, stream_offset)?;
            for (number, entry) in stream_entries {
                merged.entry(number).or_insert(entry);
            }
        }
        let previous = match trailer.get(b"Prev") {
            None => {
                if newest_trailer.is_none() {
                    newest_trailer = Some(trailer);
                }
                break;
            }
            Some(Object::Integer(previous)) => *previous,
            Some(_) => {
                return Err(Error::Malformed {
                    what: "trailer /Prev must be an integer",
                    offset: Some(offset),
                })
            }
        };
        let previous = usize::try_from(previous).map_err(|_| Error::Malformed {
            what: "trailer /Prev must not be negative",
            offset: Some(offset),
        })?;
        if newest_trailer.is_none() {
            newest_trailer = Some(trailer);
        }
        if previous >= data.len() {
            return Err(Error::Malformed {
                what: "trailer /Prev is past the end of the file",
                offset: Some(offset),
            });
        }
        offset = previous;
    }
    let trailer = newest_trailer.ok_or(Error::Malformed {
        what: "no cross-reference section could be read",
        offset: None,
    })?;
    Ok((merged, trailer))
}

/// One `xref` table and the trailer that follows it.
/// Read the cross-reference section at `offset`.
///
/// A section is either a classic `xref` table or a `/Type /XRef` stream; the
/// first token decides which, because both are reached the same way from
/// `startxref`, `/Prev` and `/XRefStm`.
fn parse_xref_section(
    data: &[u8],
    offset: usize,
) -> Result<(BTreeMap<u32, XrefEntry>, Dictionary)> {
    let slice = data.get(offset..).ok_or(Error::Malformed {
        what: "cross-reference offset is past the end of the file",
        offset: Some(offset),
    })?;
    let mut parser = Parser::new(slice);
    match parser.next_token()? {
        Token::Keyword(keyword) if keyword == b"xref" => parse_xref_table(data, offset, parser),
        Token::Integer(_) => parse_xref_stream(data, offset),
        _ => Err(Error::Malformed {
            what: "startxref does not point at an xref table or stream",
            offset: Some(offset),
        }),
    }
}

/// Read a classic `xref` table; `parser` has already consumed the keyword.
fn parse_xref_table(
    slice_source: &[u8],
    offset: usize,
    mut parser: Parser<'_>,
) -> Result<(BTreeMap<u32, XrefEntry>, Dictionary)> {
    let slice = slice_source;
    let mut entries = BTreeMap::new();
    loop {
        match parser.peek_token()? {
            Token::Keyword(keyword) if keyword == b"trailer" => {
                parser.next_token()?;
                break;
            }
            Token::Integer(_) => {
                let first = as_integer(parser.next_token()?)?;
                let count = as_integer(parser.next_token()?)?;
                let first = u32::try_from(first).map_err(|_| Error::Malformed {
                    what: "cross-reference subsection start",
                    offset: Some(offset),
                })?;
                // Each entry needs at least three short tokens; refuse counts
                // that cannot possibly fit in the remaining bytes.
                let count = usize::try_from(count)
                    .ok()
                    .filter(|count| *count <= slice.len())
                    .ok_or(Error::Malformed {
                        what: "cross-reference subsection count",
                        offset: Some(offset),
                    })?;
                for index in 0..count {
                    let entry_offset = as_integer(parser.next_token()?)?;
                    let generation = as_integer(parser.next_token()?)?;
                    let kind = match parser.next_token()? {
                        Token::Keyword(kind) => kind,
                        _ => {
                            return Err(Error::Malformed {
                                what: "cross-reference entry kind",
                                offset: Some(offset),
                            })
                        }
                    };
                    let generation = u16::try_from(generation).map_err(|_| Error::Malformed {
                        what: "cross-reference generation",
                        offset: Some(offset),
                    })?;
                    let index = u32::try_from(index).map_err(|_| Error::Malformed {
                        what: "cross-reference subsection count",
                        offset: Some(offset),
                    })?;
                    let number = first.checked_add(index).ok_or(Error::Malformed {
                        what: "cross-reference object number",
                        offset: Some(offset),
                    })?;
                    let entry = match kind.as_slice() {
                        b"n" => XrefEntry::InUse {
                            offset: usize::try_from(entry_offset).map_err(|_| {
                                Error::Malformed {
                                    what: "cross-reference offset must not be negative",
                                    offset: Some(offset),
                                }
                            })?,
                            generation,
                        },
                        b"f" => XrefEntry::Free { generation },
                        _ => {
                            return Err(Error::Malformed {
                                what: "cross-reference entry kind must be n or f",
                                offset: Some(offset),
                            })
                        }
                    };
                    entries.entry(number).or_insert(entry);
                }
            }
            Token::Eof => {
                return Err(Error::Malformed {
                    what: "cross-reference table without a trailer",
                    offset: Some(offset),
                })
            }
            _ => {
                return Err(Error::Malformed {
                    what: "cross-reference subsection header",
                    offset: Some(offset),
                })
            }
        }
    }
    let trailer = match parser.parse_object()? {
        Object::Dictionary(trailer) => trailer,
        _ => {
            return Err(Error::Malformed {
                what: "trailer is not a dictionary",
                offset: Some(offset),
            })
        }
    };
    Ok((entries, trailer))
}

/// Read a `/Type /XRef` cross-reference stream at `offset` (§7.5.8).
fn parse_xref_stream(data: &[u8], offset: usize) -> Result<(BTreeMap<u32, XrefEntry>, Dictionary)> {
    let slice = data.get(offset..).ok_or(Error::Malformed {
        what: "cross-reference stream offset is past the end of the file",
        offset: Some(offset),
    })?;
    let mut parser = Parser::new(slice);
    let stream = match parser.parse_indirect_object()?.object {
        Object::Stream(stream) => stream,
        _ => {
            return Err(Error::Malformed {
                what: "startxref points at an object that is not a cross-reference stream",
                offset: Some(offset),
            })
        }
    };
    let dict = stream.dict().clone();
    if dict.get(b"Type") != Some(&Object::name("XRef")) {
        return Err(Error::Malformed {
            what: "cross-reference stream without /Type /XRef",
            offset: Some(offset),
        });
    }
    let widths = xref_widths(&dict)?;
    let index = xref_index(&dict)?;
    let chain = FilterSpec::from_dict(&dict)?;
    let decoded = filter::decode_chain(stream.data(), &chain)?;
    let row_bytes: usize = widths.iter().map(|width| usize::from(*width)).sum();
    if row_bytes == 0 {
        return Err(Error::Malformed {
            what: "/W of a cross-reference stream must not be all zero",
            offset: Some(offset),
        });
    }
    let mut entries = BTreeMap::new();
    let mut position = 0;
    for (first, count) in index {
        if usize::try_from(count).map_or(true, |count| count > decoded.len()) {
            return Err(Error::Malformed {
                what: "cross-reference stream /Index count is implausible",
                offset: Some(offset),
            });
        }
        for index in 0..count {
            let row = decoded
                .get(position..position + row_bytes)
                .ok_or(Error::unexpected_eof("cross-reference stream entry"))?;
            position += row_bytes;
            let mut cursor = 0;
            let kind = read_xref_field(row, &mut cursor, widths[0], 1)?;
            let field_two = read_xref_field(row, &mut cursor, widths[1], 0)?;
            let field_three = read_xref_field(row, &mut cursor, widths[2], 0)?;
            let number = first.checked_add(index).ok_or(Error::Malformed {
                what: "cross-reference stream object number overflows",
                offset: Some(offset),
            })?;
            let entry = match kind {
                0 => XrefEntry::Free {
                    generation: u16::try_from(field_three).unwrap_or(u16::MAX),
                },
                1 => XrefEntry::InUse {
                    offset: usize::try_from(field_two).map_err(|_| Error::Malformed {
                        what: "cross-reference stream offset is out of range",
                        offset: Some(offset),
                    })?,
                    generation: u16::try_from(field_three).unwrap_or(u16::MAX),
                },
                2 => XrefEntry::Compressed {
                    stream: u32::try_from(field_two).map_err(|_| Error::Malformed {
                        what: "object stream number is out of range",
                        offset: Some(offset),
                    })?,
                    index: u32::try_from(field_three).map_err(|_| Error::Malformed {
                        what: "object stream index is out of range",
                        offset: Some(offset),
                    })?,
                },
                other => {
                    return Err(Error::Malformed {
                        what: match other {
                            3.. => "unknown cross-reference stream entry type",
                            _ => "cross-reference stream entry type",
                        },
                        offset: Some(offset),
                    })
                }
            };
            entries.entry(number).or_insert(entry);
        }
    }
    Ok((entries, dict))
}

/// The three field widths from `/W`.
fn xref_widths(dict: &Dictionary) -> Result<[u8; 3]> {
    let Some(Object::Array(items)) = dict.get(b"W") else {
        return Err(Error::Malformed {
            what: "cross-reference stream needs a /W array",
            offset: None,
        });
    };
    if items.len() != 3 {
        return Err(Error::Malformed {
            what: "/W must hold exactly three integers",
            offset: None,
        });
    }
    let mut widths = [0_u8; 3];
    for (slot, item) in widths.iter_mut().zip(items) {
        let width = item
            .as_integer()
            .and_then(|value| u8::try_from(value).ok())
            .filter(|width| *width <= 8)
            .ok_or(Error::Malformed {
                what: "/W entries must be integers from 0 to 8",
                offset: None,
            })?;
        *slot = width;
    }
    Ok(widths)
}

/// The `(first, count)` subsections from `/Index`, defaulting to `0 … /Size`.
fn xref_index(dict: &Dictionary) -> Result<Vec<(u32, u32)>> {
    let Some(index) = dict.get(b"Index") else {
        let size = dict
            .get(b"Size")
            .and_then(Object::as_integer)
            .and_then(|size| u32::try_from(size).ok())
            .ok_or(Error::Malformed {
                what: "cross-reference stream needs a positive /Size",
                offset: None,
            })?;
        return Ok(vec![(0, size)]);
    };
    let Object::Array(items) = index else {
        return Err(Error::Malformed {
            what: "/Index must be an array of integer pairs",
            offset: None,
        });
    };
    if items.len() % 2 != 0 {
        return Err(Error::Malformed {
            what: "/Index must hold an even number of integers",
            offset: None,
        });
    }
    let mut subsections = Vec::with_capacity(items.len() / 2);
    for pair in items.chunks(2) {
        let first = pair[0].as_integer().and_then(|v| u32::try_from(v).ok());
        let count = pair[1].as_integer().and_then(|v| u32::try_from(v).ok());
        match (first, count) {
            (Some(first), Some(count)) => subsections.push((first, count)),
            _ => {
                return Err(Error::Malformed {
                    what: "/Index pairs must be non-negative integers",
                    offset: None,
                })
            }
        }
    }
    Ok(subsections)
}

/// One big-endian field of a cross-reference stream entry.
fn read_xref_field(row: &[u8], cursor: &mut usize, width: u8, default: u64) -> Result<u64> {
    let width = usize::from(width);
    if width == 0 {
        return Ok(default);
    }
    let field = row.get(*cursor..*cursor + width).ok_or(Error::Malformed {
        what: "cross-reference stream entry is shorter than /W says",
        offset: None,
    })?;
    *cursor += width;
    Ok(field
        .iter()
        .fold(0_u64, |value, byte| (value << 8) | u64::from(*byte)))
}

fn as_integer(token: Token) -> Result<i64> {
    match token {
        Token::Integer(value) => Ok(value),
        _ => Err(Error::Malformed {
            what: "cross-reference entries must be integers",
            offset: None,
        }),
    }
}

#[cfg(test)]
mod tests {
    use super::{find_last, header_version, parse_xref_section, read_xref_chain, startxref_offset};

    /// A minimal classic-xref PDF built with correct offsets.
    fn minimal_pdf() -> Vec<u8> {
        let mut out = Vec::new();
        out.extend_from_slice(b"%PDF-2.0\n");
        let first = out.len();
        out.extend_from_slice(b"1 0 obj\n<< /Type /Catalog >>\nendobj\n");
        let xref = out.len();
        out.extend_from_slice(b"xref\n0 2\n");
        out.extend_from_slice(b"0000000000 65535 f \n");
        out.extend_from_slice(format!("{first:010} 00000 n \n").as_bytes());
        out.extend_from_slice(b"trailer\n<< /Size 2 /Root 1 0 R >>\n");
        out.extend_from_slice(format!("startxref\n{xref}\n%%EOF\n").as_bytes());
        out
    }

    #[test]
    fn header_versions_are_sniffed() {
        assert_eq!(header_version(b"%PDF-2.0\nrest").expect("header"), "2.0");
        assert_eq!(
            header_version(b"junk%PDF-2.0\r\nrest").expect("header"),
            "2.0",
            "the header may sit within the first 1024 bytes"
        );
        assert!(header_version(b"%PDF-1.7\n").is_err());
        assert!(header_version(b"no header at all").is_err());
        assert!(header_version(b"").is_err());
    }

    #[test]
    fn startxref_is_found_after_the_tables() {
        let pdf = minimal_pdf();
        let offset = startxref_offset(&pdf).expect("startxref");
        assert!(pdf[offset..].starts_with(b"xref"));
        assert!(startxref_offset(b"%PDF-2.0\n%%EOF").is_err());
        assert!(startxref_offset(b"%PDF-2.0\nstartxref\n999999\n%%EOF").is_err());
    }

    #[test]
    fn cross_reference_sections_parse_subsections() {
        let pdf = minimal_pdf();
        let offset = startxref_offset(&pdf).expect("startxref");
        let (entries, trailer) = parse_xref_section(&pdf, offset).expect("section");
        assert_eq!(entries.len(), 2);
        assert_eq!(trailer.get(b"Size"), Some(&crate::Object::Integer(2)));
        let (merged, newest) = read_xref_chain(&pdf).expect("chain");
        assert_eq!(merged.len(), 2);
        assert_eq!(newest.get(b"Size"), Some(&crate::Object::Integer(2)));
    }

    fn dict_of(entries: &[(&str, crate::Object)]) -> crate::Dictionary {
        let mut dict = crate::Dictionary::new();
        for (name, value) in entries {
            dict.insert(name.as_bytes(), value.clone());
        }
        dict
    }

    #[test]
    fn xref_stream_widths_and_index_have_spec_defaults() {
        let dict = dict_of(&[
            ("Size", crate::Object::Integer(7)),
            (
                "W",
                crate::Object::array([
                    crate::Object::Integer(1),
                    crate::Object::Integer(4),
                    crate::Object::Integer(2),
                ]),
            ),
        ]);
        assert_eq!(super::xref_widths(&dict).expect("widths"), [1, 4, 2]);
        assert_eq!(
            super::xref_index(&dict).expect("default index"),
            vec![(0, 7)],
            "/Index defaults to one subsection covering 0..Size"
        );

        let dict = dict_of(&[
            (
                "W",
                crate::Object::array([
                    crate::Object::Integer(1),
                    crate::Object::Integer(4),
                    crate::Object::Integer(2),
                ]),
            ),
            (
                "Index",
                crate::Object::array([
                    crate::Object::Integer(4),
                    crate::Object::Integer(1),
                    crate::Object::Integer(7),
                    crate::Object::Integer(1),
                ]),
            ),
        ]);
        assert_eq!(
            super::xref_index(&dict).expect("explicit index"),
            vec![(4, 1), (7, 1)]
        );

        let bad = dict_of(&[("Size", crate::Object::Integer(1))]);
        assert!(super::xref_widths(&bad).is_err(), "/W is required");
        let bad = dict_of(&[
            ("Size", crate::Object::Integer(1)),
            ("W", crate::Object::array([crate::Object::Integer(9)])),
        ]);
        assert!(super::xref_widths(&bad).is_err(), "/W larger than 8 bytes");
        let bad = dict_of(&[
            ("Size", crate::Object::Integer(1)),
            ("W", crate::Object::array([crate::Object::Integer(0)])),
        ]);
        assert!(super::xref_widths(&bad).is_err(), "/W must not be all zero");
    }

    #[test]
    fn zero_width_fields_take_their_defaults() {
        // `w0 = 0` means "type 1" and `w2 = 0` means "generation 0".
        let row = [0x00, 0x00, 0x01];
        let mut cursor = 0;
        assert_eq!(
            super::read_xref_field(&row, &mut cursor, 0, 1).expect("default"),
            1
        );
        assert_eq!(cursor, 0, "a zero-width field consumes nothing");
        assert_eq!(
            super::read_xref_field(&row, &mut cursor, 2, 0).expect("value"),
            0x0000
        );
        assert_eq!(
            super::read_xref_field(&row, &mut cursor, 1, 0).expect("value"),
            1
        );
        assert_eq!(cursor, 3);
        // Reading past the row is an error, not a panic.
        assert!(super::read_xref_field(&row, &mut cursor, 2, 0).is_err());
    }

    #[test]
    fn cross_reference_streams_map_all_three_entry_types() {
        // Three rows of /W [1 2 1]: free, in-use, compressed.
        let mut data =
            b"%PDF-2.0\n7 0 obj\n<< /Type /XRef /Size 3 /Length 12 /W [1 2 1] >>\nstream\n"
                .to_vec();
        let offset = 9;
        data.extend_from_slice(&[0, 0, 0, 0]);
        data.extend_from_slice(&[1, 0x03, 0xE7, 5]);
        data.extend_from_slice(&[2, 0x00, 0x07, 3]);
        data.extend_from_slice(b"\nendstream\nendobj\n");
        let (entries, dict) = super::parse_xref_stream(&data, offset).expect("section");
        assert_eq!(dict.get(b"Size"), Some(&crate::Object::Integer(3)));
        assert_eq!(
            entries.get(&0),
            Some(&super::XrefEntry::Free { generation: 0 })
        );
        assert_eq!(
            entries.get(&1),
            Some(&super::XrefEntry::InUse {
                offset: 999,
                generation: 5
            })
        );
        assert_eq!(
            entries.get(&2),
            Some(&super::XrefEntry::Compressed {
                stream: 7,
                index: 3
            })
        );
    }

    #[test]
    fn a_cross_reference_stream_without_type_is_rejected() {
        let mut data = b"7 0 obj\n<< /Size 3 /Length 4 /W [1 2 1] >>\nstream\n".to_vec();
        let offset = 9;
        data.extend_from_slice(&[0, 0, 0, 0]);
        data.extend_from_slice(b"\nendstream\nendobj\n");
        assert!(super::parse_xref_stream(&data, offset).is_err());
    }

    #[test]
    fn find_last_returns_the_final_match() {
        assert_eq!(find_last(b"abab", b"ab"), Some(2));
        assert_eq!(find_last(b"abab", b"zz"), None);
        assert_eq!(find_last(b"a", b"ab"), None);
    }
}