flpdf 0.2.0

A pure-Rust PDF toolkit modeled on qpdf — read, rewrite, linearize, and inspect PDFs.
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
use crate::parser::{parse_indirect_object, Parser};
use crate::{filters, Diagnostic, Diagnostics, Dictionary, Error, Object, ObjectRef, Result};
use std::collections::{BTreeMap, HashSet};
use std::io::{Read, Seek, SeekFrom};

#[derive(Debug, Clone)]
pub struct LoadedXref {
    pub version: String,
    pub startxref: u64,
    pub entries: BTreeMap<ObjectRef, XrefOffset>,
    pub trailer: Dictionary,
    pub last_xref_form: XrefForm,
    pub repair_diagnostics: Diagnostics,
}

#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum XrefOffset {
    Free { next: u32 },
    Offset(u64),
    Compressed { stream: u32, index: u32 },
}

#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum XrefForm {
    Table,
    Stream,
}

/// Load the cross-reference table and trailer dictionary from `reader`, with
/// the qpdf-style recovery pass disabled (strict parse).
///
/// # Errors
///
/// Calls [`load_xref_and_trailer_with_repair`] with repair disabled, so it
/// propagates the same errors that function raises when `allow_repair` is
/// `false`:
///
/// - [`Error::Io`] when reading the input fails.
/// - [`Error::Parse`] when the PDF header, `startxref`, or a cross-reference
///   section is malformed (including a `startxref`/`/Prev` offset that does not
///   fit `usize` and a circular `/Prev` chain).
/// - [`Error::Missing`] when a required cross-reference stream entry (such as
///   `/Size` or `/W`) is absent.
/// - [`Error::Unsupported`] when a cross-reference stream uses an unsupported
///   object or entry type.
pub fn load_xref_and_trailer<R: Read + Seek>(reader: &mut R) -> Result<LoadedXref> {
    load_xref_and_trailer_with_repair(reader, false)
}

/// Load the cross-reference table and trailer dictionary from `reader`, running
/// the qpdf-style recovery pass when `allow_repair` is `true`.
///
/// # Errors
///
/// - [`Error::Io`] when seeking or reading the input fails.
/// - [`Error::Parse`] when the PDF header is missing or its version is not
///   UTF-8 (this check runs before any repair fallback). When `allow_repair`
///   is `false`, also when `startxref`, a cross-reference table or stream, or a
///   `/Prev` chain is malformed (including offsets that do not fit `usize` and a
///   circular `/Prev` chain). When `allow_repair` is `true`, such failures are
///   recorded as diagnostics and recovered by a linear scan, which itself
///   raises [`Error::Parse`] when no cross-reference entries can be recovered or
///   no `trailer` dictionary is found.
/// - [`Error::Missing`] when a required cross-reference stream entry (such as
///   `/Size` or `/W`) is absent and `allow_repair` is `false`.
/// - [`Error::Unsupported`] when a cross-reference stream uses an unsupported
///   object or entry type and `allow_repair` is `false`.
pub fn load_xref_and_trailer_with_repair<R: Read + Seek>(
    reader: &mut R,
    allow_repair: bool,
) -> Result<LoadedXref> {
    let mut bytes = Vec::new();
    reader.seek(SeekFrom::Start(0))?;
    reader.read_to_end(&mut bytes)?;

    let version = parse_header(&bytes)?;
    let mut parse_errors = Vec::new();
    let startxref = match parse_startxref(&bytes) {
        Ok(offset) => offset,
        Err(error) if allow_repair => {
            parse_errors.push(error);
            0
        }
        Err(error) => return Err(error),
    };
    let xref_pos = match usize::try_from(startxref) {
        Ok(xref_pos) => xref_pos,
        Err(_) if allow_repair => {
            parse_errors.push(Error::parse(0, "startxref does not fit usize"));
            0
        }
        Err(_) => return Err(Error::parse(0, "startxref does not fit usize")),
    };

    let mut loaded = match parse_xref_from_start(&bytes, xref_pos, startxref, &version) {
        Ok(loaded) => loaded,
        Err(error) if allow_repair => {
            // Report the first recorded failure; this parse error is only the
            // trigger when the startxref stage itself succeeded.
            let trigger = parse_errors.into_iter().next().unwrap_or(error);
            return recover_xref_from_linear_scan(&bytes, version, startxref, trigger);
        }
        Err(error) => return Err(error),
    };

    if let Err(error) = merge_previous_xref_sections(
        &bytes,
        &version,
        &mut loaded.entries,
        &loaded.trailer,
        allow_repair,
    ) {
        if allow_repair {
            let trigger = parse_errors.into_iter().next().unwrap_or(error);
            return recover_xref_from_linear_scan(&bytes, version, startxref, trigger);
        }
        return Err(error);
    }

    if let Some(error) = parse_errors.into_iter().next() {
        push_repair_diagnostics(&mut loaded.repair_diagnostics, &error, startxref);
    }

    Ok(loaded)
}

fn parse_xref_from_start(
    bytes: &[u8],
    xref_pos: usize,
    startxref: u64,
    version: &str,
) -> Result<LoadedXref> {
    if bytes
        .get(xref_pos..)
        .is_some_and(|tail| tail.starts_with(b"xref"))
    {
        let mut cursor = ByteCursor::new(bytes, xref_pos + 4);
        let (entries, trailer) = parse_xref_table(&mut cursor, bytes)?;
        return Ok(LoadedXref {
            version: version.to_string(),
            startxref,
            entries,
            trailer,
            last_xref_form: XrefForm::Table,
            repair_diagnostics: Diagnostics::default(),
        });
    }

    parse_xref_stream(bytes, xref_pos, startxref, version.to_string())
}

fn merge_previous_xref_sections(
    bytes: &[u8],
    version: &str,
    entries: &mut BTreeMap<ObjectRef, XrefOffset>,
    trailer: &Dictionary,
    allow_repair: bool,
) -> Result<()> {
    let mut visited = HashSet::new();
    let mut previous_offset = parse_previous_xref_offset(trailer);

    while let Some(offset) = previous_offset {
        let previous_pos = usize::try_from(offset)
            .map_err(|_| Error::parse(0, "xref /Prev does not fit usize"))?;

        if !visited.insert(offset) {
            return if allow_repair {
                Ok(())
            } else {
                Err(Error::parse(0, "xref /Prev is circular"))
            };
        }

        let previous = parse_xref_from_start(bytes, previous_pos, offset, version)?;

        for (object_ref, xref_offset) in previous.entries {
            if !entries
                .keys()
                .any(|entry_ref| entry_ref.number == object_ref.number)
            {
                entries.insert(object_ref, xref_offset);
            }
        }

        previous_offset = parse_previous_xref_offset(&previous.trailer);
    }

    Ok(())
}

fn parse_previous_xref_offset(trailer: &Dictionary) -> Option<u64> {
    trailer
        .get("Prev")
        .and_then(|offset| parse_non_negative_u64(offset, "/Prev").ok())
        .filter(|&offset| offset != 0)
}

fn recover_xref_from_linear_scan(
    bytes: &[u8],
    version: String,
    startxref: u64,
    trigger_error: Error,
) -> Result<LoadedXref> {
    let entries = recover_xref_entries(bytes)?;
    let trailer = recover_trailer(bytes)?;

    let mut repair_diagnostics = Diagnostics::default();
    push_repair_diagnostics(&mut repair_diagnostics, &trigger_error, startxref);

    Ok(LoadedXref {
        version,
        startxref,
        entries,
        trailer,
        last_xref_form: XrefForm::Table,
        repair_diagnostics,
    })
}

/// Load the cross-reference table and trailer dictionary from `reader`, with the
/// qpdf-style recovery pass always enabled (best-effort).
///
/// # Errors
///
/// Calls [`load_xref_and_trailer_with_repair`] with repair enabled, so
/// malformed cross-reference data is recovered rather than reported. It still
/// fails with:
///
/// - [`Error::Io`] when seeking or reading the input fails.
/// - [`Error::Parse`] when the PDF header is missing or its version is not
///   UTF-8, or when the linear-scan recovery cannot find any cross-reference
///   entries or a `trailer` dictionary.
pub fn load_xref_and_trailer_best_effort<R: Read + Seek>(reader: &mut R) -> Result<LoadedXref> {
    load_xref_and_trailer_with_repair(reader, true)
}

/// Recover uncompressed object offsets by replaying qpdf's `reconstruct_xref`
/// (`libqpdf/QPDF.cc`, qpdf 11.9.0): scan the file line by line, and on each line
/// whose first token sequence is `int int obj`, record the object at the offset of
/// its *number* token. Only the first token of a line is inspected, object bodies
/// are never parsed, and the last occurrence of an object in the file wins (qpdf's
/// `insertReconstructedXrefEntry` overwrites). Inspecting at most three short
/// tokens per line — never re-parsing a body to end-of-file — makes the scan
/// linear in the file size, unlike a per-candidate full-object parse which an
/// unterminated literal string can drive to quadratic cost.
fn recover_xref_entries(bytes: &[u8]) -> Result<BTreeMap<ObjectRef, XrefOffset>> {
    let mut entries = BTreeMap::new();
    let mut line_start = 0usize;
    while line_start < bytes.len() {
        let next_line_start = next_line_start(bytes, line_start);
        if let Some((object_ref, offset)) =
            scan_object_header_at_line(bytes, line_start, next_line_start)
        {
            entries.insert(object_ref, XrefOffset::Offset(offset));
        }
        line_start = next_line_start;
    }

    // qpdf records only uncompressed (type-1) entries during reconstruction and
    // declines to look inside object streams (`reconstruct_xref` trailing comment
    // in QPDF.cc). flpdf additionally recovers the objects packed in a recovered
    // `/Type /ObjStm` so they remain resolvable without a usable xref; this extra
    // pass is bounded per object to keep recovery linear (see below).
    recover_objstm_compressed_entries(bytes, &mut entries);

    if entries.is_empty() {
        return Err(Error::parse(
            0,
            "unable to recover xref entries by linear scan",
        ));
    }

    Ok(entries)
}

/// Upper bound on read-to-end fallbacks during ObjStm recovery (see
/// [`recover_objstm_compressed_entries`]). Each fallback may parse to end of
/// file, so the count is capped to keep the total work O(file size) while still
/// recovering a handful of object streams whose payloads happen to contain a
/// header-like line.
const MAX_OBJSTM_RECOVERY_FALLBACKS: u32 = 64;

/// Recover the compressed objects packed in any recovered `/Type /ObjStm`,
/// emitting `XrefOffset::Compressed` entries that point back at the stream.
///
/// Each recovered object is parsed within the window that ends at the next
/// recovered object's offset (or end-of-file for the last). The windows are
/// disjoint, so the common case is bounded by the file size — a malformed object
/// cannot drive the parse to end-of-file once per candidate. When a window does
/// not hold a complete object — a header-like line (`int int obj`) recorded
/// inside an object stream's payload became the next offset and truncated it —
/// it retries against the rest of the file so the stream's own `/Length`
/// delimits it. Those retries are capped by [`MAX_OBJSTM_RECOVERY_FALLBACKS`] so
/// a flood of stream-like candidates cannot reintroduce quadratic cost.
fn recover_objstm_compressed_entries(bytes: &[u8], entries: &mut BTreeMap<ObjectRef, XrefOffset>) {
    // The line scan only ever inserts `XrefOffset::Offset`, so every entry here is
    // an uncompressed object whose offset bounds a window.
    let mut offsets: Vec<u64> = Vec::new();
    for entry in entries.values() {
        if let XrefOffset::Offset(offset) = entry {
            offsets.push(*offset);
        }
    }
    offsets.sort_unstable();

    let mut fallbacks = MAX_OBJSTM_RECOVERY_FALLBACKS;
    for (index, &offset) in offsets.iter().enumerate() {
        let start = offset as usize;
        let window_end = offsets
            .get(index + 1)
            .map_or(bytes.len(), |next| *next as usize);
        if try_recover_objstm_in(entries, &bytes[start..window_end]) {
            continue;
        }
        // The bounded window stopped short of a complete object. Retry against
        // the rest of the file so a real ObjStm truncated by a header-like line
        // in its payload is still recovered, capped so it stays linear.
        if window_end < bytes.len() && fallbacks > 0 {
            fallbacks -= 1;
            try_recover_objstm_in(entries, &bytes[start..]);
        }
    }
}

/// Parse the indirect object in `slice`; if it is a `/Type /ObjStm`, insert its
/// packed objects' compressed entries. Returns `false` only when `slice` did not
/// contain a complete object (a parse error) — the signal that a bounded window
/// may have truncated a real stream and a wider retry is worthwhile.
fn try_recover_objstm_in(entries: &mut BTreeMap<ObjectRef, XrefOffset>, slice: &[u8]) -> bool {
    match parse_indirect_object(slice) {
        Ok((object_ref, Object::Stream(stream))) => {
            if let Some(Object::Name(type_name)) = stream.dict.get("Type") {
                if type_name.as_slice() == b"ObjStm" {
                    recover_compressed_offsets_from_objstm(entries, object_ref, &stream);
                }
            }
            true
        }
        Ok(_) => true,
        Err(_) => false,
    }
}

fn recover_compressed_offsets_from_objstm(
    entries: &mut BTreeMap<ObjectRef, XrefOffset>,
    stream_ref: ObjectRef,
    stream: &crate::Stream,
) {
    let Ok(decoded_data) = crate::filters::decode_stream_data(&stream.dict, &stream.data) else {
        return;
    };

    let object_count =
        match parse_non_negative_u64(stream.dict.get("N").unwrap_or(&Object::Integer(0)), "/N") {
            Ok(count) => match usize::try_from(count) {
                Ok(count) => count,
                Err(_) => return,
            },
            Err(_) => return,
        };

    let mut cursor = Parser::new(&decoded_data);
    for index in 0..object_count {
        let number = match cursor.integer_for_indirect() {
            Ok(number) => match parse_non_negative_i64(number, "ObjStm object number") {
                Ok(number) => number,
                Err(_) => return,
            },
            Err(_) => return,
        };
        let object_ref = match u32::try_from(number) {
            Ok(object_ref) => ObjectRef::new(object_ref, 0),
            Err(_) => return,
        };

        match cursor.integer_for_indirect() {
            Ok(offset) => {
                if parse_non_negative_i64(offset, "ObjStm object offset").is_err() {
                    return;
                }
                entries.entry(object_ref).or_insert(XrefOffset::Compressed {
                    stream: stream_ref.number,
                    index: u32::try_from(index).unwrap_or(u32::MAX),
                });
            }
            Err(_) => return,
        }
    }
}

fn parse_non_negative_i64(value: i64, name: &str) -> Result<u64> {
    if value < 0 {
        return Err(Error::parse(0, format!("{name} is negative")));
    }
    Ok(value as u64)
}

/// Push the qpdf-compatible repair warning sequence onto `diagnostics`.
///
/// qpdf (`reconstruct_xref` in `QPDF_objects.cc`, observed with qpdf 11.9.0)
/// emits the same three warnings regardless of how the damaged
/// cross-reference data is ultimately recovered: `file is damaged`, the error
/// that triggered recovery, and `Attempting to reconstruct cross-reference
/// table`. `trigger_error` is the first failure that initiated recovery;
/// subsequent failures from the retry-at-offset-0 detour are not reported
/// because qpdf has no such detour and they have no counterpart on its
/// stderr. The triggering error's warning carries that error's own byte
/// offset when available (falling back to the `startxref` offset); the
/// surrounding warnings carry no offset, matching qpdf, which reports them
/// at offset 0 and suppresses the display.
fn push_repair_diagnostics(diagnostics: &mut Diagnostics, trigger_error: &Error, startxref: u64) {
    diagnostics.push(Diagnostic::warning("file is damaged", None));
    let (message, offset) = match trigger_error {
        Error::Parse { offset, message } => (message.clone(), Some(*offset as u64)),
        other => (other.to_string(), Some(startxref)),
    };
    diagnostics.push(Diagnostic::warning(message, offset));
    diagnostics.push(Diagnostic::warning(
        "Attempting to reconstruct cross-reference table",
        None,
    ));
}

fn recover_trailer(bytes: &[u8]) -> Result<Dictionary> {
    let marker = b"trailer";
    let Some(pos) = bytes
        .windows(marker.len())
        .rposition(|window| window == marker)
    else {
        return Err(Error::parse(0, "trailer dictionary not found"));
    };

    let mut cursor = ByteCursor::new(bytes, pos + marker.len());
    cursor.skip_ws();
    let mut parser = Parser::new(&bytes[cursor.pos..]);
    match parser.object()? {
        Object::Dictionary(trailer) => Ok(trailer),
        _ => Err(Error::parse(
            cursor.pos + parser.position(),
            "trailer dictionary is not a dictionary",
        )),
    }
}

/// Return the offset just past the next end-of-line at or after `from`, or
/// `bytes.len()` when no further end-of-line exists. A run of consecutive
/// `\r`/`\n` bytes is treated as a single line terminator (mirroring qpdf's
/// `findAndSkipNextEOL`, which collapses `\r\n` and blank lines). When
/// `from < bytes.len()` the result is always strictly greater than `from`, so
/// the line scan in [`recover_xref_entries`] always makes progress.
fn next_line_start(bytes: &[u8], from: usize) -> usize {
    let mut pos = from;
    while pos < bytes.len() && !matches!(bytes[pos], b'\n' | b'\r') {
        pos += 1;
    }
    // Skip the run of end-of-line bytes so blank lines do not become their own
    // iterations; this keeps the scan linear by advancing `line_start` past the
    // whole run that a forward token read would otherwise re-scan. When no
    // end-of-line exists this loop is a no-op and `pos` is already `bytes.len()`.
    while pos < bytes.len() && matches!(bytes[pos], b'\n' | b'\r') {
        pos += 1;
    }
    pos
}

/// PDF whitespace: NUL, TAB, LF, FF, CR, and space (ISO 32000-2 Table 1).
fn is_pdf_whitespace(byte: u8) -> bool {
    matches!(byte, b'\0' | b'\t' | b'\n' | b'\x0c' | b'\r' | b' ')
}

/// PDF delimiters plus `%` (comment introducer); any of these ends a regular
/// token (ISO 32000-2 Table 2).
fn is_pdf_delimiter(byte: u8) -> bool {
    matches!(
        byte,
        b'(' | b')' | b'<' | b'>' | b'[' | b']' | b'{' | b'}' | b'/' | b'%'
    )
}

/// Advance `pos` past PDF whitespace and `%...EOL` comments — the two ignorable
/// constructs allowed between tokens — but never beyond `limit`. qpdf's
/// tokenizer skips comments in this position too, so an object header with a
/// comment between its tokens (`1 %c<EOL>0 obj`) is still recovered.
fn skip_ignorable(bytes: &[u8], mut pos: usize, limit: usize) -> usize {
    loop {
        while pos < limit && is_pdf_whitespace(bytes[pos]) {
            pos += 1;
        }
        if pos < limit && bytes[pos] == b'%' {
            while pos < limit && !matches!(bytes[pos], b'\n' | b'\r') {
                pos += 1;
            }
        } else {
            return pos;
        }
    }
}

/// A token read by the reconstruction scanner, with its byte range. Only the
/// distinctions the `int int obj` match needs are modelled.
struct ScanToken {
    start: usize,
    end: usize,
    kind: ScanKind,
}

enum ScanKind {
    /// A numeric token (`[+-]?[0-9]+` that fits `i64`).
    Integer(i64),
    /// A non-numeric regular token (compared to `obj` via the byte range).
    Word,
    /// A delimiter-led token. The scanner stops as soon as a slot it expected to
    /// be `int`/`obj` is one of these.
    Delimiter,
}

/// Read the next token whose start lies in `[from, limit)`, skipping leading
/// whitespace and `%...EOL` comments. Returns `None` when no token starts before
/// `limit` (the region is only whitespace/comments).
///
/// `limit` bounds where the token may *start*, not where it ends — a regular
/// token still runs to the next whitespace/delimiter. Passing `next_line_start`
/// for the first token keeps the scan linear: a whitespace- or comment-only line
/// is rejected after scanning only its own bytes, instead of skipping forward
/// into later lines and re-scanning that suffix on every iteration (an O(n^2)
/// blowup). Later tokens pass `bytes.len()` so an `int int obj` header may span
/// lines, matching qpdf.
fn read_scan_token(bytes: &[u8], from: usize, limit: usize) -> Option<ScanToken> {
    let start = skip_ignorable(bytes, from, limit);
    if start >= limit {
        return None;
    }
    let first = bytes[start];
    if is_pdf_delimiter(first) {
        return Some(ScanToken {
            start,
            end: start + 1,
            kind: ScanKind::Delimiter,
        });
    }

    let mut pos = start;
    while pos < bytes.len() && !is_pdf_whitespace(bytes[pos]) && !is_pdf_delimiter(bytes[pos]) {
        pos += 1;
    }
    let end = pos;
    let kind = match parse_scan_integer(&bytes[start..end]) {
        Some(value) => ScanKind::Integer(value),
        None => ScanKind::Word,
    };
    Some(ScanToken { start, end, kind })
}

/// Classify a regular token as an unsigned PDF integer (`[0-9]+` that fits
/// `i64`). Returns `None` for any token containing a non-digit (a word such as
/// `obj`) or a digit run too long to fit `i64`. Object and generation numbers in
/// real PDFs are unsigned, so a leading sign is treated as a word; this only ever
/// changes the outcome for synthetic `+N`/`-N` headers, which qpdf's
/// `obj > 0`/`gen >= 0` guards reject anyway.
fn parse_scan_integer(token: &[u8]) -> Option<i64> {
    if token.is_empty() || !token.iter().all(u8::is_ascii_digit) {
        return None;
    }
    std::str::from_utf8(token).ok()?.parse().ok()
}

/// If the line beginning at `line_start` opens with an `int int obj` token
/// sequence, return the recovered object and the offset of its number token.
///
/// Mirrors qpdf's `reconstruct_xref` per-line logic: the first token must begin
/// on this line (otherwise the line records nothing — qpdf's
/// `token_start >= next_line_start` guard, here enforced by bounding the first
/// token read to `next_line_start`), the second and third tokens may spill onto
/// following lines, and the object/generation must satisfy qpdf's
/// `insertReconstructedXrefEntry` guards (`obj > 0`, `0 <= gen < 65535`).
fn scan_object_header_at_line(
    bytes: &[u8],
    line_start: usize,
    next_line_start: usize,
) -> Option<(ObjectRef, u64)> {
    // Bounding the first token to this line is what keeps a whitespace- or
    // comment-only line from re-scanning the remaining file on every iteration.
    let number_token = read_scan_token(bytes, line_start, next_line_start)?;
    let ScanKind::Integer(obj) = number_token.kind else {
        return None;
    };

    let gen_token = read_scan_token(bytes, number_token.end, bytes.len())?;
    let ScanKind::Integer(gen) = gen_token.kind else {
        return None;
    };

    let obj_token = read_scan_token(bytes, gen_token.end, bytes.len())?;
    if !matches!(obj_token.kind, ScanKind::Word) || &bytes[obj_token.start..obj_token.end] != b"obj"
    {
        return None;
    }

    // qpdf's `insertReconstructedXrefEntry` guards (`obj > 0`, `0 <= gen < 65535`).
    if obj <= 0 || !(0..65535).contains(&gen) {
        return None;
    }
    let number = u32::try_from(obj).ok()?;
    let generation = u16::try_from(gen).ok()?;
    Some((
        ObjectRef::new(number, generation),
        number_token.start as u64,
    ))
}

fn parse_xref_table(
    cursor: &mut ByteCursor<'_>,
    bytes: &[u8],
) -> Result<(BTreeMap<ObjectRef, XrefOffset>, Dictionary)> {
    let mut entries = BTreeMap::new();
    loop {
        cursor.skip_ws();
        if cursor.starts_with(b"trailer") {
            cursor.pos += b"trailer".len();
            break;
        }

        let first = cursor.read_u32()?;
        let count = cursor.read_u32()?;
        for index in 0..count {
            cursor.skip_ws();
            let offset = cursor.read_fixed_u64(10)?;
            cursor.skip_ws();
            let generation = cursor.read_fixed_u16(5)?;
            cursor.skip_ws();
            let in_use = cursor.read_byte()?;
            cursor.skip_line();
            match in_use {
                b'f' => {
                    entries.insert(
                        ObjectRef::new(first + index, generation),
                        XrefOffset::Free {
                            next: u32::try_from(offset).map_err(|_| {
                                Error::parse(0, "free xref next object does not fit u32")
                            })?,
                        },
                    );
                }
                b'n' => {
                    entries.insert(
                        ObjectRef::new(first + index, generation),
                        XrefOffset::Offset(offset),
                    );
                }
                _ => return Err(Error::parse(0, "xref table entry status is not f or n")),
            }
        }
    }

    cursor.skip_ws();
    let mut parser = Parser::new(&bytes[cursor.pos..]);
    let trailer = match parser.object()? {
        Object::Dictionary(dict) => dict,
        _ => {
            return Err(Error::parse(
                cursor.pos + parser.position(),
                "trailer is not a dictionary",
            ));
        }
    };

    Ok((entries, trailer))
}

fn parse_xref_stream(
    bytes: &[u8],
    xref_pos: usize,
    startxref: u64,
    version: String,
) -> Result<LoadedXref> {
    let tail = bytes
        .get(xref_pos..)
        .filter(|slice| !slice.is_empty())
        .ok_or_else(|| Error::parse(xref_pos, "xref stream offset is beyond end of file"))?;
    let (_, object) = parse_indirect_object(tail).map_err(|err| err.rebase_offset(xref_pos))?;
    let stream = match object {
        Object::Stream(stream) => stream,
        _ => {
            return Err(Error::Unsupported(
                "xref stream expected an indirect object stream".to_string(),
            ))
        }
    };

    let trailer = stream.dict.clone();
    let size = parse_non_negative_u64(
        trailer
            .get("Size")
            .ok_or(Error::Missing("XRef stream /Size"))?,
        "/Size",
    )?;
    let size = u32::try_from(size).map_err(|_| Error::parse(0, "/Size does not fit u32"))?;

    let widths = parse_xref_widths(&trailer)?;
    let index = parse_xref_index(&trailer, size)?;
    let ranges = build_xref_ranges(index)?;
    let stream_data = filters::decode_stream_data(&stream.dict, &stream.data)?;
    let mut cursor = ByteCursor::new(&stream_data, 0);
    let entries = parse_xref_entries(&mut cursor, size, &ranges, widths)?;

    Ok(LoadedXref {
        version,
        startxref,
        entries,
        trailer,
        last_xref_form: XrefForm::Stream,
        repair_diagnostics: Diagnostics::default(),
    })
}

type XrefWidths = (usize, usize, usize);

fn parse_xref_widths(trailer: &Dictionary) -> Result<XrefWidths> {
    let Object::Array(values) = trailer.get("W").ok_or(Error::Missing("XRef stream /W"))? else {
        return Err(Error::parse(0, "/W must be array"));
    };

    if values.len() != 3 {
        return Err(Error::parse(0, "/W must contain three integers"));
    }

    let w0 = parse_usize(parse_non_negative_u64(&values[0], "/W[0]")?, "/W[0]")?;
    let w1 = parse_usize(parse_non_negative_u64(&values[1], "/W[1]")?, "/W[1]")?;
    let w2 = parse_usize(parse_non_negative_u64(&values[2], "/W[2]")?, "/W[2]")?;

    Ok((w0, w1, w2))
}

fn parse_xref_index(trailer: &Dictionary, size: u32) -> Result<Vec<u32>> {
    match trailer.get("Index") {
        None => Ok(vec![0, size]),
        Some(Object::Array(values)) => {
            if values.len() % 2 != 0 {
                return Err(Error::parse(
                    0,
                    "/Index must contain an even number of integers",
                ));
            }

            let mut index = Vec::with_capacity(values.len());
            for value in values {
                let integer = parse_non_negative_u64(value, "/Index")?;
                index.push(
                    integer
                        .try_into()
                        .map_err(|_| Error::parse(0, "xref /Index value must fit u32"))?,
                );
            }
            Ok(index)
        }
        _ => Err(Error::parse(0, "/Index must be array")),
    }
}

fn build_xref_ranges(index: Vec<u32>) -> Result<Vec<(u32, u32)>> {
    let mut ranges = Vec::with_capacity(index.len() / 2);
    for chunk in index.chunks_exact(2) {
        if chunk[1] == 0 {
            continue;
        }
        ranges.push((chunk[0], chunk[1]));
    }
    Ok(ranges)
}

fn parse_xref_entries(
    cursor: &mut ByteCursor<'_>,
    size: u32,
    ranges: &[(u32, u32)],
    widths: XrefWidths,
) -> Result<BTreeMap<ObjectRef, XrefOffset>> {
    let (w0, w1, w2) = widths;
    let entry_width = w0 + w1 + w2;
    if entry_width == 0 {
        return Err(Error::parse(0, "invalid cross-reference stream widths"));
    }

    let mut entries = BTreeMap::new();
    for &(start, count) in ranges {
        let start =
            usize::try_from(start).map_err(|_| Error::parse(0, "object number too large"))?;
        let count = usize::try_from(count).map_err(|_| Error::parse(0, "range count too large"))?;

        for index in 0..count {
            if start + index >= usize::try_from(size).unwrap_or(usize::MAX) {
                return Err(Error::parse(0, "xref range exceeds /Size"));
            }

            if cursor.pos + entry_width > cursor.bytes.len() {
                return Err(Error::parse(cursor.pos, "xref stream data truncated"));
            }

            let object_type = if w0 == 0 {
                1
            } else {
                let value = cursor.read_be_u64(w0)?;
                u8::try_from(value).map_err(|_| {
                    Error::parse(cursor.pos, "xref stream object type does not fit u8")
                })?
            };
            let field1 = if w1 == 0 { 0 } else { cursor.read_be_u64(w1)? };
            let field2 = if w2 == 0 { 0 } else { cursor.read_be_u64(w2)? };

            let object_number = (start + index) as u32;
            match object_type {
                0 => {
                    let next = u32::try_from(field1)
                        .map_err(|_| Error::parse(0, "free xref next object does not fit u32"))?;
                    let generation = u16::try_from(field2)
                        .map_err(|_| Error::parse(0, "generation does not fit u16"))?;
                    entries.insert(
                        ObjectRef::new(object_number, generation),
                        XrefOffset::Free { next },
                    );
                }
                1 => {
                    let generation = u16::try_from(field2)
                        .map_err(|_| Error::parse(0, "generation does not fit u16"))?;
                    entries.insert(
                        ObjectRef::new(object_number, generation),
                        XrefOffset::Offset(field1),
                    );
                }
                2 => {
                    let stream = u32::try_from(field1).map_err(|_| {
                        Error::parse(0, "xref stream object number does not fit u32")
                    })?;
                    let index = u32::try_from(field2)
                        .map_err(|_| Error::parse(0, "xref stream index does not fit u32"))?;
                    entries.insert(
                        ObjectRef::new(object_number, 0),
                        XrefOffset::Compressed { stream, index },
                    );
                }
                _ => {
                    return Err(Error::Unsupported(format!(
                        "unsupported xref entry type {object_type}"
                    )))
                }
            }
        }
    }

    Ok(entries)
}

fn parse_non_negative_u64(value: &Object, name: &str) -> Result<u64> {
    let Object::Integer(integer) = value else {
        return Err(Error::parse(0, format!("{name} is not integer")));
    };
    if *integer < 0 {
        return Err(Error::parse(0, format!("{name} is negative")));
    }
    Ok(*integer as u64)
}

fn parse_usize(value: u64, name: &str) -> Result<usize> {
    usize::try_from(value).map_err(|_| Error::parse(0, format!("{name} does not fit usize")))
}

fn parse_header(bytes: &[u8]) -> Result<String> {
    if !bytes.starts_with(b"%PDF-") {
        return Err(Error::parse(0, "missing PDF header"));
    }

    let end = bytes
        .iter()
        .position(|byte| *byte == b'\n' || *byte == b'\r')
        .unwrap_or(bytes.len());
    let header = std::str::from_utf8(&bytes[5..end])
        .map_err(|_| Error::parse(5, "PDF version is not utf-8"))?;
    Ok(header.to_string())
}

fn parse_startxref(bytes: &[u8]) -> Result<u64> {
    let marker = b"startxref";
    let Some(pos) = bytes
        .windows(marker.len())
        .rposition(|window| window == marker)
    else {
        return Err(Error::parse(bytes.len(), "can't find startxref"));
    };

    let mut cursor = ByteCursor::new(bytes, pos + marker.len());
    cursor.skip_ws();
    cursor.read_u64()
}

struct ByteCursor<'a> {
    bytes: &'a [u8],
    pos: usize,
}

impl<'a> ByteCursor<'a> {
    fn new(bytes: &'a [u8], pos: usize) -> Self {
        Self { bytes, pos }
    }

    fn starts_with(&self, token: &[u8]) -> bool {
        self.bytes[self.pos..].starts_with(token)
    }

    fn skip_ws(&mut self) {
        while matches!(
            self.bytes.get(self.pos),
            Some(b'\0' | b'\t' | b'\n' | b'\x0c' | b'\r' | b' ')
        ) {
            self.pos += 1;
        }
    }

    fn skip_line(&mut self) {
        while !matches!(self.bytes.get(self.pos), None | Some(b'\n' | b'\r')) {
            self.pos += 1;
        }
        while matches!(self.bytes.get(self.pos), Some(b'\n' | b'\r')) {
            self.pos += 1;
        }
    }

    fn read_byte(&mut self) -> Result<u8> {
        let Some(byte) = self.bytes.get(self.pos).copied() else {
            return Err(Error::parse(self.pos, "unexpected end of input"));
        };
        self.pos += 1;
        Ok(byte)
    }

    fn read_u32(&mut self) -> Result<u32> {
        let value = self.read_unsigned()?;
        u32::try_from(value).map_err(|_| Error::parse(self.pos, "number does not fit u32"))
    }

    fn read_u64(&mut self) -> Result<u64> {
        self.read_unsigned()
    }

    fn read_fixed_u64(&mut self, width: usize) -> Result<u64> {
        self.read_fixed(width)?
            .parse::<u64>()
            .map_err(|_| Error::parse(self.pos, "invalid fixed-width u64"))
    }

    fn read_be_u64(&mut self, width: usize) -> Result<u64> {
        if self.pos + width > self.bytes.len() {
            return Err(Error::parse(self.pos, "unexpected end of stream field"));
        }

        let mut value = 0u64;
        for _ in 0..width {
            value = (value << 8) | u64::from(self.bytes[self.pos]);
            self.pos += 1;
        }
        Ok(value)
    }

    fn read_fixed_u16(&mut self, width: usize) -> Result<u16> {
        self.read_fixed(width)?
            .parse::<u16>()
            .map_err(|_| Error::parse(self.pos, "invalid fixed-width u16"))
    }

    fn read_fixed(&mut self, width: usize) -> Result<&str> {
        if self.pos + width > self.bytes.len() {
            return Err(Error::parse(
                self.pos,
                "unexpected end of fixed-width field",
            ));
        }
        let text = std::str::from_utf8(&self.bytes[self.pos..self.pos + width])
            .map_err(|_| Error::parse(self.pos, "field is not utf-8"))?;
        self.pos += width;
        Ok(text)
    }

    fn read_unsigned(&mut self) -> Result<u64> {
        self.skip_ws();
        let start = self.pos;
        while matches!(self.bytes.get(self.pos), Some(b'0'..=b'9')) {
            self.pos += 1;
        }
        if start == self.pos {
            return Err(Error::parse(start, "expected unsigned integer"));
        }

        let text = std::str::from_utf8(&self.bytes[start..self.pos])
            .map_err(|_| Error::parse(start, "number is not utf-8"))?;
        text.parse::<u64>()
            .map_err(|_| Error::parse(start, "invalid unsigned integer"))
    }
}