flpdf 0.3.0

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
use crate::{Dictionary, Error, Object, ObjectRef, Result, Stream};

/// Parse a single PDF object from `input`, which must contain nothing but
/// that object (apart from trailing whitespace).
///
/// # Errors
///
/// - Returns [`Error::Parse`] if `input` does not contain a syntactically
///   valid PDF object, propagated from the underlying object parser.
/// - Returns [`Error::Parse`] with `"trailing bytes after object"` if any
///   non-whitespace bytes remain after the object has been parsed.
pub fn parse_object(input: &[u8]) -> Result<Object> {
    let mut parser = Parser::new(input);
    let object = parser.object()?;
    parser.skip_ws();
    if parser.pos != parser.input.len() {
        return Err(Error::parse(parser.pos, "trailing bytes after object"));
    }
    Ok(object)
}

/// A stream object whose `/Length` is an indirect reference `M G R`. The
/// parser cannot resolve `M` (it has no xref); it locates the payload window
/// via the spec `endstream`-scan recovery and records these bounds so the
/// reader — which *does* have the xref — can re-slice to the authoritative
/// length.
#[derive(Debug, Clone, Copy)]
pub(crate) struct IndirectStreamLength {
    /// The `/Length M G R` holder object reference.
    pub holder: ObjectRef,
    /// Byte offset of the first stream payload byte (just past the `stream`
    /// keyword's EOL), relative to the `parse_indirect_object` input slice.
    pub data_start: usize,
    /// Byte offset of the `endstream` keyword (the syntactic upper bound; the
    /// authoritative length is clamped to not exceed this), when a line-anchored
    /// `endstream` was located. `None` when no line-anchored `endstream` exists
    /// — the only writer path that produces this is
    /// [`NewlineBeforeEndstream::Never`](crate::NewlineBeforeEndstream::Never)
    /// with a payload whose last byte is not an EOL, so `endstream` sits
    /// immediately after the final content byte and is not line-anchored. In
    /// that case the holder is the EXACT content length and the reader is
    /// authoritative (there is no syntactic bound to clamp against).
    pub endstream_pos: Option<usize>,
}

pub(crate) struct ParsedIndirectObject {
    pub(crate) object_ref: ObjectRef,
    pub(crate) object: Object,
    pub(crate) indirect_length: Option<IndirectStreamLength>,
    pub(crate) empty_offset: Option<usize>,
    pub(crate) expected_endobj_offset: Option<usize>,
}

pub(crate) fn parse_indirect_object(input: &[u8]) -> Result<(ObjectRef, Object)> {
    let (object_ref, object, _) = parse_indirect_object_detailed(input)?;
    Ok((object_ref, object))
}

/// Like [`parse_indirect_object`] but also returns
/// [`IndirectStreamLength`] bounds when the parsed object is a stream whose
/// `/Length` is an indirect reference. Used by the reader to resolve the
/// authoritative length via the xref; all other callers use
/// the plain [`parse_indirect_object`] wrapper.
pub(crate) fn parse_indirect_object_detailed(
    input: &[u8],
) -> Result<(ObjectRef, Object, Option<IndirectStreamLength>)> {
    let parsed = parse_indirect_object_detailed_impl(input, false)?;
    Ok((parsed.object_ref, parsed.object, parsed.indirect_length))
}

/// qpdf's JSON inspection resolves an empty indirect object as null while
/// recording a warning. Keep that recovery opt-in so ordinary PDF reads stay
/// strict. The parsed result records both the byte offset of the `endobj` token
/// when empty-object recovery was used and the byte offset where qpdf expected
/// `endobj`.
pub(crate) fn parse_indirect_object_detailed_qpdf(input: &[u8]) -> Result<ParsedIndirectObject> {
    parse_indirect_object_detailed_impl(input, true)
}

fn parse_indirect_object_detailed_impl(
    input: &[u8],
    allow_empty_object: bool,
) -> Result<ParsedIndirectObject> {
    let mut parser = Parser::new(input);
    let number = parser.integer_for_indirect()?;
    let generation = parser.integer_for_indirect()?;
    parser.expect_keyword_for_indirect(b"obj")?;
    parser.skip_ws();
    let empty_object_offset = if allow_empty_object && parser.take_keyword_token(b"endobj") {
        Some(parser.pos - b"endobj".len())
    } else {
        None
    };
    let object = if empty_object_offset.is_some() {
        Object::Null
    } else {
        parser.top_level_no_reference = allow_empty_object;
        parser.object()?
    };
    let expected_endobj_offset = if allow_empty_object && empty_object_offset.is_none() {
        parser.skip_ws();
        (!parser.take_keyword_token(b"endobj")).then_some(parser.pos)
    } else {
        None
    };
    Ok(ParsedIndirectObject {
        object_ref: ObjectRef::new(
            u32::try_from(number).map_err(|_| Error::parse(0, "invalid indirect object number"))?,
            u16::try_from(generation)
                .map_err(|_| Error::parse(0, "invalid indirect generation"))?,
        ),
        object,
        indirect_length: parser.last_indirect_stream_len,
        empty_offset: empty_object_offset,
        expected_endobj_offset,
    })
}

pub(crate) struct Parser<'a> {
    input: &'a [u8],
    pos: usize,
    /// When `true`, `N G R` is *not* recognised as an indirect reference;
    /// the first integer is returned and `G R` are left unconsumed. Content
    /// streams never contain indirect references, so the tokenizer sets this
    /// to avoid mis-parsing operands like `0 0 1 R` (rg/RG colour ops).
    no_reference: bool,
    /// qpdf treats an indirect reference in the body of an indirect object as
    /// a malformed direct object: it returns the first integer and warns that
    /// `endobj` was expected at the generation number. References nested in an
    /// array, dictionary, or stream dictionary remain valid.
    top_level_no_reference: bool,
    /// Set by [`stream_from_dict`](Self::stream_from_dict) when a stream's
    /// `/Length` is an indirect reference, so [`parse_indirect_object_detailed`]
    /// can surface the payload window for xref-based resolution.
    last_indirect_stream_len: Option<IndirectStreamLength>,
    /// Current object-nesting recursion depth, maintained by [`object`](Self::object)
    /// to bound recursion against adversarially deep input.
    depth: usize,
}

// Maximum object-nesting depth the recursive-descent parser will accept before
// returning an error. Without this bound, deeply nested input (`[[[[…` or
// `<</A <</A …`) recurses until the stack overflows and the process aborts —
// the qpdf CVE-2018-9918 class of denial of service. 500 matches the region of
// qpdf's `parser_max_nesting` (default 499); real documents never nest this
// deep, so only adversarial input is rejected.
const MAX_PARSE_DEPTH: usize = 500;

impl<'a> Parser<'a> {
    pub(crate) fn new(input: &'a [u8]) -> Self {
        Self {
            input,
            pos: 0,
            no_reference: false,
            top_level_no_reference: false,
            last_indirect_stream_len: None,
            depth: 0,
        }
    }

    /// Like [`new`](Self::new) but with indirect-reference recognition
    /// disabled (see [`Parser::no_reference`]).
    pub(crate) fn new_no_reference(input: &'a [u8]) -> Self {
        Self {
            input,
            pos: 0,
            no_reference: true,
            top_level_no_reference: false,
            last_indirect_stream_len: None,
            depth: 0,
        }
    }

    pub(crate) fn position(&self) -> usize {
        self.pos
    }

    /// Parse a single direct object at the current position (after leading
    /// whitespace/comments). Re-exported for the content-stream tokenizer so it
    /// can reuse the operand lexer without duplicating it.
    pub(crate) fn parse_one_object(&mut self) -> Result<Object> {
        self.object()
    }

    pub(crate) fn object(&mut self) -> Result<Object> {
        // `object` is the sole recursion hub: `dictionary` values and `array`
        // elements recurse only through it, and leaf parsers do not recurse.
        // A symmetric increment/decrement here therefore bounds every nesting
        // path. Decrementing on the error early-return AND on the normal return
        // keeps `depth` balanced across both, so repeated `parse_one_object`
        // calls from the content-stream tokenizer (which reuse one parser) do
        // not accumulate depth.
        self.depth += 1;
        if self.depth > MAX_PARSE_DEPTH {
            self.depth -= 1;
            return Err(Error::parse(self.pos, "object nesting too deep"));
        }
        let result = self.object_inner();
        self.depth -= 1;
        result
    }

    fn object_inner(&mut self) -> Result<Object> {
        self.skip_ws();
        if self.starts_with(b"<<") {
            return self.dictionary();
        }
        match self.peek() {
            Some(b'[') => self.array(),
            Some(b'/') => self.name().map(Object::Name),
            Some(b'(') => self.literal_string().map(Object::String),
            Some(b'<') => self.hex_string().map(Object::String),
            Some(b'.') => self.real(),
            Some(b'-' | b'+')
                if self
                    .input
                    .get(self.pos + 1)
                    .is_some_and(|byte| *byte == b'.') =>
            {
                self.real()
            }
            Some(b't') if self.take_keyword(b"true") => Ok(Object::Boolean(true)),
            Some(b'f') if self.take_keyword(b"false") => Ok(Object::Boolean(false)),
            Some(b'n') if self.take_keyword(b"null") => Ok(Object::Null),
            Some(b'-' | b'+' | b'0'..=b'9') => self.number_or_ref(),
            _ => Err(Error::parse(self.pos, "expected PDF object")),
        }
    }

    fn dictionary(&mut self) -> Result<Object> {
        self.expect_bytes(b"<<")?;
        let mut dict = Dictionary::new();
        loop {
            self.skip_ws();
            if self.starts_with(b">>") {
                self.pos += 2;
                self.skip_ws();
                if self.starts_with(b"stream") {
                    return self.stream_from_dict(dict);
                }
                return Ok(Object::Dictionary(dict));
            }
            let key = self.name()?;
            let value = self.object()?;
            dict.insert(key, value);
        }
    }

    fn stream_from_dict(&mut self, dict: Dictionary) -> Result<Object> {
        // A usable DIRECT /Length is a non-negative Integer; everything else
        // — an indirect `M G R` reference (the form flpdf's own QDF writer now
        // emits; a common real-world shape, valid per ISO 32000-1 §7.3.8.2), an
        // ABSENT /Length, or any otherwise-unusable value — falls through to
        // the spec-sanctioned `endstream`-scan recovery path. The DIRECT fast
        // path must stay byte-identical so no currently-parsing PDF regresses.
        let direct_length = match dict.get("Length") {
            Some(Object::Integer(value)) if *value >= 0 => u64::try_from(*value).ok(),
            _ => None,
        };
        // An indirect `/Length M G R`: the parser cannot resolve M, but the
        // reader can (flpdf-9hc.27). Record the holder so the recovery branch
        // below can surface the payload window for xref-based resolution.
        let length_ref = match dict.get("Length") {
            Some(Object::Reference(r)) => Some(*r),
            _ => None,
        };

        self.expect_keyword_for_indirect(b"stream")?;
        if self.peek() == Some(b'\r') {
            self.pos += 1;
            if self.peek() == Some(b'\n') {
                self.pos += 1;
            }
        } else if self.peek() == Some(b'\n') {
            self.pos += 1;
        }

        let data_start = self.pos;
        let length = match direct_length.and_then(|n| usize::try_from(n).ok()) {
            // `checked_add` so a malformed huge direct /Length cannot overflow
            // the bounds check (debug panic / release wrap-then-OOB-slice on
            // untrusted input); on overflow fall through to endstream-scan.
            Some(length)
                if data_start
                    .checked_add(length)
                    .is_some_and(|end| end <= self.input.len()) =>
            {
                length
            }
            // Indirect / missing / invalid / out-of-range /Length: recover the
            // payload boundary by locating the line-anchored `endstream`
            // keyword (what qpdf and conformant readers do). The indirect
            // holder value is advisory; `endstream` is authoritative.
            _ => match find_line_anchored_keyword(self.input, b"endstream", data_start) {
                Some(endstream_pos) => {
                    // For an indirect /Length, surface the payload window so the
                    // reader can re-slice to the xref-resolved authoritative
                    // length. The endstream-scan result here is only the
                    // last-resort fallback when the holder is unresolvable.
                    if let Some(holder) = length_ref {
                        self.last_indirect_stream_len = Some(IndirectStreamLength {
                            holder,
                            data_start,
                            endstream_pos: Some(endstream_pos),
                        });
                    }
                    // Exclude exactly ONE framing EOL marker that the writer
                    // placed between the payload and `endstream`, so
                    // `stream.data` is the logical content. The writer then
                    // re-adds exactly one EOL, keeping QDF round-trip /
                    // idempotence byte-stable.
                    let mut end = endstream_pos;
                    if end > data_start && self.input[end - 1] == b'\n' {
                        end -= 1;
                        if end > data_start && self.input[end - 1] == b'\r' {
                            end -= 1;
                        }
                    } else if end > data_start && self.input[end - 1] == b'\r' {
                        end -= 1;
                    }
                    end - data_start
                }
                // No line-anchored `endstream` anywhere from `data_start`. The
                // only writer path that produces this is
                // `NewlineBeforeEndstream::Never` with a non-EOL-ending payload:
                // `endstream` follows the last CONTENT byte directly, so it is
                // not line-anchored and the byte-level scan cannot delimit the
                // payload. (A naive non-anchored scan is wrong — the literal
                // bytes "endstream" can occur inside the payload.)
                None => match length_ref {
                    // Indirect /Length AND a non-anchored `endstream` token
                    // exists: this is the adjacent-`endstream` case. The holder
                    // — which the reader resolves via the xref — is the EXACT
                    // content length. Surface it with `endstream_pos: None` to
                    // flag the authoritative path; `data` is an empty placeholder
                    // the reader replaces. The presence gate keeps a genuinely
                    // truncated stream (no `endstream` at all) on the error path
                    // below; the reader, not this gate, fixes the boundary.
                    Some(holder)
                        if contains_keyword_token(self.input, b"endstream", data_start) =>
                    {
                        self.last_indirect_stream_len = Some(IndirectStreamLength {
                            holder,
                            data_start,
                            endstream_pos: None,
                        });
                        return Ok(Object::Stream(Stream::new(dict, Vec::new())));
                    }
                    // No `endstream` anywhere (truncated), or no holder to
                    // recover the boundary from: fail loudly. ISO 32000-1
                    // §7.3.8.1 mandates the EOL before `endstream` precisely so a
                    // direct/absent-length stream stays parseable; its absence
                    // here is unrecoverable.
                    _ => return Err(Error::parse(self.pos, "stream data exceeds input")),
                },
            },
        };

        let data = self.input[data_start..data_start + length].to_vec();
        self.pos = data_start + length;

        self.skip_ws();
        self.expect_keyword_for_indirect(b"endstream")?;

        Ok(Object::Stream(Stream::new(dict, data)))
    }

    fn hex_string(&mut self) -> Result<Vec<u8>> {
        self.expect_byte(b'<')?;
        let mut value = Vec::new();
        let mut first_nibble: Option<u8> = None;

        while let Some(byte) = self.peek() {
            if byte == b'>' {
                self.pos += 1;
                if first_nibble.is_some() {
                    return Err(Error::parse(self.pos, "hex string has odd length"));
                }
                return Ok(value);
            }

            let nibble = match hex_value(byte) {
                Some(byte) => byte,
                None => return Err(Error::parse(self.pos, "invalid hex string")),
            };

            self.pos += 1;

            if let Some(high) = first_nibble {
                value.push((high << 4) | nibble);
                first_nibble = None;
            } else {
                first_nibble = Some(nibble);
            }
        }

        Err(Error::parse(self.pos, "unterminated hex string"))
    }

    fn array(&mut self) -> Result<Object> {
        self.expect_byte(b'[')?;
        let mut values = Vec::new();
        loop {
            self.skip_ws();
            if self.peek() == Some(b']') {
                self.pos += 1;
                return Ok(Object::Array(values));
            }
            values.push(self.object()?);
        }
    }

    fn number_or_ref(&mut self) -> Result<Object> {
        let start = self.pos;
        let first = self.integer()?;
        let saved = self.pos;
        if self.peek() == Some(b'.') {
            return self.real_with_integer_prefix(start);
        }
        if matches!(self.peek(), Some(b'e' | b'E')) {
            return self.parse_real_exponent(start);
        }
        if self.no_reference || (self.top_level_no_reference && self.depth == 1) {
            self.pos = saved;
            return Ok(Object::Integer(first));
        }
        self.skip_ws();
        if let Ok(second) = self.integer() {
            self.skip_ws();
            if self.peek() == Some(b'R') {
                self.pos += 1;
                let number = u32::try_from(first)
                    .map_err(|_| Error::parse(saved, "invalid object number"))?;
                let generation = u16::try_from(second)
                    .map_err(|_| Error::parse(saved, "invalid generation number"))?;
                return Ok(Object::Reference(ObjectRef::new(number, generation)));
            }
        }
        self.pos = saved;
        Ok(Object::Integer(first))
    }

    fn real(&mut self) -> Result<Object> {
        let start = self.pos;
        if matches!(self.peek(), Some(b'+' | b'-')) {
            self.pos += 1;
        }
        if self.peek() != Some(b'.') {
            return Err(Error::parse(start, "expected real number"));
        }
        self.pos += 1;
        while matches!(self.peek(), Some(b'0'..=b'9')) {
            self.pos += 1;
        }
        self.parse_real_exponent(start)
    }

    fn real_with_integer_prefix(&mut self, start: usize) -> Result<Object> {
        if self.peek() != Some(b'.') {
            return Err(Error::parse(start, "expected decimal point"));
        }
        self.pos += 1;
        while matches!(self.peek(), Some(b'0'..=b'9')) {
            self.pos += 1;
        }
        self.parse_real_exponent(start)
    }

    fn parse_real_exponent(&mut self, start: usize) -> Result<Object> {
        if matches!(self.peek(), Some(b'e' | b'E')) {
            self.pos += 1;
            if matches!(self.peek(), Some(b'+' | b'-')) {
                self.pos += 1;
            }
            let exponent_start = self.pos;
            while matches!(self.peek(), Some(b'0'..=b'9')) {
                self.pos += 1;
            }
            if exponent_start == self.pos {
                return Err(Error::parse(start, "invalid real"));
            }
        }

        let bytes = &self.input[start..self.pos];
        let text =
            std::str::from_utf8(bytes).map_err(|_| Error::parse(start, "real is not utf-8"))?;
        let value = text
            .parse::<f64>()
            .map_err(|_| Error::parse(start, "invalid real"))?;
        // Preserve the source literal when `value.to_string()` cannot
        // reproduce it byte-for-byte (e.g. `.4`, `0.400`, `1.0`) — required
        // for byte-identical parity with qpdf's QPDF_Real (which re-emits the
        // parsed string verbatim). When the literal already matches Rust's
        // shortest round-trip, the plain `Real(f64)` is smaller and equivalent.
        if value.to_string().as_bytes() == bytes {
            Ok(Object::Real(value))
        } else {
            Ok(Object::RealLiteral {
                value,
                literal: bytes.to_vec(),
            })
        }
    }

    fn integer(&mut self) -> Result<i64> {
        self.skip_ws();
        let start = self.pos;
        if matches!(self.peek(), Some(b'-' | b'+')) {
            self.pos += 1;
        }
        while matches!(self.peek(), Some(b'0'..=b'9')) {
            self.pos += 1;
        }
        if start == self.pos
            || matches!(self.input.get(start), Some(b'-' | b'+')) && start + 1 == self.pos
        {
            return Err(Error::parse(start, "expected integer"));
        }
        let text = std::str::from_utf8(&self.input[start..self.pos])
            .map_err(|_| Error::parse(start, "integer is not utf-8"))?;
        text.parse::<i64>()
            .map_err(|_| Error::parse(start, "invalid integer"))
    }

    pub(crate) fn integer_for_indirect(&mut self) -> Result<i64> {
        self.integer()
    }

    pub(crate) fn expect_keyword_for_indirect(&mut self, keyword: &[u8]) -> Result<()> {
        self.skip_ws();
        if self.starts_with(keyword) {
            self.pos += keyword.len();
            Ok(())
        } else {
            Err(Error::parse(self.pos, "expected indirect object keyword"))
        }
    }

    fn name(&mut self) -> Result<Vec<u8>> {
        self.expect_byte(b'/')?;
        let mut value = Vec::new();
        while let Some(byte) = self.peek() {
            if is_delimiter(byte) || is_ws(byte) {
                break;
            }
            if byte == b'#' {
                let pos = self.pos;
                self.pos += 1;
                let high = self
                    .peek()
                    .and_then(hex_value)
                    .ok_or_else(|| Error::parse(pos, "invalid name escape"))?;
                self.pos += 1;
                let low = self
                    .peek()
                    .and_then(hex_value)
                    .ok_or_else(|| Error::parse(pos, "invalid name escape"))?;
                self.pos += 1;
                value.push((high << 4) | low);
            } else {
                value.push(byte);
                self.pos += 1;
            }
        }
        if value.is_empty() {
            return Err(Error::parse(self.pos, "empty name"));
        }
        Ok(value)
    }

    fn literal_string(&mut self) -> Result<Vec<u8>> {
        self.expect_byte(b'(')?;
        let mut value = Vec::new();
        let mut depth = 0;
        while let Some(byte) = self.peek() {
            if byte == b'(' {
                depth += 1;
                value.push(byte);
                self.pos += 1;
                continue;
            }

            if byte == b')' {
                if depth == 0 {
                    self.pos += 1;
                    return Ok(value);
                }
                depth -= 1;
                value.push(byte);
                self.pos += 1;
                continue;
            }

            if byte == b'\\' {
                self.pos += 1;
                match self.peek() {
                    Some(b'n') => {
                        value.push(b'\n');
                        self.pos += 1;
                    }
                    Some(b'r') => {
                        value.push(b'\r');
                        self.pos += 1;
                    }
                    Some(b't') => {
                        value.push(b'\t');
                        self.pos += 1;
                    }
                    Some(b'b') => {
                        value.push(0x08);
                        self.pos += 1;
                    }
                    Some(b'f') => {
                        value.push(0x0c);
                        self.pos += 1;
                    }
                    Some(b'(' | b')' | b'\\') => {
                        value.push(self.peek().unwrap_or_default());
                        self.pos += 1;
                    }
                    Some(b'\r') => {
                        self.pos += 1;
                        if self.peek() == Some(b'\n') {
                            self.pos += 1;
                        }
                    }
                    Some(b'\n') => {
                        self.pos += 1;
                    }
                    Some(byte @ b'0'..=b'7') => {
                        let first = byte;
                        let mut value_byte = first - b'0';
                        self.pos += 1;
                        for _ in 0..2 {
                            if let Some(next) = self.peek() {
                                if matches!(next, b'0'..=b'7') {
                                    value_byte = (value_byte << 3) | (next - b'0');
                                    self.pos += 1;
                                } else {
                                    break;
                                }
                            }
                        }
                        value.push(value_byte);
                    }
                    Some(_) => {
                        value.push(self.peek().unwrap_or_default());
                        self.pos += 1;
                    }
                    None => return Err(Error::parse(self.pos, "unterminated literal string")),
                }
                continue;
            }

            self.pos += 1;
            value.push(byte);
        }

        Err(Error::parse(self.pos, "unterminated literal string"))
    }

    pub(crate) fn skip_ws(&mut self) {
        loop {
            while matches!(self.peek(), Some(byte) if is_ws(byte)) {
                self.pos += 1;
            }
            if self.peek() == Some(b'%') {
                while !matches!(self.peek(), None | Some(b'\n' | b'\r')) {
                    self.pos += 1;
                }
                continue;
            }
            break;
        }
    }

    fn take_keyword(&mut self, keyword: &[u8]) -> bool {
        if self.starts_with(keyword) {
            self.pos += keyword.len();
            true
        } else {
            false
        }
    }

    fn take_keyword_token(&mut self, keyword: &[u8]) -> bool {
        if !self.starts_with(keyword) {
            return false;
        }
        let following = self.input.get(self.pos + keyword.len()).copied();
        if following.is_some_and(|byte| !is_ws(byte) && !is_delimiter(byte)) {
            return false;
        }
        self.pos += keyword.len();
        true
    }

    fn expect_byte(&mut self, expected: u8) -> Result<()> {
        if self.peek() == Some(expected) {
            self.pos += 1;
            Ok(())
        } else {
            Err(Error::parse(self.pos, format!("expected byte {expected}")))
        }
    }

    fn expect_bytes(&mut self, expected: &[u8]) -> Result<()> {
        if self.starts_with(expected) {
            self.pos += expected.len();
            Ok(())
        } else {
            Err(Error::parse(self.pos, "expected token"))
        }
    }

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

    fn peek(&self) -> Option<u8> {
        self.input.get(self.pos).copied()
    }
}

/// Locate the FIRST line-anchored `keyword` at or after `from`.
///
/// A match counts only when the keyword starts a line (preceding byte is a
/// `\n`/`\r` EOL or it is the start of the buffer) AND is followed by
/// EOF/whitespace/EOL/delimiter. This mirrors the line-anchored finder used by
/// the QDF fix layer so we never match an `endstream` byte sequence that is
/// merely incidental binary content rather than the stream terminator. Kept
/// local to the parser so it carries no dependency on writer-side modules.
fn find_line_anchored_keyword(input: &[u8], keyword: &[u8], from: usize) -> Option<usize> {
    let mut i = from;
    while i + keyword.len() <= input.len() {
        if &input[i..i + keyword.len()] == keyword {
            let at_line_start = i == 0 || input[i - 1] == b'\n' || input[i - 1] == b'\r';
            let after_ok = match input.get(i + keyword.len()) {
                None => true,
                Some(&c) => is_ws(c) || is_delimiter(c),
            };
            if at_line_start && after_ok {
                return Some(i);
            }
        }
        i += 1;
    }
    None
}

/// True when `keyword` appears anywhere in `input` at or after `from`, followed
/// by EOF/whitespace/delimiter (a keyword-token boundary), WITHOUT requiring a
/// line start.
///
/// Used ONLY as a presence gate to tell the adjacent-`endstream` case
/// (`NewlineBeforeEndstream::Never`, non-EOL-ending payload — a non-anchored
/// `endstream` exists, so the reader can resolve the indirect holder
/// authoritatively) apart from a genuinely truncated stream (no `endstream` at
/// all → unrecoverable). It never determines a payload boundary, so an
/// `endstream` byte sequence that is merely incidental binary content cannot
/// mis-slice the data.
fn contains_keyword_token(input: &[u8], keyword: &[u8], from: usize) -> bool {
    let mut i = from;
    while i + keyword.len() <= input.len() {
        if &input[i..i + keyword.len()] == keyword {
            let after_ok = match input.get(i + keyword.len()) {
                None => true,
                Some(&c) => is_ws(c) || is_delimiter(c),
            };
            if after_ok {
                return true;
            }
        }
        i += 1;
    }
    false
}

pub(crate) fn is_ws(byte: u8) -> bool {
    matches!(byte, b'\0' | b'\t' | b'\n' | b'\x0c' | b'\r' | b' ')
}

pub(crate) fn is_delimiter(byte: u8) -> bool {
    matches!(
        byte,
        b'(' | b')' | b'<' | b'>' | b'[' | b']' | b'{' | b'}' | b'/' | b'%'
    )
}

fn hex_value(byte: u8) -> Option<u8> {
    match byte {
        b'0'..=b'9' => Some(byte - b'0'),
        b'a'..=b'f' => Some(byte - b'a' + 10),
        b'A'..=b'F' => Some(byte - b'A' + 10),
        _ => None,
    }
}

#[cfg(test)]
mod stream_length_tests {
    use super::{
        parse_indirect_object, parse_indirect_object_detailed, parse_indirect_object_detailed_qpdf,
    };
    use crate::{Object, ObjectRef};

    fn parse_stream(bytes: &[u8]) -> crate::Stream {
        let (_, object) = parse_indirect_object(bytes).expect("indirect object must parse");
        match object {
            Object::Stream(stream) => stream,
            other => panic!("expected a stream, got {other:?}"),
        }
    }

    // Indirect `/Length M 0 R`: the holder object is never available to the
    // byte-level parser, so the data boundary must come from the `endstream`
    // scan. flpdf-m41.
    #[test]
    fn indirect_length_resolves_via_endstream_scan() {
        let payload = b"Hello indirect length world.";
        let mut bytes = Vec::new();
        bytes.extend_from_slice(b"3 0 obj\n<< /Length 7 0 R >>\nstream\n");
        bytes.extend_from_slice(payload);
        bytes.extend_from_slice(b"\nendstream\nendobj\n");

        let stream = parse_stream(&bytes);
        assert_eq!(
            stream.data.as_slice(),
            payload,
            "indirect /Length stream data must come from the endstream boundary"
        );
    }

    // Even when an integer is reachable through the reference notation, the
    // parser must NOT trust it (the holder body is never the value here) — the
    // `endstream` keyword is authoritative. A deliberately wrong-looking holder
    // ref still yields the correct payload.
    #[test]
    fn stale_holder_value_does_not_corrupt_data() {
        // /Length references object 99 (a holder the parser never sees). The
        // real payload is 11 bytes; any holder integer is irrelevant.
        let payload = b"ABCDEFGHIJK";
        let mut bytes = Vec::new();
        bytes.extend_from_slice(b"5 0 obj\n<< /Length 99 0 R >>\nstream\n");
        bytes.extend_from_slice(payload);
        bytes.extend_from_slice(b"\nendstream\nendobj\n");

        let stream = parse_stream(&bytes);
        assert_eq!(
            stream.data.as_slice(),
            payload,
            "endstream must override any (stale) indirect holder value"
        );
    }

    // Binary payload containing the literal bytes `endstream` mid-content but
    // NOT line-anchored must not terminate early.
    #[test]
    fn non_line_anchored_endstream_in_payload_is_ignored() {
        let payload = b"xx endstream yy\x00\x01rest";
        let mut bytes = Vec::new();
        bytes.extend_from_slice(b"4 0 obj\n<< /Length 8 0 R >>\nstream\n");
        bytes.extend_from_slice(payload);
        bytes.extend_from_slice(b"\nendstream\nendobj\n");

        let stream = parse_stream(&bytes);
        assert_eq!(
            stream.data.as_slice(),
            payload,
            "an `endstream` substring not at a line start must not terminate the stream"
        );
    }

    // Regression: a normal DIRECT integer /Length must still take the
    // byte-identical fast path and slice exactly `Length` bytes.
    #[test]
    fn direct_integer_length_unchanged() {
        let payload = b"direct-length-bytes";
        let mut bytes = Vec::new();
        bytes.extend_from_slice(
            format!("2 0 obj\n<< /Length {} >>\nstream\n", payload.len()).as_bytes(),
        );
        bytes.extend_from_slice(payload);
        bytes.extend_from_slice(b"\nendstream\nendobj\n");

        let stream = parse_stream(&bytes);
        assert_eq!(
            stream.data.as_slice(),
            payload,
            "direct integer /Length must keep slicing exactly Length bytes"
        );
    }

    // No `endstream` keyword at all → the existing parse error, no hang/panic.
    #[test]
    fn missing_endstream_is_an_error() {
        let mut bytes = Vec::new();
        bytes.extend_from_slice(b"6 0 obj\n<< /Length 1 0 R >>\nstream\n");
        bytes.extend_from_slice(b"payload with no terminator");
        // intentionally no `endstream`
        assert!(
            parse_indirect_object(&bytes).is_err(),
            "an indirect-length stream with no endstream must error, not hang"
        );
    }

    #[test]
    fn contains_keyword_token_boundaries() {
        use super::contains_keyword_token;
        // Token at EOF (nothing after the keyword) counts.
        assert!(contains_keyword_token(b"xxendstream", b"endstream", 0));
        // Token followed by whitespace counts.
        assert!(contains_keyword_token(b"endstream\n", b"endstream", 0));
        // No boundary after the keyword (a longer run of regular chars) does not.
        assert!(!contains_keyword_token(b"endstreamX", b"endstream", 0));
        // Absent keyword does not.
        assert!(!contains_keyword_token(b"no keyword here", b"endstream", 0));
    }

    #[test]
    fn empty_indirect_object_recovery_is_qpdf_only_and_token_bounded() {
        let empty = b"7 0 obj\n  endobj\n";
        assert!(parse_indirect_object_detailed(empty).is_err());

        let parsed = parse_indirect_object_detailed_qpdf(empty).expect("qpdf empty recovery");
        assert_eq!(parsed.object_ref, ObjectRef::new(7, 0));
        assert_eq!(parsed.object, Object::Null);
        assert!(parsed.indirect_length.is_none());
        assert_eq!(parsed.empty_offset, Some(10));
        assert_eq!(parsed.expected_endobj_offset, None);

        assert!(parse_indirect_object_detailed_qpdf(b"7 0 obj\nendobject\nendobj\n").is_err());
    }
}