nosj 0.2.2

Pull, push and fused-cursor JSON parsers plus a streaming writer, for hosts that build native values during the parse — SIMD-accelerated (NEON, SSE2, runtime-detected AVX2)
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
//! The public pull API: a grammar-enforcing cursor over the stage-1 index.

use crate::cursor::ParseOptions;
use crate::scalars::{self, Literal, Number, ScalarError, StrPart};
use crate::stage1;

/// Reusable allocations. Keep one per thread and hand it to each [`Reader`]
/// or [`crate::parse`] call to amortize allocations across parses.
#[derive(Default)]
pub struct Buffers {
    indexes: Vec<u32>,
    pub(crate) scratch: Vec<u8>,
    frames: Vec<crate::driver::Frame>,
    /// Length of the input the index was last built for: a best-effort
    /// staleness check for the preindexed drive path (the `# Safety`
    /// contract remains authoritative).
    indexed_len: usize,
}

impl std::fmt::Debug for Buffers {
    /// Capacities only; the contents are scratch state, not data.
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("Buffers")
            .field("index_capacity", &self.indexes.capacity())
            .field("scratch_capacity", &self.scratch.capacity())
            .field("frame_capacity", &self.frames.capacity())
            .finish_non_exhaustive()
    }
}

impl Buffers {
    /// Empty buffers; allocations grow on first use and are reused after.
    #[must_use]
    pub fn new() -> Self {
        Self::default()
    }

    /// Run stage 1 (structural indexing) only: pure computation, no
    /// callbacks, no host-runtime interaction. Callers embedded in runtimes
    /// with a global lock (a GVL/GIL) can run this phase outside the lock,
    /// then finish with [`crate::parse_preindexed_utf8_unchecked`].
    pub fn preindex(&mut self, input: &[u8]) {
        crate::stage1::index(input, &mut self.indexes);
        self.indexed_len = input.len();
    }

    /// Best-effort guard for [`crate::parse_preindexed_utf8_unchecked`]:
    /// the last-indexed input length must match.
    pub(crate) fn assert_preindexed(&self, input: &[u8]) {
        debug_assert!(
            self.indexed_len == input.len(),
            "drive_preindexed input does not match the preindexed input \
             ({} vs {} bytes)",
            input.len(),
            self.indexed_len,
        );
    }

    pub(crate) fn split_for_driver(
        &mut self,
    ) -> (&mut Vec<u32>, &mut Vec<u8>, &mut Vec<crate::driver::Frame>) {
        (&mut self.indexes, &mut self.scratch, &mut self.frames)
    }
}

/// A single value-position token.
#[derive(Debug, Clone, Copy, PartialEq)]
pub enum Node<'a> {
    /// `{`; iterate with [`Reader::object_first_key`] / [`Reader::object_next_key`].
    ObjectStart,
    /// `[`; iterate with [`Reader::array_first`] / [`Reader::array_next`].
    ArrayStart,
    /// String value, decoded. Borrows the parser; consume before advancing.
    Str(&'a str),
    /// Integer representable in `i64`.
    Int(i64),
    /// Floating-point number, exactly rounded.
    Float(f64),
    /// Integer too large for i64, as its ASCII digits.
    BigInt(&'a str),
    /// `true` or `false`.
    Bool(bool),
    /// `null`.
    Null,
}

/// What went wrong, without position information (see [`ParseError`]).
#[derive(Debug, Clone, PartialEq, Eq)]
#[non_exhaustive]
pub enum ErrorKind {
    /// A byte that cannot start any JSON token.
    UnexpectedCharacter,
    /// Input ended inside a value or container.
    UnexpectedEnd,
    /// Bytes remained after the top-level value.
    TrailingCharacters,
    /// The grammar required the described token here.
    Expected(&'static str),
    /// A string, number, or literal failed to parse.
    Scalar(ScalarError),
    /// A non-empty JSON Pointer that does not start with `/` (see
    /// [`crate::pointer`]).
    InvalidPointer,
}

impl std::fmt::Display for ErrorKind {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        match self {
            Self::UnexpectedCharacter => f.write_str("unexpected character"),
            Self::UnexpectedEnd => f.write_str("unexpected end of input"),
            Self::TrailingCharacters => f.write_str("trailing characters after the document"),
            Self::Expected(what) => write!(f, "expected {what}"),
            Self::Scalar(e) => write!(f, "{e}"),
            Self::InvalidPointer => f.write_str("JSON Pointer must be empty or start with '/'"),
        }
    }
}

/// A parse failure with the byte offset where it was detected.
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct ParseError {
    /// Byte offset into the input.
    pub offset: usize,
    /// What went wrong.
    pub kind: ErrorKind,
}

impl std::fmt::Display for ParseError {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        write!(f, "{} at byte {}", self.kind, self.offset)
    }
}

impl std::error::Error for ParseError {}

/// Crate-internal constructors: every parse path builds errors through
/// these (all `#[cold]`; error construction is never the hot path).
impl ParseError {
    #[cold]
    pub(crate) fn expected(what: &'static str, offset: usize) -> ParseError {
        ParseError {
            offset,
            kind: ErrorKind::Expected(what),
        }
    }

    #[cold]
    pub(crate) fn unexpected_end(offset: usize) -> ParseError {
        ParseError {
            offset,
            kind: ErrorKind::UnexpectedEnd,
        }
    }

    #[cold]
    pub(crate) fn unexpected_character(offset: usize) -> ParseError {
        ParseError {
            offset,
            kind: ErrorKind::UnexpectedCharacter,
        }
    }

    #[cold]
    pub(crate) fn trailing_characters(offset: usize) -> ParseError {
        ParseError {
            offset,
            kind: ErrorKind::TrailingCharacters,
        }
    }

    #[cold]
    pub(crate) fn scalar(kind: ScalarError, offset: usize) -> ParseError {
        ParseError {
            offset,
            kind: ErrorKind::Scalar(kind),
        }
    }
}

/// Pull cursor over one JSON document.
pub struct Reader<'j, 'b> {
    input: &'j [u8],
    bufs: &'b mut Buffers,
    pos: usize,
    opts: ParseOptions,
}

impl std::fmt::Debug for Reader<'_, '_> {
    /// Position only; dumping the document would drown the output.
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("Reader")
            .field("token_pos", &self.pos)
            .field("input_len", &self.input.len())
            .finish_non_exhaustive()
    }
}

impl<'j, 'b> Reader<'j, 'b> {
    /// Parse a `&str`; UTF-8 validity is guaranteed by the type.
    #[must_use]
    pub fn new(input: &'j str, bufs: &'b mut Buffers) -> Self {
        // SAFETY: &str is valid UTF-8.
        unsafe { Self::from_utf8_unchecked(input.as_bytes(), bufs) }
    }

    /// Parse raw bytes the caller guarantees are valid UTF-8 (e.g. vouched
    /// for by a host runtime's cached validity state).
    ///
    /// # Safety
    ///
    /// `input` must be valid UTF-8; string tokens are handed out as `&str`
    /// without re-validation.
    pub unsafe fn from_utf8_unchecked(input: &'j [u8], bufs: &'b mut Buffers) -> Self {
        // SAFETY: forwarded contract.
        unsafe { Self::from_utf8_unchecked_with(input, bufs, ParseOptions::default()) }
    }

    /// Like [`Reader::new`], with grammar extensions from [`ParseOptions`]:
    /// the same document set [`crate::parse_with`] accepts is walkable,
    /// and skippable, here.
    #[must_use]
    pub fn new_with(input: &'j str, bufs: &'b mut Buffers, opts: ParseOptions) -> Self {
        // SAFETY: &str is valid UTF-8.
        unsafe { Self::from_utf8_unchecked_with(input.as_bytes(), bufs, opts) }
    }

    /// Like [`Reader::from_utf8_unchecked`], with grammar extensions from
    /// [`ParseOptions`].
    ///
    /// # Safety
    ///
    /// `input` must be valid UTF-8; string tokens are handed out as `&str`
    /// without re-validation.
    pub unsafe fn from_utf8_unchecked_with(
        input: &'j [u8],
        bufs: &'b mut Buffers,
        opts: ParseOptions,
    ) -> Self {
        stage1::index(input, &mut bufs.indexes);
        Self {
            input,
            bufs,
            pos: 0,
            opts,
        }
    }

    /// The `allow_nan` keyword at `off` as a float node's value, or the
    /// error the strict grammar reports there.
    #[cold]
    fn nan_keyword(&self, off: usize, strict: ParseError) -> Result<(f64, usize), ParseError> {
        if self.opts.allow_nan
            && let Some(hit) = scalars::parse_nan_keyword(self.input, off)
        {
            return Ok(hit);
        }
        Err(strict)
    }

    #[inline(always)]
    fn peek_tok(&self) -> Option<(usize, u8)> {
        self.bufs.indexes.get(self.pos).map(|&off| {
            let off = off as usize;
            (off, self.input[off])
        })
    }

    #[inline(always)]
    fn take_tok(&mut self) -> Result<(usize, u8), ParseError> {
        match self.peek_tok() {
            Some(t) => {
                self.pos += 1;
                Ok(t)
            }
            None => Err(ParseError::unexpected_end(self.input.len())),
        }
    }

    /// Parse the next value-position token.
    ///
    /// Named `next_node` (not `next`) deliberately: the parser is not an
    /// [`Iterator`], because returned [`Node`]s borrow the parser and must
    /// be consumed before advancing.
    #[inline]
    pub fn next_node(&mut self) -> Result<Node<'_>, ParseError> {
        let (off, byte) = self.take_tok()?;
        match byte {
            b'{' => Ok(Node::ObjectStart),
            b'[' => Ok(Node::ArrayStart),
            b'"' => {
                // SAFETY: `self.input` came from `&str` (or the caller's
                // UTF-8 vouch), and `off` holds the `"` just tokenized.
                let part =
                    unsafe { scalars::parse_string(self.input, off, &mut self.bufs.scratch) }
                        .map_err(|e| ParseError::scalar(e, off))?;
                Ok(Node::Str(match part.0 {
                    StrPart::Borrowed(s) | StrPart::Decoded(s) => s,
                    // The pull API's Node::Str is &str-typed; lone-surrogate
                    // content (WTF-8) is rejected here. Byte-preserving
                    // consumers should use the Sink drivers.
                    StrPart::DecodedRaw(_) => {
                        return Err(ParseError::scalar(ScalarError::LoneSurrogate, off));
                    }
                }))
            }
            b'-' | b'0'..=b'9' => {
                let num = match scalars::parse_number(self.input, off) {
                    Ok((num, _)) => num,
                    // `-Infinity` fails the number grammar first.
                    Err(e) => {
                        let (f, _) = self.nan_keyword(off, ParseError::scalar(e, off))?;
                        return Ok(Node::Float(f));
                    }
                };
                Ok(match num {
                    Number::Int(i) => Node::Int(i),
                    // The pull API folds `-0` to integer 0; hosts that
                    // keep the IEEE sign use the push API and override
                    // `Sink::negative_zero`.
                    Number::NegativeZero => Node::Int(0),
                    Number::Float(f) => Node::Float(f),
                    Number::Big(digits) => Node::BigInt(digits),
                })
            }
            b't' | b'f' | b'n' => {
                let (lit, _) = scalars::parse_literal(self.input, off)
                    .map_err(|e| ParseError::scalar(e, off))?;
                Ok(match lit {
                    Literal::True => Node::Bool(true),
                    Literal::False => Node::Bool(false),
                    Literal::Null => Node::Null,
                })
            }
            _ => {
                let (f, _) = self.nan_keyword(off, ParseError::unexpected_character(off))?;
                Ok(Node::Float(f))
            }
        }
    }

    /// Decode the key string at `off` and consume the following `:`.
    #[inline(always)]
    fn read_key(&mut self, off: usize) -> Result<Option<&str>, ParseError> {
        // The colon is the next index token regardless of the key's
        // bytes; consuming it first lets the decoded key borrow the
        // scratch buffer for the whole return.
        match self.take_tok()? {
            (_, b':') => {}
            (o, _) => return Err(ParseError::expected("':'", o)),
        }
        // SAFETY: `self.input` came from `&str` (or the caller's UTF-8
        // vouch), and `off` holds the `"` just tokenized.
        let part = unsafe { scalars::parse_string(self.input, off, &mut self.bufs.scratch) }
            .map_err(|e| ParseError::scalar(e, off))?;
        Ok(Some(match part.0 {
            StrPart::Borrowed(s) | StrPart::Decoded(s) => s,
            // See next_node: the pull API rejects WTF-8 content.
            StrPart::DecodedRaw(_) => {
                return Err(ParseError::scalar(ScalarError::LoneSurrogate, off));
            }
        }))
    }

    /// After [`Node::ObjectStart`]: the first key, or `None` for `{}`.
    /// The `:` after the key is consumed; call [`Reader::next_node`] for the value.
    #[inline]
    pub fn object_first_key(&mut self) -> Result<Option<&str>, ParseError> {
        match self.take_tok()? {
            (_, b'}') => Ok(None),
            (off, b'"') => self.read_key(off),
            (o, _) => Err(ParseError::expected("'\"' or '}'", o)),
        }
    }

    /// After an object value: the next key, or `None` at `}`.
    #[inline]
    pub fn object_next_key(&mut self) -> Result<Option<&str>, ParseError> {
        match self.take_tok()? {
            (_, b'}') => Ok(None),
            (_, b',') => match self.take_tok()? {
                (off, b'"') => self.read_key(off),
                (_, b'}') if self.opts.allow_trailing_comma => Ok(None),
                (o, _) => Err(ParseError::expected("'\"'", o)),
            },
            (o, _) => Err(ParseError::expected("',' or '}'", o)),
        }
    }

    /// After [`Node::ArrayStart`]: `true` if the array has a first element
    /// (call [`Reader::next_node`] to get it), `false` for `[]`.
    #[inline]
    pub fn array_first(&mut self) -> Result<bool, ParseError> {
        match self.peek_tok() {
            Some((_, b']')) => {
                self.pos += 1;
                Ok(false)
            }
            Some(_) => Ok(true),
            None => Err(ParseError::unexpected_end(self.input.len())),
        }
    }

    /// After an array element: `true` if another element follows.
    #[inline]
    pub fn array_next(&mut self) -> Result<bool, ParseError> {
        match self.take_tok()? {
            (_, b']') => Ok(false),
            (_, b',') => {
                if self.opts.allow_trailing_comma && matches!(self.peek_tok(), Some((_, b']'))) {
                    self.pos += 1;
                    return Ok(false);
                }
                Ok(true)
            }
            (o, _) => Err(ParseError::expected("',' or ']'", o)),
        }
    }

    /// Consume the value at value position without parsing it, returning
    /// its raw text. This is the enabler for partial parsing. Call it
    /// wherever [`Reader::next_node`] would be legal; hold the returned
    /// slice as a lazy value or feed it to any nosj entry point later.
    ///
    /// Scalar values are fully validated (their tokenizer runs to find
    /// the end). A skipped **container** is only structurally validated:
    /// brackets must balance (kinds are not matched), and the scalar
    /// tokens inside are not parsed, since the skip is a pure walk over
    /// the stage-1 index, touching no document bytes. A later full parse
    /// may therefore reject a document whose subtree this method skipped.
    #[inline]
    pub fn skip_value(&mut self) -> Result<&'j str, ParseError> {
        let (start, byte) = self.take_tok()?;
        let end = match byte {
            b'"' => {
                // SAFETY: `self.input` came from `&str` (or the caller's
                // UTF-8 vouch), and `start` holds the `"` just tokenized.
                unsafe { scalars::parse_string(self.input, start, &mut self.bufs.scratch) }
                    .map_err(|e| ParseError::scalar(e, start))?
                    .1
            }
            b'-' | b'0'..=b'9' => match scalars::parse_number(self.input, start) {
                Ok((_, end)) => end,
                // `-Infinity` fails the number grammar first.
                Err(e) => self.nan_keyword(start, ParseError::scalar(e, start))?.1,
            },
            b't' | b'f' | b'n' => {
                scalars::parse_literal(self.input, start)
                    .map_err(|e| ParseError::scalar(e, start))?
                    .1
            }
            b'{' | b'[' => self.skip_container()?,
            _ => {
                self.nan_keyword(start, ParseError::unexpected_character(start))?
                    .1
            }
        };
        // SAFETY: the constructor guarantees `input` is valid UTF-8, and
        // both boundaries fall on token edges (ASCII).
        Ok(unsafe { std::str::from_utf8_unchecked(&self.input[start..end]) })
    }

    /// Walk the index past the container whose opener was just consumed,
    /// returning the offset one past its closer. Pure index walk: depth
    /// counting on brackets, every other token entry stepped over.
    fn skip_container(&mut self) -> Result<usize, ParseError> {
        let mut depth = 1usize;
        loop {
            let (off, byte) = self.take_tok()?;
            match byte {
                b'{' | b'[' => depth += 1,
                b'}' | b']' => {
                    depth -= 1;
                    if depth == 0 {
                        return Ok(off + 1);
                    }
                }
                b'"' | b'-' | b'0'..=b'9' | b't' | b'f' | b'n' | b',' | b':' => {}
                // Keyword tokens step over like any scalar: skipped
                // content is never parsed.
                b'N' | b'I' if self.opts.allow_nan => {}
                _ => {
                    return Err(ParseError {
                        offset: off,
                        kind: ErrorKind::UnexpectedCharacter,
                    });
                }
            }
        }
    }

    /// Number of members in the container just opened by the last
    /// [`next_node`](Self::next_node) that returned [`Node::ObjectStart`] or
    /// [`Node::ArrayStart`] — pairs for an object, elements for an array. Must
    /// be called BEFORE the first `object_first_key`/`array_first`, while the
    /// cursor still sits on the container's first inner token.
    ///
    /// Read-only: it does NOT advance the cursor. Intended as a pre-sizing hint
    /// for hosts that allocate a native container up front (e.g. a Lua table via
    /// `lua_createtable(narr, nrec)`). Cost is O(container size in tokens): a
    /// pure index walk, the same one [`skip_value`](Self::skip_value) uses to
    /// step over a container, depth-counting brackets and tallying top-level
    /// commas. Returns 0 for an empty
    /// container. A trailing comma (see [`ParseOptions`]) is not counted as a
    /// member. On a malformed tail it returns whatever it counted before the
    /// index ran out — it never parses scalars, so it cannot error.
    #[must_use]
    pub fn container_len(&self) -> usize {
        let indexes = &self.bufs.indexes;
        // Empty container: the first inner token is already the closer.
        match indexes.get(self.pos).map(|&off| self.input[off as usize]) {
            Some(b'}' | b']') | None => return 0,
            _ => {}
        }
        let mut depth = 1usize;
        let mut commas = 0usize;
        let mut prev = 0u8;
        let mut i = self.pos;
        while let Some(&off) = indexes.get(i) {
            i += 1;
            let byte = self.input[off as usize];
            match byte {
                b'{' | b'[' => depth += 1,
                b'}' | b']' => {
                    depth -= 1;
                    if depth == 0 {
                        // A comma right before the closer ends no member.
                        if prev == b',' {
                            commas -= 1;
                        }
                        break;
                    }
                }
                b',' if depth == 1 => commas += 1,
                _ => {}
            }
            prev = byte;
        }
        commas + 1
    }

    /// Assert the document is complete (no trailing tokens).
    #[inline]
    pub fn finish(&mut self) -> Result<(), ParseError> {
        match self.peek_tok() {
            None => Ok(()),
            Some((o, _)) => Err(ParseError {
                offset: o,
                kind: ErrorKind::TrailingCharacters,
            }),
        }
    }
}

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

    #[test]
    fn container_len_counts_top_level_members() {
        let mut bufs = Buffers::new();
        let cases: &[(&str, usize)] = &[
            ("[]", 0),
            ("{}", 0),
            ("[1]", 1),
            (r#"{"x":1}"#, 1),
            ("[1,2,3]", 3),
            (r#"{"a":1,"b":2}"#, 2),
            ("[[1,2],[3]]", 2), // nested arrays: 2 top-level elements
            (r#"{"a":[1,2,3],"b":{"c":4}}"#, 2), // nested values don't inflate the count
            (r#"["a,b","c"]"#, 2), // commas inside strings are not structural
        ];
        for (src, want) in cases {
            let mut r = Reader::new(src, &mut bufs);
            let node = r.next_node().unwrap();
            assert!(
                matches!(node, Node::ObjectStart | Node::ArrayStart),
                "src {src}: expected a container start"
            );
            assert_eq!(r.container_len(), *want, "src {src}");
        }
    }

    /// Recursively walk one value, returning a debug rendering.
    fn walk(p: &mut Reader<'_, '_>) -> Result<String, ParseError> {
        Ok(match p.next_node()? {
            Node::ObjectStart => {
                let mut s = String::from("{");
                let mut key = p.object_first_key()?.map(str::to_string);
                while let Some(k) = key {
                    s.push_str(&k);
                    s.push('=');
                    s.push_str(&walk(p)?);
                    s.push(';');
                    key = p.object_next_key()?.map(str::to_string);
                }
                s.push('}');
                s
            }
            Node::ArrayStart => {
                let mut s = String::from("[");
                let mut more = p.array_first()?;
                while more {
                    s.push_str(&walk(p)?);
                    s.push(';');
                    more = p.array_next()?;
                }
                s.push(']');
                s
            }
            Node::Str(v) => format!("str({v})"),
            Node::Int(v) => format!("int({v})"),
            Node::Float(v) => format!("f({v})"),
            Node::BigInt(v) => format!("big({v})"),
            Node::Bool(v) => format!("b({v})"),
            Node::Null => "null".into(),
        })
    }

    fn parse(doc: &str) -> Result<String, ParseError> {
        let mut bufs = Buffers::new();
        let mut p = Reader::new(doc, &mut bufs);
        let out = walk(&mut p)?;
        p.finish()?;
        Ok(out)
    }

    fn parse_with(doc: &str, opts: ParseOptions) -> Result<String, ParseError> {
        let mut bufs = Buffers::new();
        let mut p = Reader::new_with(doc, &mut bufs, opts);
        let out = walk(&mut p)?;
        p.finish()?;
        Ok(out)
    }

    /// Skip a whole document under `opts`, returning the skipped text.
    fn skip_with(doc: &str, opts: ParseOptions) -> Result<String, ParseError> {
        let mut bufs = Buffers::new();
        let mut p = Reader::new_with(doc, &mut bufs, opts);
        let raw = p.skip_value()?.to_string();
        p.finish()?;
        Ok(raw)
    }

    /// Each extension is walkable (and skippable) exactly when its option
    /// is on; the default Reader keeps rejecting it.
    #[test]
    fn parse_options_extend_the_walk() {
        let trailing = ParseOptions {
            allow_trailing_comma: true,
            ..ParseOptions::default()
        };
        let nan = ParseOptions {
            allow_nan: true,
            ..ParseOptions::default()
        };
        let cases: &[(&str, ParseOptions, &str)] = &[
            ("[1,2,]", trailing, "[int(1);int(2);]"),
            (r#"{"a":1,}"#, trailing, "{a=int(1);}"),
            (
                r#"{"a":[1,],"b":{"c":2,},}"#,
                trailing,
                "{a=[int(1);];b={c=int(2);};}",
            ),
            ("[ 1 , ]", trailing, "[int(1);]"),
            ("[NaN,Infinity,-Infinity]", nan, "[f(NaN);f(inf);f(-inf);]"),
            (r#"{"a":NaN}"#, nan, "{a=f(NaN);}"),
            ("NaN", nan, "f(NaN)"),
        ];
        for &(doc, opts, want) in cases {
            assert_eq!(parse_with(doc, opts).unwrap(), want, "{doc}");
            assert!(parse(doc).is_err(), "{doc} must stay strict by default");
            assert_eq!(skip_with(doc, opts).unwrap(), doc, "{doc} skip");
        }
        // Options never loosen anything else.
        for bad in ["[,]", "{,}", "[1,,]", r#"{"a":1,,}"#, "[1 2]"] {
            assert!(parse_with(bad, trailing).is_err(), "{bad}");
        }
        for bad in ["[Nan]", "[NaNx]", "[-Inf]", "[Infinityy]", "[+Infinity]"] {
            assert!(parse_with(bad, nan).is_err(), "{bad}");
        }
        // Keywords inside a skipped container are stepped over like any
        // scalar token.
        assert_eq!(
            skip_with(r#"{"a":[NaN,{"b":Infinity}]}"#, nan).unwrap(),
            r#"{"a":[NaN,{"b":Infinity}]}"#
        );
        assert!(skip_with(r#"{"a":[NaN]}"#, ParseOptions::default()).is_err());
    }

    #[test]
    fn container_len_ignores_a_trailing_comma() {
        let mut bufs = Buffers::new();
        for (src, want) in [("[1,2,]", 2), (r#"{"a":1,}"#, 1), ("[[1,],2,]", 2)] {
            let mut r = Reader::new(src, &mut bufs);
            r.next_node().unwrap();
            assert_eq!(r.container_len(), want, "src {src}");
        }
    }

    #[test]
    fn round_trips() {
        assert_eq!(
            parse(r#"{"a":1,"b":[true,null]}"#).unwrap(),
            "{a=int(1);b=[b(true);null;];}"
        );
        assert_eq!(parse("[]").unwrap(), "[]");
        assert_eq!(parse("{}").unwrap(), "{}");
        assert_eq!(parse("  42  ").unwrap(), "int(42)");
        assert_eq!(parse(r#""hi""#).unwrap(), "str(hi)");
        assert_eq!(parse("[[[1]]]").unwrap(), "[[[int(1);];];]");
    }

    #[test]
    fn errors() {
        assert!(parse("").is_err());
        assert!(parse("{").is_err());
        assert!(parse(r#"{"a"}"#).is_err());
        assert!(parse(r#"{"a":}"#).is_err());
        assert!(parse("[1,]").is_err());
        assert!(parse("[1 2]").is_err());
        assert!(parse("{} {}").is_err());
        assert!(parse("truex").is_err());
        assert!(parse("01").is_err());
    }

    /// Skip every kind of value and check the returned raw slice and the
    /// parser position afterwards.
    #[test]
    fn skip_value_kinds() {
        let doc = r#"{"a":[1,2.5,"x\n",{"k":[true,null]},-0],"b":"tail"}"#;
        // Skip the whole document at the root.
        let mut bufs = Buffers::new();
        let mut p = Reader::new(doc, &mut bufs);
        assert_eq!(p.skip_value().unwrap(), doc);
        p.finish().unwrap();

        // Skip "a"'s array value, then parse "b" normally.
        let mut p = Reader::new(doc, &mut bufs);
        assert_eq!(p.next_node().unwrap(), Node::ObjectStart);
        assert_eq!(p.object_first_key().unwrap(), Some("a"));
        assert_eq!(
            p.skip_value().unwrap(),
            r#"[1,2.5,"x\n",{"k":[true,null]},-0]"#
        );
        assert_eq!(p.object_next_key().unwrap(), Some("b"));
        assert_eq!(p.next_node().unwrap(), Node::Str("tail"));
        assert_eq!(p.object_next_key().unwrap(), None);
        p.finish().unwrap();

        // Scalar skips return the exact token text (and validate it).
        for (doc, raw) in [
            ("  1234  ", "1234"),
            (" -2.5e3 ", "-2.5e3"),
            (r#" "es\tc" "#, r#""es\tc""#),
            (" true ", "true"),
            (" null ", "null"),
        ] {
            let mut p = Reader::new(doc, &mut bufs);
            assert_eq!(p.skip_value().unwrap(), raw, "{doc:?}");
            p.finish().unwrap();
        }
    }

    /// Skipping and walking must consume exactly the same tokens: skip a
    /// value, then re-parse its returned slice and get the same rendering
    /// as walking it in place.
    #[test]
    fn skip_matches_walk() {
        let doc = r#"[[1,[2,[3]]],{"a":{"b":[{"c":"d"}]}},"s",7.5,null]"#;
        let mut bufs = Buffers::new();

        // Walk each element of the root array normally...
        let mut p = Reader::new(doc, &mut bufs);
        assert_eq!(p.next_node().unwrap(), Node::ArrayStart);
        let mut walked = Vec::new();
        let mut more = p.array_first().unwrap();
        while more {
            walked.push(walk(&mut p).unwrap());
            more = p.array_next().unwrap();
        }
        p.finish().unwrap();

        // ...then skip each element and re-parse its slice.
        let mut p = Reader::new(doc, &mut bufs);
        assert_eq!(p.next_node().unwrap(), Node::ArrayStart);
        let mut slices = Vec::new();
        let mut more = p.array_first().unwrap();
        while more {
            slices.push(p.skip_value().unwrap());
            more = p.array_next().unwrap();
        }
        p.finish().unwrap();

        assert_eq!(walked.len(), slices.len());
        for (rendered, slice) in walked.iter().zip(&slices) {
            let mut bufs = Buffers::new();
            let mut p = Reader::new(slice, &mut bufs);
            assert_eq!(&walk(&mut p).unwrap(), rendered, "slice {slice:?}");
            p.finish().unwrap();
        }
    }

    #[test]
    fn skip_errors() {
        let mut bufs = Buffers::new();
        // Unterminated container.
        let mut p = Reader::new("[1,2", &mut bufs);
        assert!(p.skip_value().is_err());
        // Invalid scalar at skip position is still validated.
        let mut p = Reader::new("01", &mut bufs);
        assert!(p.skip_value().is_err());
        let mut p = Reader::new("\"unterminated", &mut bufs);
        assert!(p.skip_value().is_err());
    }

    #[test]
    fn benchmark_files_walk_clean() {
        for name in [
            "twitter",
            "canada",
            "citm_catalog",
            "tolstoy",
            "numbers",
            "mesh",
        ] {
            let path = format!(
                "{}/../../benchmark/{}.json",
                env!("CARGO_MANIFEST_DIR"),
                name
            );
            if let Ok(data) = std::fs::read_to_string(&path) {
                let mut bufs = Buffers::new();
                let mut p = Reader::new(&data, &mut bufs);
                walk(&mut p).unwrap_or_else(|e| panic!("{name}: {e}"));
                p.finish().unwrap_or_else(|e| panic!("{name}: {e}"));

                // Skipping the whole document consumes exactly the same
                // tokens and returns the trimmed source.
                let mut p = Reader::new(&data, &mut bufs);
                let slice = p.skip_value().unwrap_or_else(|e| panic!("{name}: {e}"));
                let json_ws: &[char] = &[' ', '\t', '\n', '\r'];
                assert_eq!(slice, data.trim_matches(json_ws), "{name}: root skip slice");
                p.finish().unwrap_or_else(|e| panic!("{name}: {e}"));
            }
        }
    }
}