omnist 0.2.1-alpha

use super::*;
use crate::document::{Doc, Value};
use indexmap::IndexMap;

fn obj(pairs: &[(&str, Value)]) -> Value {
    let mut m = IndexMap::new();
    for (k, v) in pairs {
        m.insert((*k).to_string(), v.clone());
    }
    Value::Object(m)
}

fn roundtrip_value(v: &Value) {
    let doc = Doc::of(v).unwrap();
    let raw = doc.to_raw();
    let text = write_oml(&raw, 2).unwrap();
    let parsed = read_oml(&text).unwrap();
    let doc2 = Doc::from_raw(parsed).unwrap();
    assert!(
        doc.eq_doc(&doc2),
        "round trip mismatch for {v:?}\n--- OML ---\n{text}"
    );
}

// -- scalar round trips (every kind) ----------------------------------------

#[test]
fn string_round_trips() {
    roundtrip_value(&obj(&[("a", Value::Str("hello world".into()))]));
}

#[test]
fn string_with_escapes_round_trips() {
    roundtrip_value(&obj(&[(
        "a",
        Value::Str("quote:\" backslash:\\ nl:\n cr:\r tab:\t ctrl:\u{1}".into()),
    )]));
}

#[test]
fn integer_round_trips_positive_and_negative() {
    roundtrip_value(&obj(&[
        ("a", Value::Int((42).into())),
        ("b", Value::Int((-42).into())),
        ("c", Value::Int((0).into())),
    ]));
}

#[test]
fn number_round_trips_decimal_and_whole_valued_float() {
    roundtrip_value(&obj(&[
        ("a", Value::Float(3.15)),
        ("b", Value::Float(1.0)),
        ("c", Value::Float(-2.0)),
        ("d", Value::Float(1e10)),
    ]));
}

#[test]
fn number_round_trips_nan_and_infinities() {
    let doc = Doc::of(&obj(&[
        ("a", Value::Float(f64::NAN)),
        ("b", Value::Float(f64::INFINITY)),
        ("c", Value::Float(f64::NEG_INFINITY)),
    ]))
    .unwrap();
    let text = write_oml(&doc.to_raw(), 2).unwrap();
    assert!(text.contains("a: nan"));
    assert!(text.contains("b: inf"));
    assert!(text.contains("c: -inf"));
    let parsed = Doc::from_raw(read_oml(&text).unwrap()).unwrap();
    let root = parsed.root();
    assert!(matches!(
        root.child("a").unwrap().value().unwrap(),
        crate::document::Scalar::Float(f) if f.is_nan()
    ));
    assert!(matches!(
        root.child("b").unwrap().value().unwrap(),
        crate::document::Scalar::Float(f) if f.is_infinite() && *f > 0.0
    ));
    assert!(matches!(
        root.child("c").unwrap().value().unwrap(),
        crate::document::Scalar::Float(f) if f.is_infinite() && *f < 0.0
    ));
}

#[test]
fn boolean_round_trips() {
    roundtrip_value(&obj(&[("a", Value::Bool(true)), ("b", Value::Bool(false))]));
}

#[test]
fn null_round_trips() {
    roundtrip_value(&obj(&[("a", Value::Null)]));
}

#[test]
fn date_round_trips() {
    roundtrip_value(&obj(&[("a", Value::Str("2024-01-01".into()))]));
}

#[test]
fn datetime_date_then_time_lookahead_canonicalizes_missing_seconds() {
    // Regression test for issue #90: the DATE-then-TIME one-token lookahead
    // merge in the scanner correctly produces a single DATETIME token, but
    // was storing the raw source text ("2024-01-01T10:30") instead of the
    // canonical form with seconds filled in, unlike Python's real
    // `datetime` object (`datetime.datetime(2024, 1, 1, 10, 30).isoformat()
    // == "2024-01-01T10:30:00"`).
    let parsed = Doc::from_raw(read_oml("a: 2024-01-01T10:30\n").unwrap()).unwrap();
    let value = parsed.root().child("a").unwrap().value().unwrap();
    assert!(
        matches!(value, crate::document::Scalar::Datetime(s) if s == "2024-01-01T10:30:00"),
        "expected canonical '2024-01-01T10:30:00', got {value:?}"
    );
}

#[test]
fn time_round_trips_in_canonical_form() {
    // Only already-canonical spellings round-trip identity-wise: a
    // missing-seconds or under-padded-fraction source spelling is
    // canonicalized on read (issue #90), so it deliberately does NOT
    // round-trip back to the original `Value::Str` -- see
    // `time_literal_without_seconds_canonicalizes_on_read` below for that
    // behavior.
    roundtrip_value(&obj(&[
        ("a", Value::Str("12:00:00".into())),
        ("b", Value::Str("12:00:00.123456".into())),
    ]));
}

#[test]
fn time_literal_without_seconds_canonicalizes_on_read() {
    let parsed = Doc::from_raw(read_oml("a: 12:00\n").unwrap()).unwrap();
    let value = parsed.root().child("a").unwrap().value().unwrap();
    assert!(
        matches!(value, crate::document::Scalar::Time(s) if s == "12:00:00"),
        "expected canonical '12:00:00', got {value:?}"
    );
}

#[test]
fn time_literal_with_short_fraction_zero_pads_on_read() {
    let parsed = Doc::from_raw(read_oml("a: 12:00:00.5\n").unwrap()).unwrap();
    let value = parsed.root().child("a").unwrap().value().unwrap();
    assert!(
        matches!(value, crate::document::Scalar::Time(s) if s == "12:00:00.500000"),
        "expected canonical '12:00:00.500000', got {value:?}"
    );
}

// A bare TIME literal (not DATETIME) may itself carry a UTC offset
// (`TIME_RE` in schema.rs has its own `off` group, independent of
// DATETIME's) -- exercises the scanner's offset-lookahead for a bare
// TIME token specifically, and `canonicalize_time_captures`'s offset
// branch.
#[test]
fn bare_time_literal_with_utc_offset_reads_as_a_genuine_temporal_leaf() {
    let parsed = read_oml("a: 12:00:00+05:00\n").unwrap();
    let RawNode::Edges(edges) = &parsed else {
        panic!("expected edges, got {parsed:?}");
    };
    assert_eq!(
        edges[0],
        (
            "a".to_string(),
            RawNode::Leaf(crate::document::Scalar::Time("12:00:00+05:00".to_string()))
        )
    );
}

#[test]
fn datetime_round_trips_with_and_without_utc_offset() {
    roundtrip_value(&obj(&[
        ("a", Value::Str("2024-01-01T12:00:00".into())),
        ("b", Value::Str("2024-01-01T12:00:00+02:00".into())),
        ("c", Value::Str("2024-01-01T12:00:00-05:30".into())),
    ]));
}

#[test]
fn nested_object_round_trips() {
    roundtrip_value(&obj(&[(
        "a",
        obj(&[("b", Value::Int((1).into())), ("c", Value::Str("x".into()))]),
    )]));
}

#[test]
fn repeated_label_contiguous_round_trips() {
    roundtrip_value(&obj(&[(
        "tag",
        Value::Array(vec![Value::Str("x".into()), Value::Str("y".into())]),
    )]));
}

#[test]
fn empty_document_round_trips() {
    let doc = Doc::from_raw(RawNode::Edges(vec![])).unwrap();
    let text = write_oml(&doc.to_raw(), 2).unwrap();
    assert_eq!(text, "");
    let parsed = Doc::from_raw(read_oml(&text).unwrap()).unwrap();
    assert!(doc.eq_doc(&parsed));
}

#[test]
fn bare_top_level_scalar_round_trips() {
    let doc = Doc::of(&Value::Int((7).into())).unwrap();
    let text = write_oml(&doc.to_raw(), 2).unwrap();
    assert_eq!(text, "7");
    let parsed = Doc::from_raw(read_oml(&text).unwrap()).unwrap();
    assert!(doc.eq_doc(&parsed));
}

// -- interleaved repeated labels: the case Value/IndexMap cannot represent --

#[test]
fn interleaved_repeated_labels_round_trip_exactly_via_raw_node() {
    let raw = RawNode::Edges(vec![
        (
            "b".to_string(),
            RawNode::Leaf(crate::document::Scalar::Int((1).into())),
        ),
        (
            "c".to_string(),
            RawNode::Leaf(crate::document::Scalar::Int((2).into())),
        ),
        (
            "b".to_string(),
            RawNode::Leaf(crate::document::Scalar::Int((3).into())),
        ),
    ]);
    let text = write_oml(&raw, 2).unwrap();
    assert_eq!(text, "b: 1\nc: 2\nb: 3");
    let parsed = read_oml(&text).unwrap();
    assert_eq!(parsed, raw);
}

// -- the four TS-bug-derived regressions ------------------------------------

// omnist-ts#37 / #70: write_oml must call the shared depth guard itself.
#[test]
fn write_oml_calls_the_shared_depth_guard() {
    fn nest_raw(levels: usize) -> RawNode {
        let mut n = RawNode::Leaf(crate::document::Scalar::Int((0).into()));
        for _ in 0..levels {
            n = RawNode::Edges(vec![("a".to_string(), n)]);
        }
        n
    }
    // At exactly MAX_DEPTH it's accepted (boundary, matches document.rs's
    // own "reject only depth > MAX_DEPTH" convention).
    assert!(write_oml(&nest_raw(document::MAX_DEPTH), 2).is_ok());
    // One past it, write_oml must reject on its own -- this RawNode was
    // never passed through Doc::from_raw's guarded construction, so if
    // write_oml didn't call check_write_depth itself, this would write
    // an over-deep tree with no error, byte-identical to the omnist-ts
    // bug (a writer trusting an upstream check that never ran).
    let err = write_oml(&nest_raw(document::MAX_DEPTH + 1), 2).unwrap_err();
    assert!(err.to_string().contains("maximum depth"));
    // Same guard, compact writer entry point.
    let err2 = write_oml_compact(&nest_raw(document::MAX_DEPTH + 1)).unwrap_err();
    assert!(err2.to_string().contains("maximum depth"));
}

// omnist-ts#51: writeOml erased a datetime UTC offset.
#[test]
fn write_oml_preserves_datetime_utc_offset_exactly() {
    let raw = RawNode::Edges(vec![(
        "meeting".to_string(),
        RawNode::Leaf(crate::document::Scalar::Str(
            "2024-06-01T09:30:00+02:00".to_string(),
        )),
    )]);
    let text = write_oml(&raw, 2).unwrap();
    assert!(
        text.contains("+02:00"),
        "offset must survive the write unchanged: {text}"
    );
    let parsed = read_oml(&text).unwrap();
    assert_eq!(parsed, raw);
}

// omnist-ts#52: a bare TIME literal did not round-trip (became a quoted
// string on write-then-read). Now expressed via a real `Scalar::Time`
// (issue #105, superseding issue #99's `RawNode::TemporalLeaf` write-hint
// tag) -- a genuinely temporal value, not a plain string that merely
// looks like one; see `plain_string_shaped_like_a_time_stays_quoted`
// below for the case this variant exists to distinguish from.
#[test]
fn bare_time_literal_round_trips_as_a_time_not_a_quoted_string() {
    let raw = RawNode::Edges(vec![(
        "a".to_string(),
        RawNode::Leaf(crate::document::Scalar::Time("12:00".to_string())),
    )]);
    let text = write_oml(&raw, 2).unwrap();
    // Must write as a bare, unquoted time literal, not `"12:00"`.
    assert_eq!(text, "a: 12:00");
    // Reading it back canonicalizes the missing seconds (issue #90) and
    // is itself a genuinely-read bare literal, so the parsed value is
    // `Scalar::Time("12:00:00")`, not a byte-identical round trip of `raw`.
    let expected = RawNode::Edges(vec![(
        "a".to_string(),
        RawNode::Leaf(crate::document::Scalar::Time("12:00:00".to_string())),
    )]);
    let parsed = read_oml(&text).unwrap();
    assert_eq!(parsed, expected);
}

// issue #99: a plain string that merely *looks* like a time must stay
// quoted on write -- writing it bare would silently promote it to a
// genuine temporal literal on the next read (a different Document).
#[test]
fn plain_string_shaped_like_a_time_stays_quoted() {
    let raw = RawNode::Edges(vec![(
        "a".to_string(),
        RawNode::Leaf(crate::document::Scalar::Str("12:00:00".to_string())),
    )]);
    let text = write_oml(&raw, 2).unwrap();
    assert_eq!(text, "a: \"12:00:00\"");
    // Reading it back must reproduce the identical plain string, not a
    // `TemporalLeaf`.
    let parsed = read_oml(&text).unwrap();
    assert_eq!(parsed, raw);
}

// issue #99's own repro: a JSON-shaped plain string that looks like a
// date must not become a genuine bare temporal literal through OML.
#[test]
fn plain_string_shaped_like_a_date_stays_quoted() {
    let raw = RawNode::Edges(vec![(
        "d".to_string(),
        RawNode::Leaf(crate::document::Scalar::Str("2024-01-01".to_string())),
    )]);
    let text = write_oml(&raw, 2).unwrap();
    assert_eq!(text, "d: \"2024-01-01\"");
    let parsed = read_oml(&text).unwrap();
    assert_eq!(parsed, raw);
}

// The companion case: a genuinely date/datetime-kinded value (e.g. read
// from OML's own bare grammar) writes bare, no quotes.
#[test]
fn genuine_temporal_leaf_writes_bare_for_date_and_datetime_too() {
    let raw = RawNode::Edges(vec![
        (
            "d".to_string(),
            RawNode::Leaf(crate::document::Scalar::Date("2024-01-01".to_string())),
        ),
        (
            "dt".to_string(),
            RawNode::Leaf(crate::document::Scalar::Datetime(
                "2024-01-01T12:30:00".to_string(),
            )),
        ),
    ]);
    let text = write_oml(&raw, 2).unwrap();
    assert_eq!(text, "d: 2024-01-01\ndt: 2024-01-01T12:30:00");
    // Compact mode dispatches through the same `write_scalar` (issue
    // #105), separate from pretty mode's own call site -- covered here
    // too rather than only through the pretty-mode assertion above.
    assert_eq!(
        write_oml_compact(&raw).unwrap(),
        "d: 2024-01-01; dt: 2024-01-01T12:30:00"
    );
}

// A schema-directed materialize upgrade to a Date-kinded field is the
// other real provenance source (alongside OML's own bare-literal
// grammar) -- confirm it also produces bare OML output, per issue #99's
// vector comment naming this exact case.
#[test]
fn materialize_upgraded_date_writes_bare_through_oml() {
    use crate::schema::{DATE, Field, Record, Ref, Schema};
    let root = Record::new(vec![Field::required("d", DATE).unwrap()]).unwrap();
    let mut env = IndexMap::new();
    env.insert("Root".to_string(), root);
    let schema = Schema::new(Ref::new("Root"), env).unwrap();
    let input = RawNode::Edges(vec![(
        "d".to_string(),
        RawNode::Leaf(crate::document::Scalar::Str("2024-01-01".to_string())),
    )]);
    let materialized = crate::materialize::materialize(&input, Some(&schema)).unwrap();
    let text = write_oml(&materialized, 2).unwrap();
    assert_eq!(text, "d: 2024-01-01");
}

// omnist-ts#36-equivalent: writer escaping must be all-occurrences, not
// first-match-only (a non-global regex-style replace would under-sanitize).
#[test]
fn string_escaping_is_all_occurrences_not_first_match_only() {
    let s = "a\"b\"c\"d\\e\\f";
    let written = write_scalar(&crate::document::Scalar::Str(s.to_string()));
    assert_eq!(written, r#""a\"b\"c\"d\\e\\f""#);
    // Every quote and every backslash escaped -- not just the first of each.
    assert_eq!(written.matches("\\\"").count(), 3);
    assert_eq!(written.matches("\\\\").count(), 2);
}

// -- malformed input / error positions --------------------------------------

#[test]
fn stray_character_reports_position() {
    let err = read_oml("a: 1\n@").unwrap_err();
    assert!(err.message.contains("stray character"));
    assert_eq!(err.line, 2);
}

#[test]
fn unterminated_string_is_an_error() {
    let err = read_oml("a: \"unterminated").unwrap_err();
    assert!(err.message.contains("unterminated string"));
}

#[test]
fn unterminated_raw_string_is_an_error() {
    let err = read_oml("a: 'unterminated").unwrap_err();
    assert!(err.message.contains("unterminated raw string"));
}

#[test]
fn unterminated_multiline_string_is_an_error() {
    let err = read_oml("a: \"\"\"unterminated").unwrap_err();
    assert!(err.message.contains("unterminated multiline string"));
}

#[test]
fn control_character_in_string_is_rejected() {
    let err = read_oml("a: \"bad\u{1}char\"").unwrap_err();
    assert!(err.message.contains("control character"));
}

#[test]
fn control_character_in_multiline_string_is_rejected() {
    let err = read_oml("a: \"\"\"bad\u{1}char\"\"\"").unwrap_err();
    assert!(err.message.contains("control character"));
}

#[test]
fn bare_word_is_rejected_as_a_value() {
    let err = read_oml("a: bareword").unwrap_err();
    assert!(err.message.contains("bare word"));
}

#[test]
fn reserved_word_cannot_be_a_bare_label() {
    // At the very top of a document, `null` parses as the null scalar
    // (looks_like_edge deliberately excludes reserved words -- matching
    // Python's `_looks_like_edge`), so the reserved-word-as-label check
    // only fires once the parser is unambiguously in edge/label position,
    // e.g. inside a `{ }` block.
    let err = read_oml("a: { null: 1 }").unwrap_err();
    assert!(err.message.contains("reserved word"));
}

// issue #75: capitalized/mixed-case spellings of the reserved words are
// bare idents, not the keyword -- `matches_word` compares byte-for-byte
// (see scanner.rs), so only the exact lowercase spelling is recognized.
// A bare ident used as a *value* isn't a valid token on its own (only
// null/true/false/nan/inf/-inf are), so it surfaces as a "bare word" error
// -- mirroring Python's `test_capitalized_nan_is_bare_ident_not_keyword`
// and the TS port's equivalent case.
#[test]
fn capitalized_reserved_words_are_bare_idents_not_keywords() {
    for word in ["NAN", "NaN", "INF", "Inf", "NULL", "Null", "TRUE", "False"] {
        let src = format!("a: {word}");
        let err =
            read_oml(&src).expect_err(&format!("expected {word:?} to be rejected as a bare word"));
        assert!(
            err.message.contains("bare word"),
            "{word:?}: expected a bare-word error, got {:?}",
            err.message
        );
    }
}

// issue #75: `nan`/`inf`/`-inf` (and the other reserved words) are valid
// field labels once quoted -- `is_bare_label` explicitly excludes them
// from ever being written unquoted (writer.rs), and a quoted string is
// always a valid label on read, so these round-trip exactly like any
// other label. Mirrors Python's grammar-doc conformance tests
// (`test_oml_ex9_nan_is_a_number_token_never_a_label` /
// `test_oml_ex10_quoted_nan_is_a_valid_label`) and the TS port's
// parametrized "-inf"/"nan"/"inf"-as-a-label test.
#[test]
fn quoted_reserved_words_are_valid_labels_and_round_trip() {
    for label in ["nan", "inf", "-inf", "null", "true", "false"] {
        roundtrip_value(&obj(&[(label, Value::Int((1).into()))]));
    }
}

#[test]
fn invalid_date_reports_a_clear_error() {
    let err = read_oml("a: 2024-02-30").unwrap_err();
    assert!(err.message.contains("invalid date"));
}

#[test]
fn invalid_time_reports_a_clear_error() {
    let err = read_oml("a: 25:00:00").unwrap_err();
    assert!(err.message.contains("invalid time"));
}

#[test]
fn invalid_datetime_reports_a_clear_error() {
    let err = read_oml("a: 2024-13-01T12:00:00").unwrap_err();
    assert!(err.message.contains("invalid datetime"));
}

#[test]
fn empty_array_is_rejected() {
    let err = read_oml("a: []").unwrap_err();
    assert!(err.message.contains("empty array"));
}

#[test]
fn nested_array_is_rejected() {
    let err = read_oml("a: [[1]]").unwrap_err();
    assert!(err.message.contains("nested array"));
}

#[test]
fn missing_colon_after_label_reports_position() {
    // At the document's very top level, "a 1" doesn't look like an edge
    // attempt at all (looks_like_edge needs a colon right after the first
    // token, which isn't there), so "a" alone is read as a bare-word-value
    // error instead. The "expected ':'" error is unambiguous once already
    // inside a `{ }` block, where a label is mandatory.
    let err = read_oml("r: { a 1 }").unwrap_err();
    assert!(err.message.contains("expected ':'"));
}

#[test]
fn missing_closing_brace_is_an_error() {
    let err = read_oml("a: { b: 1").unwrap_err();
    assert!(err.message.contains("expected '}'"));
}

#[test]
fn missing_separator_between_edges_is_an_error() {
    let err = read_oml("a: 1 b: 2").unwrap_err();
    assert!(err.message.contains("expected a separator"));
}

#[test]
fn trailing_content_after_document_body_is_an_error() {
    let err = read_oml("1 2").unwrap_err();
    assert!(err.message.contains("unexpected trailing content"));
}

#[test]
fn unquoted_field_label_must_be_a_valid_ident_or_string() {
    let err = read_oml("a: 1\n5: 2").unwrap_err();
    assert!(
        err.message.contains("unexpected trailing content") || err.message.contains("expected")
    );
}

#[test]
fn integer_digit_cap_is_enforced() {
    let digits = "1".repeat(MAX_INT_DIGITS + 1);
    let src = format!("a: {digits}");
    let err = read_oml(&src).unwrap_err();
    assert!(err.message.contains("digit"));
    assert!(err.message.contains("4300"));
}

#[test]
fn integer_at_the_digit_cap_boundary_parses() {
    // Exactly MAX_INT_DIGITS digits passes the cap check, and (issue #104)
    // `Scalar::Int` is arbitrary-precision, so this is now a real,
    // successfully-parsed value, not an i64-range error.
    let digits = "1".repeat(MAX_INT_DIGITS);
    let src = format!("a: {digits}");
    let parsed = Doc::from_raw(read_oml(&src).unwrap()).unwrap();
    let value = parsed.root().child("a").unwrap().value().unwrap();
    assert!(
        matches!(value, crate::document::Scalar::Int(i) if i.to_string().len() == MAX_INT_DIGITS),
        "got {value:?}"
    );
}

#[test]
fn integer_cap_does_not_false_positive_on_a_long_identifier() {
    // A guard-against-false-positives check (issue #10): an identifier
    // that merely *contains* a long digit run is not an INTEGER token at
    // all (IDENT must start with a letter/underscore), so the cap must
    // never fire on it.
    let ident = format!("x{}", "1".repeat(MAX_INT_DIGITS + 5));
    let err = read_oml(&format!("a: {ident}")).unwrap_err();
    assert!(err.message.contains("bare word"));
    assert!(!err.message.contains("digit"));
}

#[test]
fn beyond_i64_integer_parses_arbitrary_precision() {
    // Issue #104: no i64 ceiling anymore -- a 20-digit literal is a real,
    // correctly-parsed value.
    let parsed = Doc::from_raw(read_oml("a: 99999999999999999999").unwrap()).unwrap();
    let value = parsed.root().child("a").unwrap().value().unwrap();
    assert_eq!(
        value,
        &crate::document::Scalar::Int(
            num_bigint::BigInt::parse_bytes(b"99999999999999999999", 10).unwrap()
        )
    );
}

#[test]
fn invalid_unicode_escape_is_rejected() {
    let err = read_oml(r#"a: "\uZZZZ""#).unwrap_err();
    assert!(err.message.contains(r"invalid \u escape"));
}

#[test]
fn unpaired_high_surrogate_is_rejected() {
    let err = read_oml(r#"a: "\ud800""#).unwrap_err();
    assert!(err.message.contains("unpaired high surrogate"));
}

#[test]
fn unpaired_low_surrogate_is_rejected() {
    let err = read_oml(r#"a: "\udc00""#).unwrap_err();
    assert!(err.message.contains("unpaired low surrogate"));
}

#[test]
fn valid_surrogate_pair_decodes_to_the_combined_codepoint() {
    // U+1F600 (grinning face) as a UTF-16 surrogate pair.
    let node = read_oml(r#"a: "😀""#).unwrap();
    match node {
        RawNode::Edges(edges) => {
            assert_eq!(edges.len(), 1);
            match &edges[0].1 {
                RawNode::Leaf(crate::document::Scalar::Str(s)) => {
                    assert_eq!(s, "\u{1F600}");
                }
                other => panic!("expected a string leaf, got {other:?}"),
            }
        }
        other => panic!("expected edges, got {other:?}"),
    }
}

#[test]
fn invalid_escape_character_is_rejected() {
    let err = read_oml(r#"a: "\q""#).unwrap_err();
    assert!(err.message.contains("invalid escape"));
}

#[test]
fn unterminated_escape_sequence_is_rejected() {
    let err = read_oml("a: \"\\").unwrap_err();
    assert!(err.message.contains("unterminated escape"));
}

// -- OML-Extended on read: raw strings (E2) and triple-quoted (E3) ---------

#[test]
fn raw_string_e2_reads_with_no_escape_processing() {
    let node = read_oml(r"a: 'no \n escapes here \\'").unwrap();
    match node {
        RawNode::Edges(edges) => match &edges[0].1 {
            RawNode::Leaf(crate::document::Scalar::Str(s)) => {
                assert_eq!(s, r"no \n escapes here \\");
            }
            other => panic!("expected a string leaf, got {other:?}"),
        },
        other => panic!("expected edges, got {other:?}"),
    }
}

#[test]
fn triple_quoted_multiline_e3_strips_opening_newline() {
    let node = read_oml("a: \"\"\"\nline one\nline two\"\"\"").unwrap();
    match node {
        RawNode::Edges(edges) => match &edges[0].1 {
            RawNode::Leaf(crate::document::Scalar::Str(s)) => {
                assert_eq!(s, "line one\nline two");
            }
            other => panic!("expected a string leaf, got {other:?}"),
        },
        other => panic!("expected edges, got {other:?}"),
    }
}

#[test]
fn triple_quoted_multiline_e3_allows_embedded_quotes_up_to_two() {
    let node = read_oml("a: \"\"\"has \"\" two quotes\"\"\"").unwrap();
    match node {
        RawNode::Edges(edges) => match &edges[0].1 {
            RawNode::Leaf(crate::document::Scalar::Str(s)) => {
                assert_eq!(s, "has \"\" two quotes");
            }
            other => panic!("expected a string leaf, got {other:?}"),
        },
        other => panic!("expected edges, got {other:?}"),
    }
}

// -- writer form details ----------------------------------------------------

#[test]
fn write_oml_compact_joins_edges_with_semicolons() {
    let raw = RawNode::Edges(vec![
        (
            "a".to_string(),
            RawNode::Leaf(crate::document::Scalar::Int((1).into())),
        ),
        (
            "b".to_string(),
            RawNode::Leaf(crate::document::Scalar::Int((2).into())),
        ),
    ]);
    assert_eq!(write_oml_compact(&raw).unwrap(), "a: 1; b: 2");
}

#[test]
fn write_oml_pretty_indents_nested_objects() {
    let raw = RawNode::Edges(vec![(
        "a".to_string(),
        RawNode::Edges(vec![(
            "b".to_string(),
            RawNode::Leaf(crate::document::Scalar::Int((1).into())),
        )]),
    )]);
    assert_eq!(write_oml(&raw, 2).unwrap(), "a: {\n  b: 1\n}");
}

#[test]
fn write_oml_writes_empty_object_compactly_in_both_modes() {
    let raw = RawNode::Edges(vec![("a".to_string(), RawNode::Edges(vec![]))]);
    assert_eq!(write_oml(&raw, 2).unwrap(), "a: {}");
    assert_eq!(write_oml_compact(&raw).unwrap(), "a: {}");
}

#[test]
fn labels_needing_quotes_are_quoted_on_write() {
    let raw = RawNode::Edges(vec![
        (
            "has space".to_string(),
            RawNode::Leaf(crate::document::Scalar::Int((1).into())),
        ),
        (
            "null".to_string(),
            RawNode::Leaf(crate::document::Scalar::Int((2).into())),
        ),
        (
            "1leading".to_string(),
            RawNode::Leaf(crate::document::Scalar::Int((3).into())),
        ),
    ]);
    let text = write_oml(&raw, 2).unwrap();
    assert!(text.contains("\"has space\": 1"));
    assert!(text.contains("\"null\": 2"));
    assert!(text.contains("\"1leading\": 3"));
}

#[test]
fn bare_identifier_label_is_written_unquoted() {
    let raw = RawNode::Edges(vec![(
        "under_score-and-dash".to_string(),
        RawNode::Leaf(crate::document::Scalar::Int((1).into())),
    )]);
    assert_eq!(write_oml(&raw, 2).unwrap(), "under_score-and-dash: 1");
}

// -- comments / separators ---------------------------------------------------

#[test]
fn comments_and_semicolons_are_accepted_as_separators() {
    let node = read_oml("a: 1 # a comment\n; b: 2").unwrap();
    match node {
        RawNode::Edges(edges) => assert_eq!(edges.len(), 2),
        other => panic!("expected edges, got {other:?}"),
    }
}

#[test]
fn crlf_line_endings_are_accepted_as_separators() {
    let node = read_oml("a: 1\r\nb: 2").unwrap();
    match node {
        RawNode::Edges(edges) => assert_eq!(edges.len(), 2),
        other => panic!("expected edges, got {other:?}"),
    }
}

#[test]
fn lone_carriage_return_is_a_stray_character() {
    let err = read_oml("a: 1\rb: 2").unwrap_err();
    assert!(err.message.contains("stray character"));
}

#[test]
fn utf8_bom_is_stripped() {
    let node = read_oml("\u{feff}a: 1").unwrap();
    match node {
        RawNode::Edges(edges) => assert_eq!(edges.len(), 1),
        other => panic!("expected edges, got {other:?}"),
    }
}

#[test]
fn bare_label_starting_with_a_multibyte_unicode_letter_scans_correctly() {
    // Regression for issue #43's byte-offset scanner rewrite: `scan_word`
    // used to assume its first char was exactly 1 byte when computing the
    // rest-of-word scan start -- a bare word beginning with a multi-byte
    // Unicode letter (`é`, 2 bytes in UTF-8) would misalign that offset if
    // the fix regressed. Confirm it reads as a bare-word error (not
    // `null`/`true`/`false`), not silently truncated or panicking.
    let err = read_oml("éxyz").unwrap_err();
    assert!(
        err.message.contains("bare word") && err.message.contains("éxyz"),
        "got {err:?}"
    );
}

#[test]
fn multibyte_content_inside_a_comment_does_not_corrupt_subsequent_scanning() {
    // Regression for issue #43: comment-skipping used to advance by a flat
    // `+= 1` per char, which would land mid-character on multi-byte
    // comment content if the byte-offset rewrite regressed. Confirm a
    // comment containing multi-byte UTF-8 content is skipped cleanly and
    // the following edge still reads correctly.
    let doc = read_oml("# café ☕ comment\na: 1").unwrap();
    match doc {
        RawNode::Edges(edges) => {
            assert_eq!(
                edges,
                vec![("a".to_string(), RawNode::Leaf(Scalar::Int((1).into())))]
            );
        }
        other => panic!("expected edges, got {other:?}"),
    }
}

#[test]
fn multibyte_string_content_round_trips_and_error_after_it_reports_correct_line() {
    // Regression for issue #43: string-body scanning used to advance by a
    // flat `+= 1` per char; multi-byte string content (emoji, accented
    // letters) would misalign the byte offset if that fix regressed.
    // Also confirms `line_col`'s byte-offset line counting still reports
    // the correct line number for an error *after* a line containing
    // multi-byte content.
    let doc = read_oml("a: \"café \u{1F600}\"").unwrap();
    assert_eq!(
        doc,
        RawNode::Edges(vec![(
            "a".to_string(),
            RawNode::Leaf(Scalar::Str("café \u{1F600}".to_string()))
        )])
    );

    let err = read_oml("a: \"café \u{1F600}\"\nb @\n").unwrap_err();
    assert_eq!(err.line, 2, "got {err:?}");
}

#[test]
fn top_level_brace_document_is_equivalent_to_the_bare_edge_list() {
    let a = read_oml("a: 1").unwrap();
    let b = read_oml("{ a: 1 }").unwrap();
    assert_eq!(a, b);
}

#[test]
fn array_syntax_expands_to_repeated_edges() {
    let node = read_oml("tag: [\"x\", \"y\", \"z\"]").unwrap();
    match node {
        RawNode::Edges(edges) => {
            assert_eq!(edges.len(), 3);
            assert!(edges.iter().all(|(l, _)| l == "tag"));
        }
        other => panic!("expected edges, got {other:?}"),
    }
}

#[test]
fn array_syntax_allows_a_trailing_comma() {
    let node = read_oml("tag: [1, 2,]").unwrap();
    match node {
        RawNode::Edges(edges) => assert_eq!(edges.len(), 2),
        other => panic!("expected edges, got {other:?}"),
    }
}

#[test]
fn array_of_brace_subtrees_round_trips() {
    let raw = RawNode::Edges(vec![
        (
            "item".to_string(),
            RawNode::Edges(vec![(
                "v".to_string(),
                RawNode::Leaf(crate::document::Scalar::Int((1).into())),
            )]),
        ),
        (
            "item".to_string(),
            RawNode::Edges(vec![(
                "v".to_string(),
                RawNode::Leaf(crate::document::Scalar::Int((2).into())),
            )]),
        ),
    ]);
    let text = write_oml_compact(&raw).unwrap();
    let parsed = read_oml(&text).unwrap();
    assert_eq!(parsed, raw);
}

// -- Doc-level depth guard, ported from #4's pattern ------------------------

#[test]
fn read_oml_document_depth_guard_via_doc_from_raw() {
    // The reader itself also enforces MAX_DEPTH while parsing nested `{ }`
    // (parse_brace_value checks before descending), independent of
    // write_oml's own guard.
    let mut src = String::new();
    for _ in 0..(document::MAX_DEPTH + 1) {
        src.push_str("a: {");
    }
    src.push('1');
    for _ in 0..(document::MAX_DEPTH + 1) {
        src.push('}');
    }
    let err = read_oml(&src).unwrap_err();
    assert!(err.message.contains("maximum depth"));
}

// -- coverage: multiline string edge cases ----------------------------------

#[test]
fn multiline_string_strips_crlf_opening_newline() {
    let node = read_oml("a: \"\"\"\r\nline one\"\"\"").unwrap();
    match node {
        RawNode::Edges(edges) => match &edges[0].1 {
            RawNode::Leaf(crate::document::Scalar::Str(s)) => assert_eq!(s, "line one"),
            other => panic!("expected a string leaf, got {other:?}"),
        },
        other => panic!("expected edges, got {other:?}"),
    }
}

#[test]
fn multiline_string_decodes_escapes_inside_the_body() {
    let node = read_oml("a: \"\"\"tab:\\there\"\"\"").unwrap();
    match node {
        RawNode::Edges(edges) => match &edges[0].1 {
            RawNode::Leaf(crate::document::Scalar::Str(s)) => assert_eq!(s, "tab:\there"),
            other => panic!("expected a string leaf, got {other:?}"),
        },
        other => panic!("expected edges, got {other:?}"),
    }
}

#[test]
fn multiline_string_allows_a_run_of_more_than_three_quotes_as_content() {
    // A run of 4 quotes: the first 3 close the string, the 4th is literal
    // content of the *next* token attempt -- mirrors the Python reference's
    // `run >= 3` closing rule (only the first 3 are ever "the delimiter").
    // 4 trailing quotes: the first 3 close the string, the 4th becomes a
    // literal `"` appended to the content (mirrors the Python reference).
    let node = read_oml("a: \"\"\"x\"\"\"\"").unwrap();
    match node {
        RawNode::Edges(edges) => match &edges[0].1 {
            RawNode::Leaf(crate::document::Scalar::Str(s)) => assert_eq!(s, "x\""),
            other => panic!("expected a string leaf, got {other:?}"),
        },
        other => panic!("expected edges, got {other:?}"),
    }
}

// -- coverage: surrogate-pair escape decoding -------------------------------

#[test]
fn escaped_surrogate_pair_decodes_to_the_combined_codepoint() {
    let node = read_oml(r#"a: "😀""#).unwrap();
    match node {
        RawNode::Edges(edges) => match &edges[0].1 {
            RawNode::Leaf(crate::document::Scalar::Str(s)) => assert_eq!(s, "\u{1F600}"),
            other => panic!("expected a string leaf, got {other:?}"),
        },
        other => panic!("expected edges, got {other:?}"),
    }
}

// -- coverage: number/date/time near-miss shapes falling back correctly ----

#[test]
fn four_digits_not_followed_by_dash_is_a_plain_integer() {
    let node = read_oml("a: 1234").unwrap();
    match node {
        RawNode::Edges(edges) => {
            assert_eq!(
                edges[0].1,
                RawNode::Leaf(crate::document::Scalar::Int((1234).into()))
            );
        }
        other => panic!("expected edges, got {other:?}"),
    }
}

#[test]
fn date_shape_missing_second_dash_falls_back_and_errors_on_trailing_content() {
    // "2024-01x": the second dash never appears, so try_date gives up on
    // the date/datetime/time interpretation and falls through to a plain
    // integer read of "2024"; the remaining "-01x" then fails to form a
    // valid document -- an error either way, but *not* a date-shape error.
    let err = read_oml("a: 2024-01x").unwrap_err();
    assert!(!err.message.contains("invalid date"));
}

#[test]
fn two_digit_integer_is_not_mistaken_for_a_time() {
    let node = read_oml("a: 12").unwrap();
    match node {
        RawNode::Edges(edges) => {
            assert_eq!(
                edges[0].1,
                RawNode::Leaf(crate::document::Scalar::Int((12).into()))
            );
        }
        other => panic!("expected edges, got {other:?}"),
    }
}

// -- coverage: writer entry points on a bare top-level node -----------------

#[test]
fn write_oml_compact_on_a_bare_leaf_node() {
    let raw = RawNode::Leaf(crate::document::Scalar::Int((5).into()));
    assert_eq!(write_oml(&raw, 2).unwrap(), "5");
    assert_eq!(write_oml_compact(&raw).unwrap(), "5");
}

#[test]
fn write_oml_on_a_bare_temporal_leaf_node() {
    // A whole document that's a single genuinely-temporal leaf (no
    // wrapping edges) -- `write_oml`/`write_oml_compact`'s own
    // `Scalar::Date` top-level arm, distinct from the same variant
    // appearing nested under an edge (already covered elsewhere).
    let raw = RawNode::Leaf(crate::document::Scalar::Date("2024-01-01".to_string()));
    assert_eq!(write_oml(&raw, 2).unwrap(), "2024-01-01");
    assert_eq!(write_oml_compact(&raw).unwrap(), "2024-01-01");
}

#[test]
fn write_oml_pretty_on_multiple_top_level_edges() {
    let raw = RawNode::Edges(vec![
        (
            "a".to_string(),
            RawNode::Leaf(crate::document::Scalar::Int((1).into())),
        ),
        (
            "b".to_string(),
            RawNode::Leaf(crate::document::Scalar::Int((2).into())),
        ),
    ]);
    assert_eq!(write_oml(&raw, 2).unwrap(), "a: 1\nb: 2");
}

// -- coverage: parse-error display / field paths ----------------------------

#[test]
fn parse_array_with_multiple_scalar_elements_and_no_trailing_comma() {
    let node = read_oml("tag: [1, 2, 3]").unwrap();
    match node {
        RawNode::Edges(edges) => assert_eq!(edges.len(), 3),
        other => panic!("expected edges, got {other:?}"),
    }
}

#[test]
fn parse_label_rejects_a_non_label_token() {
    let err = read_oml("a: { 1: 2 }").unwrap_err();
    assert!(err.message.contains("expected a label"));
}

#[test]
fn parse_value_rejects_a_bad_token_where_a_scalar_is_expected() {
    let err = read_oml("a: :").unwrap_err();
    assert!(err.message.contains("expected a value"));
}

#[test]
fn array_close_error_when_neither_comma_nor_bracket() {
    let err = read_oml("a: [1 2]").unwrap_err();
    assert!(err.message.contains("expected ',' or ']'"));
}

// -- coverage: remaining escape / scanner / parser branches -----------------

#[test]
fn string_decodes_solidus_backspace_and_formfeed_escapes() {
    let node = read_oml(r#"a: "\/ \b \f""#).unwrap();
    match node {
        RawNode::Edges(edges) => match &edges[0].1 {
            RawNode::Leaf(crate::document::Scalar::Str(s)) => {
                assert_eq!(s, "/ \u{8} \u{c}");
            }
            other => panic!("expected a string leaf, got {other:?}"),
        },
        other => panic!("expected edges, got {other:?}"),
    }
}

#[test]
fn escaped_surrogate_pair_via_explicit_u_escapes_decodes_correctly() {
    // 😀 is the UTF-16 surrogate pair for U+1F600 (grinning
    // face) -- unlike `escaped_surrogate_pair_decodes_to_the_combined_
    // codepoint` (which embeds the literal character), this exercises the
    // scanner's own `\uXXXX\uYYYY` pairing/combination code path directly:
    // the source text below contains the literal six-character escape
    // sequences, not the pre-combined UTF-8 character.
    let node = read_oml("a: \"\\uD83D\\uDE00\"").unwrap();
    match node {
        RawNode::Edges(edges) => match &edges[0].1 {
            RawNode::Leaf(crate::document::Scalar::Str(s)) => assert_eq!(s, "\u{1F600}"),
            other => panic!("expected a string leaf, got {other:?}"),
        },
        other => panic!("expected edges, got {other:?}"),
    }
}

#[test]
fn lone_minus_not_followed_by_digit_or_inf_is_a_stray_character() {
    let err = read_oml("a: -x").unwrap_err();
    assert!(err.message.contains("stray character '-'"));
}

#[test]
fn decimal_number_with_an_exponent_round_trips() {
    // Not a write-then-read round trip on purpose: Rust's `f64::Display`
    // never emits scientific notation (unlike Python's `repr`), so
    // `write_oml` never actually *produces* a decimal-with-exponent
    // literal (`1.5e10`) -- it always writes the equivalent plain decimal.
    // This is a direct-read test of the NUMDEC-with-exponent scanner
    // branch, which OML-Extended source *can* still contain.
    let node = read_oml("a: 1.5e10").unwrap();
    match node {
        RawNode::Edges(edges) => {
            assert_eq!(
                edges[0].1,
                RawNode::Leaf(crate::document::Scalar::Float(1.5e10))
            );
        }
        other => panic!("expected edges, got {other:?}"),
    }
}

#[test]
fn exponent_only_number_forms_are_recognized() {
    let cases = ["1e10", "1e+10", "1e-10", "-1e5"];
    for src in cases {
        let node = read_oml(&format!("a: {src}")).unwrap();
        match node {
            RawNode::Edges(edges) => {
                assert!(
                    matches!(edges[0].1, RawNode::Leaf(crate::document::Scalar::Float(_))),
                    "{src} should scan as a float"
                );
            }
            other => panic!("expected edges, got {other:?}"),
        }
    }
}

#[test]
fn quoted_string_label_at_the_top_level_is_recognized_as_an_edge() {
    // Exercises looks_like_edge's and parse_label's `Str` arms directly --
    // every other test in this suite uses a bare identifier label.
    let node = read_oml(r#""a b": 1"#).unwrap();
    match node {
        RawNode::Edges(edges) => {
            assert_eq!(edges[0].0, "a b");
            assert_eq!(
                edges[0].1,
                RawNode::Leaf(crate::document::Scalar::Int((1).into()))
            );
        }
        other => panic!("expected edges, got {other:?}"),
    }
}

#[test]
fn reserved_word_at_the_document_top_level_is_read_as_the_scalar_not_a_label() {
    // Exercises `looks_like_edge`'s reserved-word-returns-false arm
    // directly: "null" at the very top of a document is the null scalar
    // (looks_like_edge deliberately excludes reserved words), so parsing
    // stops there and the trailing ": 1" is reported as unexpected
    // trailing content, not a reserved-label error (that error only fires
    // once genuinely in label position -- see
    // `reserved_word_cannot_be_a_bare_label` above).
    let err = read_oml("null: 1").unwrap_err();
    assert!(err.message.contains("unexpected trailing content"));
}

#[test]
fn missing_separator_error_names_a_quoted_string_token() {
    // Exercises `tok_display`'s `Str` arm: the *offending* token in the
    // error message is itself a quoted string, not an identifier/number.
    let err = read_oml(r#"a: 1 "b""#).unwrap_err();
    assert!(err.message.contains("expected a separator"));
    assert!(err.message.contains("\"b\""));
}

// -- issue #75: capitalized / quoted-label boundary cases for the reserved
// spellings (nan/inf/-inf, and null/true/false for symmetry with the other
// two ports) ------------------------------------------------------------

#[test]
fn capitalized_null_true_false_is_a_bare_ident_not_the_keyword() {
    // `RESERVED` (parser.rs) is matched case-sensitively, mirroring Python's
    // `test_capitalized_nan_is_bare_ident_not_keyword` / TS's "capitalized
    // NaN is a bare ident, not the keyword": a capitalized spelling of a
    // reserved word is just an ordinary identifier, valid as a bare label
    // and round-tripping like any other bare ident -- never mistaken for
    // the lowercase keyword it capitalizes.
    for word in ["Null", "True", "False", "NULL", "TRUE", "FALSE"] {
        let raw = RawNode::Edges(vec![(
            word.to_string(),
            RawNode::Leaf(crate::document::Scalar::Int((1).into())),
        )]);
        let text = write_oml(&raw, 2).unwrap();
        assert_eq!(
            text,
            format!("{word}: 1"),
            "expected a bare unquoted label for {word:?}"
        );
        let parsed = read_oml(&text).unwrap();
        assert_eq!(parsed, raw, "round trip mismatch for {word:?}");

        // Also valid, and rejected the same way as any other unquoted
        // string, when used as a *value* -- a capitalized spelling is not
        // recognized as the `null`/`true`/`false` literal.
        let err = read_oml(&format!("a: {word}")).unwrap_err();
        assert!(
            err.message.contains("bare word"),
            "expected a bare-word error for value {word:?}, got {err:?}"
        );
    }
}

#[test]
fn capitalized_nan_inf_is_a_bare_ident_not_the_keyword() {
    // Same case-sensitivity property as above, but for `nan`/`inf`: the
    // scanner's `matches_word` (scanner.rs) is a byte-exact ASCII compare,
    // so `NAN`/`INF`/`NaN`/`Inf` never take the float-keyword scan path at
    // all -- they fall through to ordinary identifier scanning, same as
    // any other bare word.
    for word in ["NAN", "INF", "NaN", "Inf"] {
        let raw = RawNode::Edges(vec![(
            word.to_string(),
            RawNode::Leaf(crate::document::Scalar::Int((1).into())),
        )]);
        let text = write_oml(&raw, 2).unwrap();
        assert_eq!(
            text,
            format!("{word}: 1"),
            "expected a bare unquoted label for {word:?}"
        );
        let parsed = read_oml(&text).unwrap();
        assert_eq!(parsed, raw, "round trip mismatch for {word:?}");

        // As a value, a capitalized spelling is not recognized as the
        // nan/inf float keyword -- it's an unquoted bare word, rejected
        // the same way any other stray identifier value would be.
        let err = read_oml(&format!("a: {word}")).unwrap_err();
        assert!(
            err.message.contains("bare word"),
            "expected a bare-word error for value {word:?}, got {err:?}"
        );
    }

    // `-INF`/`-NaN`/`-Inf`: unlike `NAN`/`INF`, a leading `-` is never part
    // of identifier scanning (scan_minus owns it), and scan_minus only
    // recognizes the exact lowercase `-inf` spelling before falling back to
    // "next char must be a digit" -- so any capitalized `-`-prefixed
    // spelling is a scan-level stray character, not a bare word and not a
    // valid label token at all (it can never even reach parse_label).
    for word in ["-INF", "-NaN", "-Inf"] {
        let err = read_oml(&format!("a: {word}")).unwrap_err();
        assert!(
            err.message.contains("stray character '-'"),
            "expected a stray-character error for {word:?}, got {err:?}"
        );
    }
}

#[test]
fn bare_lowercase_nan_inf_neg_inf_cannot_be_a_label() {
    // Companion to `reserved_word_cannot_be_a_bare_label`: unlike
    // `null`/`true`/`false`, lowercase `nan`/`inf`/`-inf` are *not* in the
    // parser's `RESERVED` list -- the scanner intercepts them as
    // `TokKind::Float` before general identifier scanning ever runs (see
    // `scan_word`/`scan_minus`), so `parse_label` never sees them as an
    // `Ident` and the "reserved word" branch never fires for them. They're
    // still correctly rejected as a label, just via the "expected a label"
    // branch instead -- a real, if subtle, divergence from the
    // null/true/false error message that's worth pinning down explicitly.
    for (word, expected_fragment) in [
        ("nan", "expected a label"),
        ("inf", "expected a label"),
        ("-inf", "expected a label"),
    ] {
        let err = read_oml(&format!("a: {{ {word}: 1 }}")).unwrap_err();
        assert!(
            err.message.contains(expected_fragment),
            "expected {expected_fragment:?} for label {word:?}, got {err:?}"
        );
    }
}

#[test]
fn a_label_starting_with_nan_or_inf_but_continuing_with_underscore_is_a_real_identifier() {
    // Regression: `word_boundary_ok` (scanner.rs) checked
    // alphanumeric-or-'-' to decide whether "nan"/"inf" ends a reserved
    // float keyword, but identifier continuation (`scan_word`'s own loop)
    // also allows '_' -- so a label like `nan_` was wrongly tokenized as
    // the keyword `nan` (TokKind::Float(NAN)) followed by a stray `_`,
    // instead of the single identifier `nan_`. Found by proptest fuzzing
    // (fuzz.rs::oml_round_trips) against a real Value with an object field
    // labeled "nan_".
    for label in ["nan_", "inf_", "nan_x", "infinity_ish"] {
        let doc = read_oml(&format!("a: {{ {label}: 1 }}")).unwrap();
        let RawNode::Edges(edges) = &doc else {
            panic!("expected edges")
        };
        let RawNode::Edges(inner) = &edges[0].1 else {
            panic!("expected edges")
        };
        assert_eq!(inner[0].0, label, "label {label:?} was mis-tokenized");
    }
}

#[test]
fn quoted_reserved_spelling_is_a_valid_label_and_round_trips() {
    // Grammar-doc conformance, mirroring Python's
    // `test_oml_ex9_nan_is_a_number_token_never_a_label` /
    // `test_oml_ex10_quoted_nan_is_a_valid_label` and TS's parametrized
    // reserved-spelling-as-a-label test: quoting any of the reserved
    // spellings -- the null/true/false keywords as well as the nan/inf/-inf
    // number spellings -- turns it into a perfectly ordinary string label,
    // which must round-trip through read -> write -> read unchanged.
    for word in ["null", "true", "false", "nan", "inf", "-inf"] {
        let raw = RawNode::Edges(vec![(
            word.to_string(),
            RawNode::Leaf(crate::document::Scalar::Int((1).into())),
        )]);
        let text = write_oml(&raw, 2).unwrap();
        // None of these are valid bare labels (either RESERVED, or -- for
        // nan/inf -- explicitly excluded in `is_bare_label`, or -- for
        // -inf -- disqualified by the leading `-`), so the writer must
        // quote every one of them.
        assert_eq!(
            text,
            format!("{:?}: 1", word),
            "expected a quoted label for {word:?}"
        );
        let parsed = read_oml(&text).unwrap();
        assert_eq!(parsed, raw, "round trip mismatch for quoted {word:?}");
    }
}

#[test]
fn test_oml_depth_limit_boundary_and_consistency() {
    use crate::oml::{read_oml, write_oml};

    // Exactly at MAX_DEPTH (200): 199 levels of "a: { " wrapping "z: 1", leaf is at depth 200
    let valid_oml = "a: { ".repeat(199) + "z: 1" + &" }".repeat(199);
    let raw = read_oml(&valid_oml).expect("should accept depth 200");
    assert!(write_oml(&raw, 2).is_ok());

    // Exceeds MAX_DEPTH: 200 levels of "a: { " wrapping "z: 1", leaf is at depth 201
    let invalid_oml = "a: { ".repeat(200) + "z: 1" + &" }".repeat(200);
    let err = read_oml(&invalid_oml).unwrap_err();
    assert!(err.to_string().contains("maximum depth"));
}

#[test]
fn test_read_oml_node_count_limit() {
    use crate::document::MAX_NODES;
    use crate::oml::read_oml;

    // At the limit: (MAX_NODES - 1) edges + 1 root node = MAX_NODES
    let at_limit = "a: 0
"
    .repeat(MAX_NODES - 1);
    assert!(read_oml(&at_limit).is_ok());

    // One past the limit: MAX_NODES edges + 1 root node = MAX_NODES + 1
    let past_limit = "a: 0
"
    .repeat(MAX_NODES);
    let err = read_oml(&past_limit).unwrap_err();
    assert!(err.to_string().contains("maximum node count"));
}