yaml_edit/

lex.rs

//! Lexer for YAML files.

/// Whitespace and formatting validation errors
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct WhitespaceError {
    /// The error message
    pub message: String,
    /// The byte range where the error occurred
    pub range: std::ops::Range<usize>,
    /// Error category
    pub category: WhitespaceErrorCategory,
}

/// Categories of whitespace errors
#[derive(Debug, Clone, PartialEq, Eq)]
pub enum WhitespaceErrorCategory {
    /// Tab character used for indentation (forbidden in YAML)
    TabIndentation,
    /// Line too long according to configured limit
    LineTooLong,
    /// Mixed line ending styles
    MixedLineEndings,
    /// Invalid scalar indentation
    InvalidIndentation,
}

/// YAML Concrete Syntax Tree (CST) node types.
///
/// This enum defines all possible node types in the YAML syntax tree, representing both
/// lexical tokens (from the lexer) and semantic nodes (created by the parser).
///
/// # Tree Hierarchy
///
/// The YAML syntax tree follows this general structure:
///
/// ```text
/// ROOT
/// ├── DOCUMENT*
/// │   ├── DIRECTIVE* (optional, e.g., %YAML 1.2)
/// │   ├── DOC_START? (optional ---)
/// │   ├── MAPPING | SEQUENCE | SCALAR | TAGGED_NODE
/// │   └── DOC_END? (optional ...)
/// └── WHITESPACE | NEWLINE | COMMENT (between documents)
///
/// MAPPING
/// ├── MAPPING_ENTRY*
/// │   ├── KEY
/// │   │   └── SCALAR | SEQUENCE | MAPPING (YAML 1.2 allows complex keys)
/// │   ├── COLON
/// │   ├── WHITESPACE?
/// │   └── VALUE
/// │       └── SCALAR | SEQUENCE | MAPPING | TAGGED_NODE
/// ├── NEWLINE
/// ├── INDENT
/// └── COMMENT?
///
/// SEQUENCE  
/// ├── SEQUENCE_ENTRY*
/// │   ├── DASH
/// │   ├── WHITESPACE?
/// │   └── SCALAR | SEQUENCE | MAPPING | TAGGED_NODE
/// ├── NEWLINE
/// ├── INDENT
/// └── COMMENT?
///
/// SCALAR
/// └── STRING | INT | FLOAT | BOOL | NULL
///
/// TAGGED_NODE
/// ├── TAG (e.g., !!str, !custom)
/// ├── WHITESPACE?
/// └── SCALAR | MAPPING | SEQUENCE
/// ```
///
/// # Node Categories
///
/// ## Structural Nodes (created by parser)
/// - **ROOT**: Top-level container for the entire document
/// - **DOCUMENT**: A single YAML document (separated by --- or ...)
/// - **MAPPING**: Key-value pairs `{key: value}` or block style
/// - **SEQUENCE**: Lists `[item1, item2]` or block style with `-`
/// - **SCALAR**: Leaf values (strings, numbers, booleans, null)
/// - **TAGGED_NODE**: Values with explicit type tags `!!str "hello"`
///
/// ## Container Nodes (created by parser)
/// - **MAPPING_ENTRY**: A single key-value pair within a mapping
/// - **SEQUENCE_ENTRY**: A single item within a sequence
/// - **KEY**: The key part of a key-value pair (can contain complex types)
/// - **VALUE**: The value part of a key-value pair
///
/// ## Lexical Tokens (from lexer)
/// - **Punctuation**: COLON, DASH, COMMA, etc.
/// - **Brackets**: LEFT_BRACKET, RIGHT_BRACKET, LEFT_BRACE, RIGHT_BRACE
/// - **Literals**: STRING, INT, FLOAT, BOOL, NULL
/// - **YAML-specific**: TAG, ANCHOR, REFERENCE, MERGE_KEY
/// - **Document markers**: DOC_START (---), DOC_END (...)
/// - **Formatting**: WHITESPACE, NEWLINE, INDENT, COMMENT
///
/// ## Special Cases
///
/// ### Complex Keys (YAML 1.2.2)
/// Keys can be sequences or mappings, not just scalars:
/// ```yaml
/// [1, 2]: value        # Sequence key
/// {a: b}: value        # Mapping key
/// ```
///
/// ### Tagged Values
/// Values can have explicit type information:
/// ```yaml
/// number: !!int "123"  # Force string "123" to be treated as integer
/// binary: !!binary |   # Base64 encoded binary data
///   R0lGODlhDAAMAIQ...
/// ```
///
/// ### Block Scalars
/// Multi-line strings with special parsing rules:
/// ```yaml
/// literal: |           # PIPE indicates literal scalar
///   Line 1
///   Line 2
/// folded: >            # GREATER indicates folded scalar  
///   Long text that
///   gets folded
/// ```
///
/// The tree preserves all original formatting, comments, and whitespace,
/// enabling lossless round-trip parsing and precise source location tracking.
#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)]
#[repr(u16)]
#[allow(non_camel_case_types, clippy::upper_case_acronyms)]
pub enum SyntaxKind {
    // Structural
    /// Root node of the syntax tree
    ROOT = 0,
    /// A YAML document
    DOCUMENT,
    /// A YAML sequence (list)
    SEQUENCE,
    /// A YAML mapping (key-value pairs)
    MAPPING,
    /// A YAML scalar value
    SCALAR,
    /// A YAML alias reference (e.g., *anchor_name)
    ALIAS,
    /// A YAML tagged scalar (tag + value)
    TAGGED_NODE,
    /// Parse error marker
    ERROR,

    // Tokens
    /// Dash character '-'
    DASH,
    /// Plus character '+'
    PLUS,
    /// Colon character ':'
    COLON,
    /// Question mark '?'
    QUESTION,
    /// Left bracket '['
    LEFT_BRACKET,
    /// Right bracket ']'
    RIGHT_BRACKET,
    /// Left brace '{'
    LEFT_BRACE,
    /// Right brace '}'
    RIGHT_BRACE,
    /// Comma ','
    COMMA,
    /// Pipe '|'
    PIPE,
    /// Greater than '>'
    GREATER,
    /// Ampersand '&'
    AMPERSAND,
    /// Asterisk '*'
    ASTERISK,
    /// Exclamation '!'
    EXCLAMATION,
    /// Percent '%'
    PERCENT,
    /// At symbol '@'
    AT,
    /// Backtick '`'
    BACKTICK,
    /// Double quote '"'
    QUOTE,
    /// Single quote "'"
    SINGLE_QUOTE,

    // Document markers
    /// Document start marker '---'
    DOC_START,
    /// Document end marker '...'
    DOC_END,

    // Parser-generated semantic nodes
    /// A mapping key (created by parser from context)
    KEY,
    /// A value in key-value pair (created by parser from context)
    VALUE,
    /// A complete mapping entry (key-value pair with associated tokens)
    MAPPING_ENTRY,
    /// A sequence entry (item with associated tokens)
    SEQUENCE_ENTRY,

    // Content tokens (from lexer)
    /// String literal (quoted or unquoted identifier)
    STRING,
    /// Unterminated string (missing closing quote)
    UNTERMINATED_STRING,
    /// Integer literal
    INT,
    /// Float literal
    FLOAT,
    /// Boolean literal (true/false)
    BOOL,
    /// Null literal
    NULL,
    /// YAML tag like '!tag'
    TAG,
    /// YAML anchor like '&anchor'
    ANCHOR,
    /// YAML reference like '*reference'
    REFERENCE,
    /// YAML merge key '<<'
    MERGE_KEY,
    /// YAML directive like '%YAML 1.2'
    DIRECTIVE,

    // Whitespace and formatting
    /// Spaces and tabs
    WHITESPACE,
    /// Newline characters
    NEWLINE,
    /// Leading whitespace that determines structure
    INDENT,
    /// Comments starting with '#'
    COMMENT,

    // Special
    /// UTF-8 Byte Order Mark (BOM) - U+FEFF at start of a document
    BOM,
    /// End of file marker
    EOF,
}

impl From<SyntaxKind> for rowan::SyntaxKind {
    fn from(kind: SyntaxKind) -> Self {
        Self(kind as u16)
    }
}

/// Decide whether the whitespace run starting at `ws_idx` is internal to a
/// plain scalar or a real terminator. Internal whitespace is followed by
/// more plain-scalar content on the same line; a terminator is followed by
/// end-of-line, end-of-input, a `#` comment marker, a `:` that ends the
/// scalar, or a structural indicator. Quotes and hyphens remain scalar
/// content here; flow delimiters terminate only in flow context.
///
/// Caller must guarantee the character at `ws_idx` is a space or tab.
fn plain_scalar_continues_past_whitespace(input: &str, ws_idx: usize, flow_depth: u32) -> bool {
    let rest = &input[ws_idx..];
    // Find the first non-whitespace char in the run.
    let (offset, next) = match rest.char_indices().find(|(_, c)| *c != ' ' && *c != '\t') {
        Some(v) => v,
        None => return false,
    };
    match next {
        '\n' | '\r' | '#' => false,
        ',' | '[' | ']' | '{' | '}' if flow_depth > 0 => false,
        ':' => !is_colon_a_mapping_indicator(input, ws_idx + offset, flow_depth),
        // Once a scalar has begun these are all content on the next word:
        // the signs, the quotes, the flow indicators, and the node
        // properties and block headers, which are indicators only at the
        // start of a node. So `+ ?: v` is keyed `+ ?`, as `+ -: v` was.
        '-' | '+' | '?' | '|' | '>' | '&' | '*' | '!' | '%' | '\'' | '"' | ',' | '[' | ']'
        | '{' | '}' => true,
        c if is_yaml_special(c) => false,
        _ => true,
    }
}

/// Read an anchor or alias name (`ns-anchor-name`).
///
/// Per YAML 1.2 an anchor name is a run of non-space characters excluding
/// the flow indicators, so `-`, `*`, `:` and the rest are ordinary name
/// characters: saphyr reads `&xT*U---` as the single anchor `xT*U---`.
/// Reading these with the general scalar reader stopped at the first
/// YAML-special character and stranded the remainder.
fn read_anchor_name_from<'a>(
    chars: &mut std::iter::Peekable<std::str::CharIndices<'a>>,
    input: &'a str,
    start_idx: usize,
) -> &'a str {
    let mut end_idx = start_idx;
    while let Some((idx, ch)) = chars.peek().copied() {
        if is_white_or_break(ch) || matches!(ch, ',' | '[' | ']' | '{' | '}') {
            break;
        }
        end_idx = idx + ch.len_utf8();
        chars.next();
    }
    &input[start_idx..end_idx]
}

/// Is a `:` at `input[colon_idx]` a mapping indicator (rather than
/// plain-scalar content)?
///
/// Per YAML 1.2 the answer is context-dependent. `:` is an indicator
/// when:
///   - followed by whitespace or EOF, in any context; or
///   - inside a flow collection, followed by a flow terminator
///     (`,`, `]`, `}`).
///
/// It is *not* an indicator when embedded in a plain scalar (`http://`,
/// `::vector`, timestamp bodies, etc.).
///
/// `colon_idx` must point at the `:` byte itself; only the char that
/// follows is inspected. Callers already in flow context should pass a
/// non-zero `flow_depth`.
fn is_colon_a_mapping_indicator(input: &str, colon_idx: usize, flow_depth: u32) -> bool {
    // Per YAML 1.2 section 7.4 a `:` directly after a JSON-like node -- a
    // closing `}` or `]`, or a quoted scalar -- separates that key from its
    // value in flow context, with no space required: `[ {a: b}:c ]` holds
    // the mapping `{{a: b}: c}`, as the YAML test suite's 9MMW expects.
    if flow_depth > 0
        && input[..colon_idx]
            .chars()
            .next_back()
            .is_some_and(|ch| matches!(ch, '}' | ']' | '"' | '\''))
    {
        return true;
    }
    is_colon_followed_by_separator(input, colon_idx, flow_depth)
}

/// Is the `:` at `input[colon_idx]` followed by what makes it a mapping
/// indicator after content that is not a JSON-like node, such as a plain
/// scalar's own text?
fn is_colon_followed_by_separator(input: &str, colon_idx: usize, flow_depth: u32) -> bool {
    let after = input[colon_idx + 1..].chars().next();
    match after {
        None => true,
        Some(ch) if is_white_or_break(ch) => true,
        Some(ch) if flow_depth > 0 && matches!(ch, ',' | ']' | '}') => true,
        _ => false,
    }
}

/// Is a `#` at `input[hash_idx]` the start of a comment (rather than
/// scalar content)?
///
/// Per YAML 1.2 section 6.6, `#` starts a comment only when preceded
/// by whitespace (or is at the start of the input / a fresh line).
/// Inside a plain scalar with no preceding whitespace, `#` is scalar
/// content (URLs, `#frag`, etc.).
fn is_hash_a_comment_start(input: &str, hash_idx: usize) -> bool {
    hash_idx == 0
        || input[..hash_idx]
            .chars()
            .next_back()
            .is_some_and(is_white_or_break)
}

/// Scan a double-quoted string body, consuming through the closing quote.
///
/// Returns the end offset and whether a closing quote was found; a backslash
/// escapes the next character, so `"a\""` runs to the second `"`.
fn scan_double_quoted(
    chars: &mut std::iter::Peekable<std::str::CharIndices<'_>>,
    start_idx: usize,
) -> (usize, bool) {
    let mut end_idx = start_idx + 1;
    let mut escaped = false;

    while let Some((idx, ch)) = chars.peek() {
        let (current_idx, current_ch) = (*idx, *ch);
        end_idx = current_idx + current_ch.len_utf8();
        chars.next();

        if escaped {
            escaped = false;
        } else if current_ch == '\\' {
            escaped = true;
        } else if current_ch == '"' {
            return (end_idx, true);
        }
    }
    (end_idx, false)
}

/// Scan a single-quoted string body, consuming through the closing quote.
///
/// Returns the end offset and whether a closing quote was found. A doubled
/// `''` is an escaped quote and does not end the string.
fn scan_single_quoted(
    chars: &mut std::iter::Peekable<std::str::CharIndices<'_>>,
    start_idx: usize,
) -> (usize, bool) {
    let mut end_idx = start_idx + 1;

    while let Some((idx, ch)) = chars.peek() {
        let (current_idx, current_ch) = (*idx, *ch);
        end_idx = current_idx + current_ch.len_utf8();
        chars.next();

        if current_ch == '\'' {
            if let Some((next_idx, '\'')) = chars.peek() {
                // Doubled quote: an escaped `'`, not the end.
                end_idx = *next_idx + 1;
                chars.next();
            } else {
                return (end_idx, true);
            }
        }
    }
    (end_idx, false)
}

/// Whether the flow collection opening at `open_idx` starts an implicit
/// key: it closes on its line, with the brackets matching, and a `:`
/// mapping indicator follows on that line.
fn flow_collection_is_implicit_key(input: &str, open_idx: usize) -> bool {
    let line = input[open_idx..]
        .split(['\n', '\r'])
        .next()
        .unwrap_or_default();
    let mut chars = line.char_indices().peekable();
    let mut closers = Vec::new();
    let mut prev = ' ';
    while let Some((idx, ch)) = chars.next() {
        match ch {
            '[' => closers.push(']'),
            '{' => closers.push('}'),
            ']' | '}' => {
                if closers.pop() != Some(ch) {
                    return false;
                }
                if closers.is_empty() {
                    // The `:` need not follow straight away: `[q] x: v` is a
                    // key holding two nodes, which the validator reports.
                    let rest = &line[idx + 1..];
                    let comment = [" #", "\t#"].iter().filter_map(|c| rest.find(c)).min();
                    return rest[..comment.unwrap_or(rest.len())]
                        .match_indices(':')
                        .any(|(i, _)| is_colon_a_mapping_indicator(line, idx + 1 + i, 0));
                }
            }
            // A quote mid-scalar (`[a'b]`) is content.
            '"' | '\'' if matches!(prev, ' ' | '\t' | '[' | '{' | ',' | ':') => {
                let (_, closed) = if ch == '"' {
                    scan_double_quoted(&mut chars, idx)
                } else {
                    scan_single_quoted(&mut chars, idx)
                };
                if !closed {
                    return false;
                }
            }
            '#' if matches!(prev, ' ' | '\t') => return false,
            _ => {}
        }
        prev = ch;
    }
    false
}

/// Whether a `-` or `+` following `line_prefix`, the text of its line before
/// it, is a block scalar header's chomping indicator: the last character
/// other than an indentation digit is the header's `|` or `>`.
fn is_chomping_indicator(line_prefix: &str) -> bool {
    line_prefix
        .bytes()
        .rev()
        .find(|b| !b.is_ascii_digit())
        .is_some_and(|b| b == b'|' || b == b'>')
}

/// Tokenize YAML input with whitespace validation
pub fn lex(input: &str) -> Vec<(SyntaxKind, &str)> {
    let (tokens, _) = lex_with_validation(input);
    tokens
}

/// Configuration for whitespace and formatting validation
pub struct ValidationConfig {
    /// Maximum line length (None = no limit)
    pub max_line_length: Option<usize>,
    /// Whether to enforce consistent line endings
    pub enforce_consistent_line_endings: bool,
}

impl Default for ValidationConfig {
    fn default() -> Self {
        Self {
            max_line_length: Some(120), // Default to 120 characters
            enforce_consistent_line_endings: true,
        }
    }
}

/// Tokenize YAML input with whitespace and formatting validation
pub fn lex_with_validation(input: &str) -> (Vec<(SyntaxKind, &str)>, Vec<WhitespaceError>) {
    lex_with_validation_config(input, &ValidationConfig::default())
}

/// Record a "line too long" error if the line ending at `end` exceeds the
/// configured maximum.
fn check_line_length(
    errors: &mut Vec<WhitespaceError>,
    config: &ValidationConfig,
    line_start: usize,
    end: usize,
) {
    let Some(max_len) = config.max_line_length else {
        return;
    };
    let line_length = end - line_start;
    if line_length > max_len {
        errors.push(WhitespaceError {
            message: format!("Line too long ({line_length} > {max_len} characters)"),
            range: line_start..end,
            category: WhitespaceErrorCategory::LineTooLong,
        });
    }
}

/// Record the line ending style on first sight, and flag any later one that
/// disagrees with it.
fn note_line_ending<'a>(
    errors: &mut Vec<WhitespaceError>,
    config: &ValidationConfig,
    detected: &mut Option<&'a str>,
    line_ending: &'a str,
    range: std::ops::Range<usize>,
) {
    if !config.enforce_consistent_line_endings {
        return;
    }
    match *detected {
        Some(seen) if seen != line_ending => errors.push(WhitespaceError {
            message: "Inconsistent line endings detected".to_string(),
            range,
            category: WhitespaceErrorCategory::MixedLineEndings,
        }),
        Some(_) => {}
        None => *detected = Some(line_ending),
    }
}

/// Tokenize YAML input with custom validation configuration
pub fn lex_with_validation_config<'a>(
    input: &'a str,
    config: &ValidationConfig,
) -> (Vec<(SyntaxKind, &'a str)>, Vec<WhitespaceError>) {
    let mut state = LexState::new(input, config);
    while let Some((start_idx, ch)) = state.chars.next() {
        state.lex_char(start_idx, ch);
    }
    state.finish()
}

/// The lexer's state while it walks the input.
struct LexState<'a, 'c> {
    input: &'a str,
    config: &'c ValidationConfig,
    chars: std::iter::Peekable<std::str::CharIndices<'a>>,
    tokens: Vec<(SyntaxKind, &'a str)>,
    errors: Vec<WhitespaceError>,
    /// Offset of the start of the current line.
    line_start: usize,
    detected_line_ending: Option<&'a str>,
    /// Depth of the flow collections the lexer is inside.
    flow_depth: u32,
    /// The block scalar whose body is still running. All of the body is
    /// literal text, so a `{` or `[` there must not open a flow collection:
    /// the depth would never come back down and every later `,` in the file
    /// would lex as a delimiter.
    block_scalar: Option<BlockScalarBody>,
    /// Number of tokens when the last block scalar body ended. The text
    /// tokens before it may be body lines, which no line continues.
    block_scalar_end: usize,
}

impl<'a, 'c> LexState<'a, 'c> {
    fn new(input: &'a str, config: &'c ValidationConfig) -> Self {
        Self {
            input,
            config,
            chars: input.char_indices().peekable(),
            tokens: Vec::with_capacity(input.len() / 8), // Pre-allocate based on estimate
            errors: Vec::new(),
            line_start: 0,
            detected_line_ending: None,
            flow_depth: 0,
            block_scalar: None,
            block_scalar_end: 0,
        }
    }

    /// Push a token for `input[start..end]`.
    fn push(&mut self, kind: SyntaxKind, start: usize, end: usize) {
        self.tokens.push((kind, &self.input[start..end]));
    }

    /// Push `input[start..end]` as a plain scalar token.
    fn push_scalar(&mut self, start: usize, end: usize) {
        let text = &self.input[start..end];
        self.tokens.push((classify_scalar(text), text));
    }

    /// Push a quoted-string token, marking it UNTERMINATED_STRING when the
    /// closing quote was never reached.
    fn push_quoted_string(&mut self, start: usize, end: usize, found_closing: bool) {
        let kind = if found_closing {
            SyntaxKind::STRING
        } else {
            SyntaxKind::UNTERMINATED_STRING
        };
        self.push(kind, start, end);
    }

    /// Read a plain scalar that starts at `start`, whose body begins at
    /// `body_start`, classify it, and push it as one token.
    ///
    /// The characters before `body_start` have already been consumed by the
    /// caller (`-`, `+`, `.`, ...) and stay part of the scalar text.
    fn push_plain_scalar(&mut self, start: usize, body_start: usize, absorb_spaces: bool) {
        let end = self.read_plain_scalar_body(body_start, absorb_spaces, false);
        self.push_scalar(start, end);
    }

    /// Read a plain-scalar body starting at `start_idx`, returning where it
    /// ends. `:` is content unless it is a mapping indicator, `#` unless it
    /// starts a comment, and the indicators that only count at the start of
    /// a node are content here.
    ///
    /// `absorb_spaces` folds an internal whitespace run into the scalar when
    /// more scalar content follows, as a plain scalar spanning several words
    /// requires. A block scalar's chomping indicator (`|-`, `>-`) is a token
    /// in its own right, so that caller passes false.
    ///
    /// `stop_at_chomping_plus` ends the body at a `+` after nothing but
    /// digits, for a scalar that is a block scalar header's indentation
    /// indicator, so that `|2+` keeps the `+` as its own token.
    fn read_plain_scalar_body(
        &mut self,
        start_idx: usize,
        absorb_spaces: bool,
        stop_at_chomping_plus: bool,
    ) -> usize {
        let input = self.input;
        let flow_depth = self.flow_depth;
        let mut end_idx = start_idx;
        while let Some((idx, ch)) = self.chars.peek().copied() {
            // Intra-line whitespace belongs to the scalar when more scalar
            // content follows on the same line. Without this a space ended
            // the body, so `-{ [a]: v }` lexed `[` as a flow collection
            // rather than scalar content.
            if absorb_spaces && (ch == ' ' || ch == '\t') {
                if !plain_scalar_continues_past_whitespace(input, idx, flow_depth) {
                    break;
                }
                while let Some((wi, wc)) = self.chars.peek().copied() {
                    if wc != ' ' && wc != '\t' {
                        break;
                    }
                    end_idx = wi + wc.len_utf8();
                    self.chars.next();
                }
                continue;
            }
            if is_white_or_break(ch) {
                break;
            }
            // What comes before the `:` is this scalar's text, even a quote
            // in it: `[p":a]` holds the scalar `p":a`.
            if ch == ':' && is_colon_followed_by_separator(input, idx, flow_depth) {
                break;
            }
            if ch == '#' && is_hash_a_comment_start(input, idx) {
                break;
            }
            if stop_at_chomping_plus
                && ch == '+'
                && input[start_idx..idx].bytes().all(|b| b.is_ascii_digit())
            {
                break;
            }
            // `-` and `+` are indicators only where a node starts (a sequence
            // entry, or a block-scalar chomping suffix); inside a body they
            // are content, so `++` is the single scalar `++`.
            // `?`, `|`, `>`, `&`, `*`, `!` and `%` are indicators only at the
            // start of a node (YAML 1.2 section 7.1), so they are content
            // here too: `+?`, `a|b` and `a: b%c` are each one scalar.
            if is_yaml_special_except(ch, "-+:#'\"?|>&*!%")
                && !(flow_depth == 0 && matches!(ch, ',' | '[' | ']' | '{' | '}'))
            {
                break;
            }
            end_idx = idx + ch.len_utf8();
            self.chars.next();
        }
        end_idx
    }

    /// Whether a node, and so a node property, block scalar header or
    /// explicit key, may start here. Nothing starts inside a block scalar's
    /// body, which is literal text: `a: |\n  >'\nb: ''\n` holds `>'`, where
    /// reading the `>` as a header left the quote to open a quoted scalar
    /// that ran on into `b`.
    fn node_can_start(&self) -> bool {
        !in_block_scalar_body(self.block_scalar, self.line_start)
            && node_property_can_start(&self.tokens)
    }

    /// Whether a tag, anchor or alias may start at `idx`. A line continuing
    /// a plain scalar is inside that node already, so there they are
    /// content, read by plain-scalar rules: as a tag `!b:` in
    /// `{ a\n  !b: 1 }` took the `:` that follows the key.
    fn property_can_start(&self, idx: usize) -> bool {
        self.node_can_start() && !self.opens_plain_scalar_continuation_line(idx)
    }

    /// Whether a block scalar header may start at `idx`.
    fn block_scalar_header_can_start(&self, idx: usize) -> bool {
        self.flow_depth == 0
            && self.node_can_start()
            && !self.opens_plain_scalar_continuation_line(idx)
    }

    /// Whether the character at `idx` opens a line following a plain scalar.
    ///
    /// Such a line either continues that scalar or starts a new implicit
    /// key, and an implicit key fits on one line.
    /// In block context the line has to be indented past the collection
    /// the scalar is in: at its column a line starts the next entry, as in
    /// `a: 1\n'b: 2\n`.
    fn opens_plain_scalar_continuation_line(&self, idx: usize) -> bool {
        use SyntaxKind::*;
        self.input[self.line_start..idx]
            .chars()
            .all(|c| c == ' ' || c == '\t')
            && (self.flow_depth > 0
                || parent_column(&self.tokens)
                    .map_or(true, |column| idx - self.line_start > column))
            && self
                .tokens
                .iter()
                .enumerate()
                .rev()
                .find(|(_, (kind, _))| !matches!(kind, WHITESPACE | INDENT | NEWLINE))
                .is_some_and(|(i, (kind, _))| {
                    i >= self.block_scalar_end
                        && matches!(kind, STRING | INT | FLOAT | BOOL | NULL | MERGE_KEY)
                })
    }

    /// Lex the token starting with `ch`, which has just been consumed.
    fn lex_char(&mut self, start_idx: usize, ch: char) {
        use SyntaxKind::*;

        match ch {
            // A BOM is not content and takes no column, so the line starts
            // after it.
            '\u{FEFF}'
                if self.flow_depth == 0
                    && start_idx == self.line_start
                    && bom_starts_document(&self.tokens, self.input, start_idx) =>
            {
                self.push(BOM, start_idx, start_idx + ch.len_utf8());
                self.line_start = start_idx + ch.len_utf8();
            }
            // Context-aware hyphen handling
            '-' => self.lex_dash(start_idx),
            '+' => self.lex_plus(start_idx),
            ':' => self.lex_colon(start_idx),
            // An explicit-key indicator, like the node properties below, only
            // at the start of a node: `a ?b` is the scalar `a ?b`, not a key
            // indicator inside it, and so is `a\n? b`.
            //
            // Per YAML 1.2 `c-complex-mapping-key` the `?` must also be
            // followed by a space or a line break to be an indicator. Without
            // that test `a: ?!!r{2}x` opened an explicit key and then read
            // `!!r` as a tag, which stranded the rest of the line; PyYAML
            // reads the value as the plain scalar `?!!r{2}x`.
            //
            // In a flow collection a `?` before a flow indicator cannot
            // start a plain scalar (`ns-plain-first`), so `{?, a}` has an
            // empty explicit key, as PyYAML reads it. The validator reports
            // the space missing after it.
            '?' if self.property_can_start(start_idx)
                && self.chars.peek().map_or(true, |(_, next)| {
                    is_white_or_break(*next)
                        || (self.flow_depth > 0 && matches!(next, ',' | '[' | ']' | '{' | '}'))
                }) =>
            {
                self.push(QUESTION, start_idx, start_idx + 1)
            }
            // A flow indicator inside a block scalar's body is literal
            // text, so it neither opens nor closes a collection. Read the
            // rest of the line as the scalar content it is.
            //
            // The same goes for a `#`: the body is taken verbatim, so
            // `- >\n  # detected\n` holds the content `# detected` rather
            // than a comment (test suite 4QFQ). A line that does not reach
            // the body's column has left it, where a comment is a comment
            // again (T26H).
            '[' | ']' | '{' | '}' | ',' | '#'
                if in_block_scalar_body(self.block_scalar, self.line_start) =>
            {
                self.push_plain_scalar(start_idx, start_idx + 1, true);
            }
            // So is one opening a line after a plain scalar, unless it opens
            // a collection that is an implicit key: opening a collection in
            // `f\n[\n---\n` hid the `---` that ends the document, as one
            // after a `]` did in `f\n]{\n---\n` and after `[]` in
            // `f\n[][\n---\n`.
            '[' | '{' | ']' | '}'
                if self.flow_depth == 0
                    && self.opens_plain_scalar_continuation_line(start_idx)
                    && !(matches!(ch, '[' | '{')
                        && flow_collection_is_implicit_key(self.input, start_idx)) =>
            {
                self.push_plain_scalar(start_idx, start_idx + 1, true);
            }
            '[' => {
                self.flow_depth += 1;
                self.push(LEFT_BRACKET, start_idx, start_idx + 1);
            }
            ']' => {
                self.flow_depth = self.flow_depth.saturating_sub(1);
                self.push(RIGHT_BRACKET, start_idx, start_idx + 1);
            }
            '{' => {
                self.flow_depth += 1;
                self.push(LEFT_BRACE, start_idx, start_idx + 1);
            }
            '}' => {
                self.flow_depth = self.flow_depth.saturating_sub(1);
                self.push(RIGHT_BRACE, start_idx, start_idx + 1);
            }
            // A `,` separates entries only inside a flow collection. In block
            // context it is ordinary plain-scalar content (`a: x\n  ,y`), so
            // fall through to the catch-all arm, which already keeps flow
            // indicators inside a block scalar.
            ',' if self.flow_depth > 0 => self
                .tokens
                .push((COMMA, &self.input[start_idx..start_idx + 1])),
            // A block-scalar header, like the node properties below, only at
            // the start of a node: `a|b` is the scalar `a|b`, not a header
            // inside it. `>` reaches this arm only at a node start already,
            // because the catch-all treats it as scalar content.
            //
            // And only in block context: YAML 1.2 section 8.1 gives block
            // scalars a block-context production only, so inside a flow
            // collection a `|` is plain content and `[|, x]` has two
            // entries, as saphyr reads it.
            //
            // A line after a plain scalar continues it, as a block scalar
            // cannot be an implicit key: `1\n|x\n  # t\n` is `1 |x`.
            '|' if self.block_scalar_header_can_start(start_idx) => {
                self.block_scalar = Some(BlockScalarBody::new(
                    &self.tokens,
                    self.input,
                    self.line_start,
                ));
                self.push(PIPE, start_idx, start_idx + 1)
            }
            '>' if self.block_scalar_header_can_start(start_idx) => {
                self.block_scalar = Some(BlockScalarBody::new(
                    &self.tokens,
                    self.input,
                    self.line_start,
                ));
                self.push(GREATER, start_idx, start_idx + 1)
            }
            // `<<` is a merge key only when it is the whole plain scalar, per
            // the YAML 1.1 merge type: `<<foo`, `<<#` and `<< x` are scalars
            // of their own.
            '<' => {
                self.lex_plain_scalar(start_idx, ch);
                if let Some(token) = self.tokens.last_mut().filter(|t| t.1 == "<<") {
                    token.0 = MERGE_KEY;
                }
            }
            // Like tags, only at the start of a node: `a: x &anc` is the
            // scalar `x &anc`, not an anchored value.
            '&' if self.property_can_start(start_idx) => {
                // Check if this is an anchor definition
                let name = read_anchor_name_from(&mut self.chars, self.input, start_idx + 1);
                if !name.is_empty() {
                    self.push(ANCHOR, start_idx, start_idx + 1 + name.len());
                } else {
                    self.push(AMPERSAND, start_idx, start_idx + 1);
                }
            }
            '*' if self.property_can_start(start_idx) => {
                // Check if this is an alias reference
                let name = read_anchor_name_from(&mut self.chars, self.input, start_idx + 1);
                if !name.is_empty() {
                    self.push(REFERENCE, start_idx, start_idx + 1 + name.len());
                } else {
                    self.push(ASTERISK, start_idx, start_idx + 1);
                }
            }
            // A quote is literal text in a block scalar's body, and content on
            // a line continuing a plain scalar, as the flow indicators above:
            // `a: b\n  'c\nd: 1\n` holds `b 'c`, where scanning it as quoted
            // ran on to the next quote in the file, and `[b\n "x", c]` holds
            // `b "x"`.
            '"' | '\''
                if in_block_scalar_body(self.block_scalar, self.line_start)
                    || self.opens_plain_scalar_continuation_line(start_idx) =>
            {
                self.push_plain_scalar(start_idx, start_idx + 1, true);
            }
            '"' => {
                let (end_idx, found_closing) = scan_double_quoted(&mut self.chars, start_idx);
                self.push_quoted_string(start_idx, end_idx, found_closing);
            }
            '\'' => {
                let (end_idx, found_closing) = scan_single_quoted(&mut self.chars, start_idx);
                self.push_quoted_string(start_idx, end_idx, found_closing);
            }

            // Document end
            '.' => self.lex_dot(start_idx),

            // Comments.
            //
            // Per YAML 1.2 section 6.6, `#` starts a comment only when
            // preceded by whitespace. At outer-token boundary we have
            // one of two situations:
            //   - `#` at the start of the document / a fresh line, or
            //     after a whitespace/newline: legitimate comment.
            //   - `#` glued to the tail of a non-whitespace token
            //     (e.g., `]#foo`): the input violates 4.6.6. The
            //     validator's check_comment_token_whitespace flags
            //     this at validation time; the lexer still emits a
            //     COMMENT token so downstream error reporting has
            //     something to point at.
            //
            // A `#` glued to plain-scalar content never gets here: the
            // scalar reader takes it in (`a#: v`, `:#: v`).
            '#' => self.lex_comment(start_idx),

            // Tags
            //
            // Only at the start of a node: mid-scalar a `!` is ordinary
            // content (`a: x !!b` is the scalar `x !!b`), which the
            // catch-all arm below reads with plain-scalar rules.
            '!' if self.property_can_start(start_idx) => self.lex_tag(start_idx),

            '%' => self.lex_percent(start_idx),

            // Newlines
            '\n' | '\r' => self.lex_line_break(start_idx, ch),

            // Whitespace (spaces and tabs)
            ' ' | '\t' => self.lex_whitespace(start_idx, ch),

            // Everything else is scalar content
            _ => self.lex_plain_scalar(start_idx, ch),
        }
    }

    fn lex_dash(&mut self, start_idx: usize) {
        use SyntaxKind::*;
        if self.flow_depth == 0
            && start_idx == self.line_start
            && self.input[start_idx..].starts_with("---")
            && self.input[start_idx + 3..]
                .chars()
                .next()
                .map_or(true, ends_document_marker)
        {
            self.chars.next();
            self.chars.next();
            self.push(DOC_START, start_idx, start_idx + 3);
        } else {
            // Check if this hyphen should be treated as a sequence marker
            // It's a sequence marker if:
            // 1. It's at the beginning of a line (after optional indentation)
            // 2. OR it follows a value context (after ? or : plus whitespace)
            // AND it's followed by whitespace or end of input

            // Check if preceded only by whitespace from start of line
            let line_start_pos = self.line_start;
            let before_dash = &self.input[line_start_pos..start_idx];
            let only_whitespace_before = before_dash.chars().all(|c| c == ' ' || c == '\t');

            // Check if the previous non-whitespace token was ? or :
            // indicating a value context where sequences are allowed.
            //
            // A DASH counts too: an entry's own node may be a nested
            // sequence, so `- - x` is `[[x]]`, as saphyr and PyYAML
            // both read it and as value.rs already writes it.
            let after_value_indicator = self
                .tokens
                .iter()
                .rev()
                .find(|(kind, _)| !matches!(kind, WHITESPACE | INDENT))
                .is_some_and(|(kind, _)| matches!(kind, QUESTION | COLON | DASH | DOC_START));

            // Check if followed by whitespace or end of input
            let followed_by_whitespace_or_end = self
                .chars
                .peek()
                .map_or(true, |(_, next_ch)| is_white_or_break(*next_ch));

            // A line continuing a plain scalar holds content: `e\n- [` is
            // the scalar `e - [`.
            let is_sequence_marker = ((only_whitespace_before
                && !self.opens_plain_scalar_continuation_line(start_idx))
                || after_value_indicator)
                && followed_by_whitespace_or_end;

            if is_sequence_marker {
                self.push(DASH, start_idx, start_idx + 1);
            } else {
                // This hyphen is part of a scalar value. Use the
                // plain-scalar body reader so embedded `:` (not
                // followed by whitespace) stays inside the scalar.
                //
                // A `-` closing a block-scalar header (`|-`, `>2-`)
                // is a chomping indicator and stands alone, so it
                // must not swallow the space that follows it. Same
                // test as the `+` arm below.
                self.push_plain_scalar(
                    start_idx,
                    start_idx + 1,
                    !is_chomping_indicator(&self.input[self.line_start..start_idx]),
                );
            }
        }
    }

    fn lex_plus(&mut self, start_idx: usize) {
        // `+` is a chomping indicator when it closes a block-scalar header
        // (`|+`, `|2+`, `>+`, `>2+`). Anywhere else it is content, so it
        // starts a plain scalar: `+.INF` is one FLOAT, `+ x: v` is keyed
        // `+ x`, and `+: v` is keyed `+`. A bare PLUS token no parse rule
        // claims would leave the `+` invisible.
        if is_chomping_indicator(&self.input[self.line_start..start_idx]) {
            self.push(SyntaxKind::PLUS, start_idx, start_idx + 1);
        } else {
            self.push_plain_scalar(start_idx, start_idx + 1, true);
        }
    }

    fn lex_colon(&mut self, start_idx: usize) {
        use SyntaxKind::*;
        // Colon at token boundary. It's a mapping indicator when
        // its lookahead matches is_colon_a_mapping_indicator, OR
        // when we're in flow context and the immediately
        // preceding meaningful token was a JSON-like key (a quoted
        // scalar or a flow collection), which YAML 1.2 lets an
        // adjacent value follow:
        //     { "foo"\n  :bar }    -- COLON after "foo"
        // vs.
        //     [ ::vector ]         -- `:` starts a plain scalar
        //     [ a\n :b ]           -- `:b` continues the scalar `a`
        let preceded_by_json_key = self.flow_depth > 0
            && self
                .tokens
                .iter()
                .rev()
                .find(|(k, _)| !matches!(k, WHITESPACE | INDENT | NEWLINE | COMMENT))
                .is_some_and(|(k, text)| {
                    matches!(k, RIGHT_BRACKET | RIGHT_BRACE)
                        || (*k == STRING && text.starts_with(['"', '\'']))
                });
        if preceded_by_json_key
            || is_colon_a_mapping_indicator(self.input, start_idx, self.flow_depth)
        {
            self.push(COLON, start_idx, start_idx + 1);
        } else {
            // This colon starts (or continues) a plain scalar such
            // as a URL, `::vector`, or a timestamp. The shared body
            // reader knows the rules: a `:` not followed by space is
            // content, flow indicators are content in block context,
            // and a `#` glued to content is not a comment.
            self.push_plain_scalar(start_idx, start_idx + 1, true);
        }
    }

    fn lex_dot(&mut self, start_idx: usize) {
        use SyntaxKind::*;
        // Check for three dots (document end marker)
        // `...` is the document-end marker only when it is the whole
        // line, in block context: the same three conditions the `---`
        // arm applies. `...w` is the plain scalar `...w`, as both
        // saphyr and PyYAML read it, and `---x` already was.
        let is_doc_end = self.flow_depth == 0
            && start_idx == self.line_start
            && self.input[start_idx..].starts_with("...")
            && self.input[start_idx + 3..]
                .chars()
                .next()
                .map_or(true, ends_document_marker);
        if is_doc_end && self.chars.peek() == Some(&(start_idx + 1, '.')) {
            self.chars.next(); // consume second .
            if self.chars.peek() == Some(&(start_idx + 2, '.')) {
                self.chars.next(); // consume third .
                self.push(DOC_END, start_idx, start_idx + 3);
            } else {
                // Two dots -- continue as plain scalar body (allows
                // embedded `-` and `:` per YAML plain-scalar rules).
                self.push_plain_scalar(start_idx, start_idx + 2, true);
            }
        } else {
            // Single dot -- part of plain scalar body.
            self.push_plain_scalar(start_idx, start_idx + 1, true);
        }
    }

    fn lex_comment(&mut self, start_idx: usize) {
        use SyntaxKind::*;
        let mut end_idx = start_idx + 1;
        while let Some((idx, ch)) = self.chars.peek() {
            if *ch == '\n' || *ch == '\r' {
                break;
            }
            end_idx = *idx + ch.len_utf8();
            self.chars.next();
        }
        self.push(COMMENT, start_idx, end_idx);
    }

    fn lex_tag(&mut self, start_idx: usize) {
        use SyntaxKind::*;
        // Handle tag indicators - both ! and !!
        let mut end_idx = start_idx + 1;

        // A verbatim tag `!<uri>` runs to its closing `>` and may hold
        // any URI character, commas included, so take it whole.
        if let Some((_, '<')) = self.chars.peek() {
            self.chars.next(); // consume '<'
            end_idx = start_idx + 2;
            // An unterminated `!<x` stops at the line break, which
            // the main loop then handles; the text still round-trips
            // and the parser reports the malformed tag.
            while let Some((idx, ch)) = self.chars.peek() {
                if matches!(ch, '\n' | '\r') {
                    break;
                }
                end_idx = *idx + ch.len_utf8();
                let closing = *ch == '>';
                self.chars.next();
                if closing {
                    break;
                }
            }
            self.push(TAG, start_idx, end_idx);
            return;
        }

        // Check for double exclamation (global tag)
        if let Some((_, '!')) = self.chars.peek() {
            self.chars.next(); // consume the second !
            end_idx = start_idx + 2;
        } else {
            // A named handle `!e!` is a run of word characters
            // closed by another `!`.
            let word = self.input[start_idx + 1..]
                .find(|c: char| !(c.is_ascii_alphanumeric() || c == '-'))
                .map_or(self.input.len(), |i| start_idx + 1 + i);
            if word > start_idx + 1 && self.input[word..].starts_with('!') {
                while self.chars.peek().is_some_and(|(idx, _)| *idx <= word) {
                    self.chars.next();
                }
                end_idx = word + 1;
            }
        }

        // Read the tag name after the ! or !!
        while let Some((idx, ch)) = self.chars.peek() {
            if !is_tag_char(*ch) {
                break;
            }
            end_idx = *idx + ch.len_utf8();
            self.chars.next();
        }

        self.push(TAG, start_idx, end_idx);
    }

    fn lex_percent(&mut self, start_idx: usize) {
        use SyntaxKind::*;
        // A directive is only a `%` that opens a line. Anywhere else,
        // inside a flow collection, or in a block scalar's body
        // (`--- |\n%!PS-Adobe-2.0\n`), it is ordinary scalar content.
        if self.flow_depth > 0
            || start_idx != self.line_start
            || in_block_scalar_body(self.block_scalar, self.line_start)
        {
            // Treat as part of a plain scalar. The shared body reader
            // knows the rules this needs: flow indicators are content
            // in block context, and an internal whitespace run stays
            // in the scalar when more content follows on the line, so
            // `r % {` is the single scalar `r % {`.
            self.push_plain_scalar(start_idx, start_idx + 1, true);
        } else {
            // At the start of a line in block context, % starts a directive
            let mut end_idx = start_idx + 1;
            while let Some((idx, ch)) = self.chars.peek() {
                if *ch == '\n' || *ch == '\r' {
                    break;
                }
                end_idx = *idx + ch.len_utf8();
                self.chars.next();
            }
            self.push(DIRECTIVE, start_idx, end_idx);
        }
    }

    /// Lex a line break: `\n`, `\r\n` or a lone `\r`.
    fn lex_line_break(&mut self, start_idx: usize, ch: char) {
        check_line_length(&mut self.errors, self.config, self.line_start, start_idx);

        let (line_ending, end_pos) = if ch == '\n' {
            ("\n", start_idx + 1)
        } else if let Some((_, '\n')) = self.chars.peek() {
            self.chars.next();
            ("\r\n", start_idx + 2)
        } else {
            ("\r", start_idx + 1)
        };

        note_line_ending(
            &mut self.errors,
            self.config,
            &mut self.detected_line_ending,
            line_ending,
            start_idx..end_pos,
        );

        self.push(SyntaxKind::NEWLINE, start_idx, end_pos);
        self.line_start = end_pos;
        let in_body = self.block_scalar.is_some();
        track_block_scalar_body(
            &self.tokens,
            self.input,
            self.line_start,
            &mut self.block_scalar,
        );
        if in_body && self.block_scalar.is_none() {
            self.block_scalar_end = self.tokens.len();
        }
    }

    fn lex_whitespace(&mut self, start_idx: usize, ch: char) {
        use SyntaxKind::*;
        let mut end_idx = start_idx + 1;
        let mut has_tabs = ch == '\t';

        while let Some((idx, ch)) = self.chars.peek() {
            if *ch != ' ' && *ch != '\t' {
                break;
            }
            if *ch == '\t' {
                has_tabs = true;
            }
            end_idx = *idx + 1;
            self.chars.next();
        }

        // White space starting a line is its indentation. A line also starts
        // after a BOM, which takes no column.
        let is_indentation = start_idx == self.line_start;

        if is_indentation {
            // Check for tab characters in indentation (forbidden in YAML)
            if has_tabs {
                self.errors.push(WhitespaceError {
                    message: "Tab character used for indentation (forbidden in YAML)".to_string(),
                    range: start_idx..end_idx,
                    category: WhitespaceErrorCategory::TabIndentation,
                });
            }
            self.push(INDENT, start_idx, end_idx);
        } else {
            self.push(WHITESPACE, start_idx, end_idx);
        }
    }

    fn lex_plain_scalar(&mut self, start_idx: usize, ch: char) {
        // Only a header's indentation indicator can precede its chomping
        // `+`. A `|` reaching here is content: `1\n|+\n` is `1 |+`.
        let after_header = ch.is_ascii_digit()
            && matches!(
                self.tokens.last(),
                Some((SyntaxKind::PIPE | SyntaxKind::GREATER, _))
            );
        let end = self.read_plain_scalar_body(start_idx + ch.len_utf8(), true, after_header);
        self.push_scalar(start_idx, end);
    }

    fn finish(self) -> (Vec<(SyntaxKind, &'a str)>, Vec<WhitespaceError>) {
        let mut whitespace_errors = self.errors;
        // Check the final line length if there's no trailing newline
        if let Some(max_len) = self.config.max_line_length {
            let final_line_length = self.input.len() - self.line_start;
            if final_line_length > max_len && final_line_length > 0 {
                whitespace_errors.push(WhitespaceError {
                    message: format!("Line too long ({final_line_length} > {max_len} characters)"),
                    range: self.line_start..self.input.len(),
                    category: WhitespaceErrorCategory::LineTooLong,
                });
            }
        }
        (self.tokens, whitespace_errors)
    }
}

/// Map a plain-scalar's semantic classification onto the concrete
/// [`SyntaxKind`] the tokenizer emits. The classification itself lives
/// in [`ScalarValue::classify_plain`](crate::ScalarValue::classify_plain);
/// this function only translates enum variants.
fn classify_scalar(text: &str) -> SyntaxKind {
    use crate::CoreScalarType;
    match crate::scalar::ScalarValue::classify_plain(text) {
        CoreScalarType::Null => SyntaxKind::NULL,
        CoreScalarType::Boolean => SyntaxKind::BOOL,
        CoreScalarType::Integer => SyntaxKind::INT,
        CoreScalarType::Float => SyntaxKind::FLOAT,
        CoreScalarType::String => SyntaxKind::STRING,
    }
}

/// Common set of YAML special characters
const YAML_SPECIAL_CHARS: &str = ":+-?[]{},'|>&*!%\"#";

/// Check if a character has special meaning in YAML
/// Whether a node property (`!tag`, `&anchor`, `*alias`) may start here.
///
/// Per YAML 1.2 §7.1 these apply to the start of a node. Once a plain
/// scalar has begun on the line, a `!` or `&` is ordinary content, so
/// `a: x !!b` is the scalar `x !!b` rather than a tagged node. A preceding
/// space is not enough to start a new node: what matters is whether the
/// last token was scalar content.
/// Whether the character ends a `---` or `...` marker.
///
/// YAML 1.2 `s-white` is space and tab only, and a marker may also end at a
/// line break. Rust's `char::is_whitespace` is Unicode-wide and counts
/// U+00A0 and friends, which are ordinary scalar content here: `"...\u{a0}-"`
/// is one plain scalar, as saphyr and PyYAML both read it.
fn ends_document_marker(ch: char) -> bool {
    matches!(ch, ' ' | '\t' | '\n' | '\r')
}

/// Whether a BOM at the start of a line is a document prefix.
///
/// YAML 1.2 allows a BOM at the start of any document in a stream, not only
/// the first, so one may follow a `---` or `...` line or precede a `---`.
/// Anywhere else it is ordinary content.
fn bom_starts_document(tokens: &[(SyntaxKind, &str)], input: &str, bom_idx: usize) -> bool {
    use SyntaxKind::*;
    let after_marker = tokens
        .iter()
        .rev()
        .find(|(kind, _)| !matches!(kind, WHITESPACE | INDENT | NEWLINE | COMMENT))
        .map_or(true, |(kind, _)| matches!(kind, DOC_START | DOC_END));
    let rest = &input[bom_idx + '\u{FEFF}'.len_utf8()..];
    let before_marker =
        rest.starts_with("---") && rest[3..].chars().next().map_or(true, ends_document_marker);
    after_marker || before_marker
}

fn node_property_can_start(tokens: &[(SyntaxKind, &str)]) -> bool {
    for (kind, _) in tokens.iter().rev() {
        match kind {
            // Skip the layout between the property and what precedes it.
            SyntaxKind::WHITESPACE | SyntaxKind::INDENT => continue,
            // A node starts after these.
            SyntaxKind::NEWLINE
            | SyntaxKind::COLON
            | SyntaxKind::DASH
            | SyntaxKind::QUESTION
            | SyntaxKind::COMMA
            | SyntaxKind::LEFT_BRACKET
            | SyntaxKind::LEFT_BRACE
            | SyntaxKind::DOC_START
            | SyntaxKind::BOM
            | SyntaxKind::TAG
            | SyntaxKind::ANCHOR
            // A bare `&` or `*` is punctuation rather than content, so a
            // property may still follow it.
            | SyntaxKind::AMPERSAND
            | SyntaxKind::ASTERISK => return true,
            // Anything else is content the property would sit inside.
            _ => return false,
        }
    }
    // Nothing before it: the very start of the input is a node start.
    true
}

/// Whether the line starting at `line_start` carries no content.
fn line_is_blank(input: &str, line_start: usize) -> bool {
    input[line_start..]
        .split(['\n', '\r'])
        .next()
        .unwrap_or_default()
        .trim_matches([' ', '\t'])
        .is_empty()
}

/// Leading whitespace of the line starting at `line_start`.
fn line_indent(input: &str, line_start: usize) -> usize {
    let line = &input[line_start..];
    line.len() - line.trim_start_matches([' ', '\t']).len()
}

/// A block scalar whose body the lexer is reading.
#[derive(Clone, Copy)]
struct BlockScalarBody {
    /// Start of the line carrying the header; the body begins on the next.
    header_line_start: usize,
    /// Index of the header's `|` or `>` token.
    header_token: usize,
    /// Column an explicit indentation indicator counts from.
    parent_column: usize,
    /// Leftmost column a body line may start at. A document's root block
    /// scalar sits at indentation -1, so its body may start at column 0;
    /// any other body has to be indented past its header's line and the
    /// collection it is a node of: in `- a: |\n  b: 1\n` the `b` line is
    /// the mapping's next entry.
    min_indent: usize,
    /// The body's own column, set by an explicit indentation indicator or
    /// else by its first content line.
    indent: Option<usize>,
}

impl BlockScalarBody {
    /// Start tracking for a header found on the line at `line_start`, with
    /// `tokens` holding everything lexed before it.
    fn new(tokens: &[(SyntaxKind, &str)], input: &str, line_start: usize) -> Self {
        // Only `---` and node properties before the header, with nothing
        // indenting the line, make it the document's root node.
        let is_root = tokens
            .iter()
            .rev()
            .take_while(|(kind, _)| !matches!(kind, SyntaxKind::NEWLINE | SyntaxKind::BOM))
            .all(|(kind, _)| {
                matches!(
                    kind,
                    SyntaxKind::DOC_START
                        | SyntaxKind::WHITESPACE
                        | SyntaxKind::TAG
                        | SyntaxKind::ANCHOR
                )
            });
        let parent_column = parent_column(tokens).unwrap_or(0);
        let min_indent = if is_root {
            0
        } else {
            line_indent(input, line_start).max(parent_column) + 1
        };
        Self {
            header_line_start: line_start,
            header_token: tokens.len(),
            parent_column,
            min_indent,
            indent: None,
        }
    }
}

/// The column of the block collection a node starting after `tokens` is a
/// node of, or `None` at the document root.
///
/// This is the column as the parser records it: the `-` or `?`
/// introducing the node, or for a mapping value the start of its key (the
/// `:` itself when the key is empty or explicit, as an explicit key's
/// mapping sits at its `?` and `:`). When the node opens its line, or
/// continues a plain scalar, the collection is the one the last line with
/// an indicator leaves open. A block scalar's indentation indicator counts
/// from this column, taking the root's as 0 as the parser and libyaml do,
/// although YAML 1.2 puts the root at -1.
fn parent_column(tokens: &[(SyntaxKind, &str)]) -> Option<usize> {
    use SyntaxKind::*;
    for line in tokens.rsplit(|(kind, _)| matches!(kind, NEWLINE | BOM)) {
        let Some(last) = line.iter().rposition(|(kind, _)| {
            !matches!(
                kind,
                WHITESPACE
                    | INDENT
                    | COMMENT
                    | TAG
                    | ANCHOR
                    | STRING
                    | INT
                    | FLOAT
                    | BOOL
                    | NULL
                    | MERGE_KEY
            )
        }) else {
            continue;
        };
        let column = |index: usize| -> usize {
            line[..index]
                .iter()
                .map(|(_, text)| text.chars().count())
                .sum()
        };
        return match line[last].0 {
            DASH | QUESTION => Some(column(last)),
            COLON => {
                // The key starts after the block indicators opening the
                // line. An explicit `:` is one of them, as its value may be
                // a compact mapping: in `: b: |1` that sits at `b`.
                let key = line[..last]
                    .iter()
                    .position(|(kind, _)| {
                        !matches!(kind, WHITESPACE | INDENT | DASH | QUESTION | COLON)
                    })
                    .unwrap_or(last);
                Some(column(key))
            }
            DOC_START | DOC_END => None,
            _ => Some(0),
        };
    }
    None
}

/// Read a block scalar header's indicators from the tokens following its
/// `|` or `>`, returning the indentation indicator if there is one, and how
/// many tokens the indicators and the whitespace between them take up. The
/// header's comment and line break are not counted.
///
/// Format: `|<indent><chomp>` or `|<chomp><indent>`, as in `|2`, `|-`,
/// `|2-` and `|-2`.
pub(crate) fn block_scalar_header<'a>(
    tokens: impl IntoIterator<Item = (SyntaxKind, &'a str)>,
) -> (Option<usize>, usize) {
    let mut explicit_indent = None;
    let mut len = 0;
    for (kind, text) in tokens {
        match kind {
            SyntaxKind::INT => {
                // The spec allows a single digit 1-9, but the lexer hands
                // over the whole run, so `|10` arrives as "10". Take it as
                // part of the header rather than leaving it for the body,
                // which would strand it.
                //
                // A chomping indicator before the digit lexes into the same
                // token (`|-2` is INT("-2")), so allow one sign.
                let digits = text.strip_prefix(['+', '-']).unwrap_or(text);
                if digits.is_empty() || !digits.chars().all(|c| c.is_ascii_digit()) {
                    break;
                }
                // A longer run is out of range, so keep only the first digit.
                explicit_indent = digits
                    .chars()
                    .next()
                    .and_then(|c| c.to_digit(10))
                    .filter(|d| *d > 0)
                    .map(|d| d as usize);
            }
            SyntaxKind::STRING => {
                // Could be a chomping indicator, or an indentation digit and
                // chomping indicator lexed together (`|3-`, `|-3`). A
                // header's indicators are ASCII, so compare characters
                // rather than bytes: `text.len()` is 2 for a single no-break
                // space.
                let mut chars = text.chars();
                let is_header = match (chars.next(), chars.next(), chars.next()) {
                    (Some(a), None, _) => a == '+' || a == '-',
                    (Some(a), Some(b), None) => {
                        let digit = |c: char| c.is_ascii_digit() && c != '0';
                        let chomp = |c: char| c == '+' || c == '-';
                        (digit(a) && chomp(b)) || (chomp(a) && digit(b))
                    }
                    _ => false,
                };
                if !is_header {
                    break;
                }
                explicit_indent = text
                    .chars()
                    .find_map(|c| c.to_digit(10))
                    .filter(|d| *d > 0)
                    .map(|d| d as usize);
            }
            // A `+` straight after the `|` or `>` lexes as PLUS rather than
            // STRING, but it is the same chomping indicator.
            SyntaxKind::PLUS | SyntaxKind::WHITESPACE => {}
            _ => break,
        }
        len += 1;
    }
    (explicit_indent, len)
}

/// Whether the line starting at `line_start` is a `---` or `...` marker.
fn line_is_document_marker(input: &str, line_start: usize) -> bool {
    let line = &input[line_start..];
    (line.starts_with("---") || line.starts_with("..."))
        && line[3..].chars().next().map_or(true, ends_document_marker)
}

/// Update the lexer's block-scalar tracking at the start of a new line.
///
/// A block scalar's body ends at the first following line that is not blank
/// and starts left of the body's column (or, before a content line has set
/// that, left of the leftmost column allowed), or at a document marker,
/// which is the only thing that ends a root body at column 0. This runs for
/// every line ending, CRLF and lone CR included: left to the `\n` arm
/// alone, a CRLF file kept the tracking set and read a later `[` as literal
/// body text rather than a flow collection.
fn track_block_scalar_body(
    tokens: &[(SyntaxKind, &str)],
    input: &str,
    line_start: usize,
    body: &mut Option<BlockScalarBody>,
) {
    let Some(current) = body else {
        return;
    };
    // An explicit indentation indicator fixes the body's column, so a body
    // line less indented than the first one still belongs to it:
    // `- |2\n   x\n  y # z\n` holds the text `y # z`.
    if current.indent.is_none() {
        if let (Some(indicator), _) =
            block_scalar_header(tokens[current.header_token + 1..].iter().copied())
        {
            current.indent = Some(current.parent_column + indicator);
        }
    }
    if line_is_blank(input, line_start) {
        return;
    }
    let indent = line_indent(input, line_start);
    if indent < current.indent.unwrap_or(current.min_indent)
        || line_is_document_marker(input, line_start)
    {
        *body = None;
    } else if current.indent.is_none() {
        // Only spaces indent, so a tab after them is content. As in the
        // parser, tabs count when the spaces alone don't reach the body.
        let spaces = crate::column::space_indent(&input[line_start..]);
        current.indent = Some(if spaces >= current.min_indent {
            spaces
        } else {
            indent
        });
    }
}

/// Whether the line starting at `line_start` is part of a block scalar's
/// body, all of which is literal text: `a: |\n  {\nb,c: 1\n` holds the
/// scalar `{`, `- >\n  # detected\n` the text `# detected` (test suite
/// 4QFQ), and `--- |\n%!PS-Adobe-2.0\n` a line that is not a directive
/// (W4TN).
///
/// [`track_block_scalar_body`] drops the tracking once a line ends the body,
/// so this only has to rule out the header's own line.
fn in_block_scalar_body(body: Option<BlockScalarBody>, line_start: usize) -> bool {
    body.is_some_and(|body| body.header_line_start != line_start)
}

/// Whether `ch` may appear in a tag shorthand's suffix.
///
/// YAML 1.2 `ns-tag-char` is any URI character except the flow indicators
/// `,`, `[`, `]`, `{`, `}` and the `!` that would start another tag. That
/// admits plenty the lexer treats as special elsewhere -- `:` in `!!ss:eq`,
/// `-` in `!!ss---` -- so a tag cannot be scanned with the general
/// [`YAML_SPECIAL_CHARS`] set.
fn is_tag_char(ch: char) -> bool {
    if ch.is_whitespace() {
        return false;
    }
    !matches!(ch, ',' | '[' | ']' | '{' | '}' | '!')
}

/// Whether the character is YAML whitespace (`s-white`, space and tab
/// only) or a line break. Unlike `char::is_whitespace` this leaves U+00A0
/// and friends as content, as [`ends_document_marker`] does.
fn is_white_or_break(ch: char) -> bool {
    matches!(ch, ' ' | '\t' | '\n' | '\r')
}

fn is_yaml_special(ch: char) -> bool {
    YAML_SPECIAL_CHARS.contains(ch)
}

/// Check if character is YAML special, with optional exclusions
fn is_yaml_special_except(ch: char, exclude: &str) -> bool {
    YAML_SPECIAL_CHARS.contains(ch) && !exclude.contains(ch)
}

/// Check whether `name` is a well-formed anchor/alias name (`ns-anchor-name`).
///
/// Per YAML 1.2 that is one or more `ns-anchor-char`s: printable characters
/// other than whitespace, line breaks, the byte order mark and the flow
/// indicators. The lexer reads a name up to whitespace or a flow
/// indicator (see [`read_anchor_name_from`]), so such a name round-trips.
pub(crate) fn is_valid_anchor_name(name: &str) -> bool {
    !name.is_empty() && name.chars().all(is_anchor_char)
}

/// `ns-anchor-char`: `ns-char` minus the flow indicators.
fn is_anchor_char(ch: char) -> bool {
    // c-printable, less s-white, b-char and the byte order mark.
    let ns_char = matches!(ch,
        '\u{21}'..='\u{7E}'
        | '\u{85}'
        | '\u{A0}'..='\u{D7FF}'
        | '\u{E000}'..='\u{FFFD}'
        | '\u{10000}'..='\u{10FFFF}')
        && ch != '\u{FEFF}';
    ns_char && !matches!(ch, ',' | '[' | ']' | '{' | '}')
}

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

    #[test]
    fn test_is_valid_anchor_name() {
        assert!(is_valid_anchor_name("shared"));
        assert!(is_valid_anchor_name("shared_1"));
        assert!(is_valid_anchor_name("anchor.dotted"));
        assert!(is_valid_anchor_name("naïve"));
        // Only the flow indicators are excluded from the indicators.
        for name in [
            "a:",
            "colon:name",
            "*star",
            "&amp",
            "-",
            "x?y",
            "#",
            "!t",
            "|>",
            "%'\"",
            "@`",
            "no\u{A0}break",
            "nel\u{85}",
        ] {
            assert!(is_valid_anchor_name(name), "{name:?}");
        }

        for name in [
            "",
            "has space",
            "has\ttab",
            "has\nnewline",
            "has\rreturn",
            "comma,name",
            "bracket]name",
            "[",
            "{",
            "}",
            "bom\u{FEFF}",
            "bell\u{7}",
            "del\u{7F}",
            "c1\u{80}",
        ] {
            assert!(!is_valid_anchor_name(name), "{name:?}");
        }
    }

    #[test]
    fn test_embedded_hyphens_stay_in_one_plain_scalar() {
        let input = format!("k: {}\n", "a-".repeat(200));
        let kinds: Vec<_> = lex(&input).into_iter().map(|(k, _)| k).collect();
        assert!(kinds.contains(&SyntaxKind::STRING) || kinds.contains(&SyntaxKind::SCALAR));
        let dash_count = kinds.iter().filter(|k| **k == SyntaxKind::DASH).count();
        assert_eq!(dash_count, 0);
        let seq = lex("- a\n- b\n");
        let seq_dashes = seq.iter().filter(|(k, _)| *k == SyntaxKind::DASH).count();
        assert_eq!(seq_dashes, 2);
    }

    /// Names accepted by [`is_valid_anchor_name`] must round-trip through the
    /// lexer as a single `REFERENCE` token, so `Alias::new` and the lexer stay
    /// in agreement.
    #[test]
    fn test_valid_anchor_names_round_trip_through_lexer() {
        for name in [
            "shared",
            "shared_1",
            "anchor.dotted",
            "naïve",
            "a:",
            "*star",
            "x#y",
            "no\u{A0}break",
        ] {
            let input = format!("*{name}");
            let tokens = lex(&input);
            assert_eq!(
                tokens,
                vec![(SyntaxKind::REFERENCE, input.as_str())],
                "name {name:?} did not round-trip as a single REFERENCE token"
            );
        }
    }

    #[test]
    fn test_simple_mapping() {
        let input = "key: value";
        let tokens = lex(input);

        assert_eq!(tokens.len(), 4);
        assert_eq!(tokens[0], (SyntaxKind::STRING, "key"));
        assert_eq!(tokens[1], (SyntaxKind::COLON, ":"));
        assert_eq!(tokens[2], (SyntaxKind::WHITESPACE, " "));
        assert_eq!(tokens[3], (SyntaxKind::STRING, "value"));
    }

    #[test]
    fn test_scalar_types() {
        // Test integer
        let tokens = lex("age: 42");
        assert_eq!(tokens[0], (SyntaxKind::STRING, "age"));
        assert_eq!(tokens[3], (SyntaxKind::INT, "42"));

        // Test float
        let tokens = lex("pi: 3.14");
        assert_eq!(tokens[0], (SyntaxKind::STRING, "pi"));
        assert_eq!(tokens[3], (SyntaxKind::FLOAT, "3.14"));

        // Test boolean true
        let tokens = lex("enabled: true");
        assert_eq!(tokens[0], (SyntaxKind::STRING, "enabled"));
        assert_eq!(tokens[3], (SyntaxKind::BOOL, "true"));

        // Test boolean false
        let tokens = lex("disabled: false");
        assert_eq!(tokens[3], (SyntaxKind::BOOL, "false"));

        // Test null
        let tokens = lex("value: null");
        assert_eq!(tokens[3], (SyntaxKind::NULL, "null"));

        // Test tilde as null
        let tokens = lex("value: ~");
        assert_eq!(tokens[3], (SyntaxKind::NULL, "~"));
    }

    #[test]
    fn test_sequences() {
        let input = "- item1\n- item2";
        let tokens = lex(input);

        assert_eq!(tokens[0], (SyntaxKind::DASH, "-"));
        assert_eq!(tokens[1], (SyntaxKind::WHITESPACE, " "));
        assert_eq!(tokens[2], (SyntaxKind::STRING, "item1"));
        assert_eq!(tokens[3], (SyntaxKind::NEWLINE, "\n"));
        assert_eq!(tokens[4], (SyntaxKind::DASH, "-"));
        assert_eq!(tokens[5], (SyntaxKind::WHITESPACE, " "));
        assert_eq!(tokens[6], (SyntaxKind::STRING, "item2"));
    }

    #[test]
    fn test_hyphen_in_scalars() {
        // Test hyphens in scalar values should not be treated as sequence markers
        let input = "Name: example-project";
        let tokens = lex(input);

        println!("Hyphen test tokens:");
        for (i, (kind, text)) in tokens.iter().enumerate() {
            println!("  {}: {:?} = {:?}", i, kind, text);
        }

        // Should get: STRING("Name"), COLON(":"), WHITESPACE(" "), STRING("example-project")
        assert_eq!(tokens.len(), 4);
        assert_eq!(tokens[0], (SyntaxKind::STRING, "Name"));
        assert_eq!(tokens[1], (SyntaxKind::COLON, ":"));
        assert_eq!(tokens[2], (SyntaxKind::WHITESPACE, " "));
        assert_eq!(tokens[3], (SyntaxKind::STRING, "example-project"));
    }

    #[test]
    fn test_hyphen_sequence_vs_scalar() {
        // Test that sequence markers are still recognized correctly
        let sequence_input = "- example-item";
        let tokens = lex(sequence_input);

        println!("Sequence hyphen tokens:");
        for (i, (kind, text)) in tokens.iter().enumerate() {
            println!("  {}: {:?} = {:?}", i, kind, text);
        }

        // Should get: DASH("-"), WHITESPACE(" "), STRING("example-item")
        assert_eq!(tokens[0], (SyntaxKind::DASH, "-"));
        assert_eq!(tokens[1], (SyntaxKind::WHITESPACE, " "));
        assert_eq!(tokens[2], (SyntaxKind::STRING, "example-item"));

        // Test scalar with hyphens in different contexts
        let scalar_input = "package-name: my-awesome-package";
        let tokens = lex(scalar_input);

        println!("Package hyphen tokens:");
        for (i, (kind, text)) in tokens.iter().enumerate() {
            println!("  {}: {:?} = {:?}", i, kind, text);
        }

        // Should get: STRING("package-name"), COLON(":"), WHITESPACE(" "), STRING("my-awesome-package")
        assert_eq!(tokens.len(), 4);
        assert_eq!(tokens[0], (SyntaxKind::STRING, "package-name"));
        assert_eq!(tokens[3], (SyntaxKind::STRING, "my-awesome-package"));
    }

    #[test]
    fn test_flow_style() {
        // Flow sequence
        let tokens = lex("[1, 2, 3]");
        assert_eq!(tokens[0], (SyntaxKind::LEFT_BRACKET, "["));
        assert_eq!(tokens[1], (SyntaxKind::INT, "1"));
        assert_eq!(tokens[2], (SyntaxKind::COMMA, ","));
        assert_eq!(tokens[3], (SyntaxKind::WHITESPACE, " "));
        assert_eq!(tokens[4], (SyntaxKind::INT, "2"));
        assert_eq!(tokens[5], (SyntaxKind::COMMA, ","));
        assert_eq!(tokens[6], (SyntaxKind::WHITESPACE, " "));
        assert_eq!(tokens[7], (SyntaxKind::INT, "3"));
        assert_eq!(tokens[8], (SyntaxKind::RIGHT_BRACKET, "]"));

        // Flow mapping
        let tokens = lex("{a: 1, b: 2}");
        assert_eq!(tokens[0], (SyntaxKind::LEFT_BRACE, "{"));
        assert_eq!(tokens[1], (SyntaxKind::STRING, "a"));
        assert_eq!(tokens[2], (SyntaxKind::COLON, ":"));
        assert_eq!(tokens[3], (SyntaxKind::WHITESPACE, " "));
        assert_eq!(tokens[4], (SyntaxKind::INT, "1"));
    }

    #[test]
    fn test_comments() {
        let input = "key: value # this is a comment\n# full line comment";
        let tokens = lex(input);

        // Find comment tokens
        let comments: Vec<_> = tokens
            .iter()
            .filter(|(kind, _)| *kind == SyntaxKind::COMMENT)
            .collect();

        assert_eq!(comments.len(), 2);
        assert_eq!(comments[0].1, "# this is a comment");
        assert_eq!(comments[1].1, "# full line comment");
    }

    #[test]
    fn test_multiline_scalar() {
        let input = "key: value\n  continued";
        let tokens = lex(input);

        // Check for indent token
        let indents: Vec<_> = tokens
            .iter()
            .filter(|(kind, _)| *kind == SyntaxKind::INDENT)
            .collect();
        assert_eq!(indents.len(), 1);
        assert_eq!(indents[0].1, "  ");
    }

    #[test]
    fn test_quoted_strings() {
        let input = r#"single: 'quoted'
double: "quoted""#;
        let tokens = lex(input);

        // Find quoted string tokens - after fix, quotes are included in STRING tokens
        let quoted_strings: Vec<_> = tokens
            .iter()
            .filter(|(kind, text)| {
                *kind == SyntaxKind::STRING && (text.starts_with('\'') || text.starts_with('"'))
            })
            .collect();
        assert_eq!(quoted_strings.len(), 2); // single and double quoted strings

        // Verify content (order depends on which appears first in the source)
        let quoted_texts: Vec<&str> = {
            let mut v: Vec<&str> = quoted_strings.iter().map(|(_, t)| *t).collect();
            v.sort();
            v
        };
        assert_eq!(quoted_texts, ["\"quoted\"", "'quoted'"]);
    }

    #[test]
    fn test_document_markers() {
        let input = "---\nkey: value\n...";
        let tokens = lex(input);

        println!("Document tokens:");
        for (i, (kind, text)) in tokens.iter().enumerate() {
            println!("  {}: {:?} = {:?}", i, kind, text);
        }

        // Check for document start and end markers
        let doc_start_count = tokens
            .iter()
            .filter(|(kind, _)| *kind == SyntaxKind::DOC_START)
            .count();
        let doc_end_count = tokens
            .iter()
            .filter(|(kind, _)| *kind == SyntaxKind::DOC_END)
            .count();
        assert_eq!(doc_start_count, 1);
        assert_eq!(doc_end_count, 1);
    }

    #[test]
    fn test_empty_input() {
        let input = "";
        let tokens = lex(input);
        println!("Empty input tokens: {:?}", tokens);
        assert_eq!(tokens.len(), 0);
    }

    #[test]
    fn test_anchors_and_aliases() {
        // Test anchor definition
        let input = "key: &anchor_name value";
        let tokens = lex(input);
        println!("Anchor tokens: {:?}", tokens);

        let anchors: Vec<_> = tokens
            .iter()
            .filter(|(kind, _)| *kind == SyntaxKind::ANCHOR)
            .collect();
        assert_eq!(anchors.len(), 1);
        assert_eq!(anchors[0].1, "&anchor_name");

        // Test alias reference
        let input = "key: *reference_name";
        let tokens = lex(input);
        println!("Reference tokens: {:?}", tokens);

        let references: Vec<_> = tokens
            .iter()
            .filter(|(kind, _)| *kind == SyntaxKind::REFERENCE)
            .collect();
        assert_eq!(references.len(), 1);
        assert_eq!(references[0].1, "*reference_name");

        // Test bare ampersand and asterisk (should not be treated as anchors/references)
        let input = "key: & *";
        let tokens = lex(input);

        let ampersands: Vec<_> = tokens
            .iter()
            .filter(|(kind, _)| *kind == SyntaxKind::AMPERSAND)
            .collect();
        assert_eq!(ampersands.len(), 1);

        let asterisks: Vec<_> = tokens
            .iter()
            .filter(|(kind, _)| *kind == SyntaxKind::ASTERISK)
            .collect();
        assert_eq!(asterisks.len(), 1);
    }

    #[test]
    fn test_indicators_mid_plain_scalar_are_scalar_content() {
        // Per YAML 1.2 §7.1 the node-property indicators `&`, `*`, `!`
        // only apply when a scalar starts with them (followed by an
        // identifier). Mid-scalar they are ordinary content and must
        // not create phantom ANCHOR / REFERENCE / TAG tokens.
        // Including space-separated: a preceding space does not start a new
        // node, so `k: a !b` is still one plain scalar.
        for input in [
            "k: 3.1&1", "k: a&b", "k: a*b", "k: a!b", "k: a?b", "k: a !b", "k: a !!b", "k: a &b",
            "k: a *b", "- a !b",
        ] {
            let tokens = lex(input);
            let bad: Vec<_> = tokens
                .iter()
                .filter(|(kind, _)| {
                    matches!(
                        kind,
                        SyntaxKind::ANCHOR
                            | SyntaxKind::REFERENCE
                            | SyntaxKind::TAG
                            | SyntaxKind::QUESTION
                    )
                })
                .collect();
            assert!(
                bad.is_empty(),
                "{:?} produced structural tokens mid-scalar: {:?}",
                input,
                bad
            );
        }
        // And a real anchor still tokenizes as ANCHOR.
        let tokens = lex("a: &real 1");
        assert!(
            tokens.iter().any(|(k, _)| *k == SyntaxKind::ANCHOR),
            "real anchor should still be recognized: {:?}",
            tokens
        );
    }

    #[test]
    fn test_merge_key_token() {
        // Test merge key '<<'
        let input = "<<: *defaults";
        let tokens = lex(input);

        let merge_keys: Vec<_> = tokens
            .iter()
            .filter(|(kind, _)| *kind == SyntaxKind::MERGE_KEY)
            .collect();
        assert_eq!(merge_keys.len(), 1);
        assert_eq!(merge_keys[0].1, "<<");

        // Test single '<' is not a merge key
        let input2 = "key: < value";
        let tokens2 = lex(input2);

        let merge_keys2: Vec<_> = tokens2
            .iter()
            .filter(|(kind, _)| *kind == SyntaxKind::MERGE_KEY)
            .collect();
        assert_eq!(merge_keys2.len(), 0, "Single < should not be a merge key");
    }

    #[test]
    fn test_angle_prefix_is_plain_scalar_not_merge_key() {
        let tokens = lex("<<foo: 1");
        let merge_keys: Vec<_> = tokens
            .iter()
            .filter(|(kind, _)| *kind == SyntaxKind::MERGE_KEY)
            .collect();
        assert_eq!(
            merge_keys.len(),
            0,
            "<<foo should be a plain key, got {:?}",
            tokens
        );
        assert!(
            tokens
                .iter()
                .any(|(kind, text)| *kind != SyntaxKind::MERGE_KEY && *text == "<<foo"),
            "expected <<foo scalar, got {:?}",
            tokens
        );

        let tokens = lex("{<<foo: 1, more: x}");
        assert_eq!(
            tokens
                .iter()
                .filter(|(kind, _)| *kind == SyntaxKind::MERGE_KEY)
                .count(),
            0,
            "flow <<foo should not be MERGE_KEY: {:?}",
            tokens
        );

        let tokens = lex("<<: *defaults");
        assert_eq!(
            tokens
                .iter()
                .filter(|(kind, text)| *kind == SyntaxKind::MERGE_KEY && *text == "<<")
                .count(),
            1
        );

        use crate::YamlFile;
        use std::str::FromStr;
        let file = YamlFile::from_str("<<foo: 1\nbar: 2\n").unwrap();
        let mapping = file.document().unwrap().as_mapping().unwrap();
        assert_eq!(mapping.keys().count(), 2);
        assert_eq!(
            mapping
                .get("<<foo")
                .unwrap()
                .as_scalar()
                .unwrap()
                .as_string(),
            "1"
        );
    }

    /// `<<` is a merge key only when the plain scalar there is exactly
    /// `<<`: a `#` or `:` glued to it, or more words after a space, are
    /// part of the scalar.
    #[test]
    fn test_merge_key_is_a_whole_plain_scalar() {
        use SyntaxKind::*;
        assert_eq!(
            lex("<<#: 1"),
            [(STRING, "<<#"), (COLON, ":"), (WHITESPACE, " "), (INT, "1")]
        );
        assert_eq!(
            lex("<<:x: 1"),
            [
                (STRING, "<<:x"),
                (COLON, ":"),
                (WHITESPACE, " "),
                (INT, "1")
            ]
        );
        assert_eq!(
            lex("<< x: 1"),
            [
                (STRING, "<< x"),
                (COLON, ":"),
                (WHITESPACE, " "),
                (INT, "1")
            ]
        );
        assert_eq!(
            lex("<<]: 1"),
            [(STRING, "<<]"), (COLON, ":"), (WHITESPACE, " "), (INT, "1")]
        );
        assert_eq!(
            lex("<< #c\n"),
            [
                (MERGE_KEY, "<<"),
                (WHITESPACE, " "),
                (COMMENT, "#c"),
                (NEWLINE, "\n")
            ]
        );
        assert_eq!(
            lex("{<<,<<: 1}"),
            [
                (LEFT_BRACE, "{"),
                (MERGE_KEY, "<<"),
                (COMMA, ","),
                (MERGE_KEY, "<<"),
                (COLON, ":"),
                (WHITESPACE, " "),
                (INT, "1"),
                (RIGHT_BRACE, "}")
            ]
        );
    }

    #[test]
    fn test_angle_prefixed_scalar_uses_plain_scalar_rules() {
        // A key starting with `<` follows the same plain-scalar rules as any
        // other key: `-`, `!`, `&` and a non-indicator `:` are ordinary
        // content, not token breaks.
        use crate::YamlFile;
        use std::str::FromStr;

        for key in [
            "<foo-bar",
            "<<foo-bar",
            "<foo!x",
            "<<foo!x",
            "<foo&x",
            "<<foo&x",
            "<foo:x",
            "<<foo:x",
        ] {
            let src = format!("{}: 1\n", key);
            let tokens = lex(&src);
            assert_eq!(
                tokens
                    .iter()
                    .filter(|(kind, _)| *kind != SyntaxKind::WHITESPACE)
                    .map(|(_, text)| *text)
                    .collect::<Vec<_>>(),
                vec![key, ":", "1", "\n"],
                "unexpected tokens for {:?}",
                src
            );

            let file = YamlFile::from_str(&src).unwrap();
            assert_eq!(file.to_string(), src);
            let mapping = file.document().unwrap().as_mapping().unwrap();
            assert_eq!(
                mapping.get(key).unwrap().as_scalar().unwrap().as_string(),
                "1"
            );
        }
    }

    #[test]
    fn test_plus_token() {
        // Test plus as standalone token
        let input = "key: |+ value";
        let tokens = lex(input);

        let plus_tokens: Vec<_> = tokens
            .iter()
            .filter(|(kind, _)| *kind == SyntaxKind::PLUS)
            .collect();
        assert_eq!(plus_tokens.len(), 1);
        assert_eq!(plus_tokens[0].1, "+");
    }

    /// A `+` is a chomping indicator only where it closes a block scalar
    /// header, and otherwise starts a plain scalar.
    #[test]
    fn test_plus_tokens() {
        use SyntaxKind::*;
        let cases: Vec<(&str, Vec<(SyntaxKind, &str)>)> = vec![
            (
                "+ x: v\n",
                vec![
                    (STRING, "+ x"),
                    (COLON, ":"),
                    (WHITESPACE, " "),
                    (STRING, "v"),
                    (NEWLINE, "\n"),
                ],
            ),
            (
                "+\tx: v\n",
                vec![
                    (STRING, "+\tx"),
                    (COLON, ":"),
                    (WHITESPACE, " "),
                    (STRING, "v"),
                    (NEWLINE, "\n"),
                ],
            ),
            (
                "+ \n",
                vec![(STRING, "+"), (WHITESPACE, " "), (NEWLINE, "\n")],
            ),
            (
                "+ #c\n",
                vec![
                    (STRING, "+"),
                    (WHITESPACE, " "),
                    (COMMENT, "#c"),
                    (NEWLINE, "\n"),
                ],
            ),
            ("+[b]\n", vec![(STRING, "+[b]"), (NEWLINE, "\n")]),
            (
                "[+[b]]\n",
                vec![
                    (LEFT_BRACKET, "["),
                    (STRING, "+"),
                    (LEFT_BRACKET, "["),
                    (STRING, "b"),
                    (RIGHT_BRACKET, "]"),
                    (RIGHT_BRACKET, "]"),
                    (NEWLINE, "\n"),
                ],
            ),
            ("+'a'\n", vec![(STRING, "+'a'"), (NEWLINE, "\n")]),
            ("+?\n", vec![(STRING, "+?"), (NEWLINE, "\n")]),
            ("+|\n", vec![(STRING, "+|"), (NEWLINE, "\n")]),
            (
                "+#: v\n",
                vec![
                    (STRING, "+#"),
                    (COLON, ":"),
                    (WHITESPACE, " "),
                    (STRING, "v"),
                    (NEWLINE, "\n"),
                ],
            ),
            (
                "+ #: v\n",
                vec![
                    (STRING, "+"),
                    (WHITESPACE, " "),
                    (COMMENT, "#: v"),
                    (NEWLINE, "\n"),
                ],
            ),
            (
                "+: v\n",
                vec![
                    (STRING, "+"),
                    (COLON, ":"),
                    (WHITESPACE, " "),
                    (STRING, "v"),
                    (NEWLINE, "\n"),
                ],
            ),
            (
                "+ : v\n",
                vec![
                    (STRING, "+"),
                    (WHITESPACE, " "),
                    (COLON, ":"),
                    (WHITESPACE, " "),
                    (STRING, "v"),
                    (NEWLINE, "\n"),
                ],
            ),
            ("+:x\n", vec![(STRING, "+:x"), (NEWLINE, "\n")]),
            (
                "+\nx\n",
                vec![
                    (STRING, "+"),
                    (NEWLINE, "\n"),
                    (STRING, "x"),
                    (NEWLINE, "\n"),
                ],
            ),
            (
                "+\r\nx\n",
                vec![
                    (STRING, "+"),
                    (NEWLINE, "\r\n"),
                    (STRING, "x"),
                    (NEWLINE, "\n"),
                ],
            ),
            (
                "- |+\n  x\n",
                vec![
                    (DASH, "-"),
                    (WHITESPACE, " "),
                    (PIPE, "|"),
                    (PLUS, "+"),
                    (NEWLINE, "\n"),
                    (INDENT, "  "),
                    (STRING, "x"),
                    (NEWLINE, "\n"),
                ],
            ),
            (
                "/|+: v\n",
                vec![
                    (STRING, "/|+"),
                    (COLON, ":"),
                    (WHITESPACE, " "),
                    (STRING, "v"),
                    (NEWLINE, "\n"),
                ],
            ),
            ("i:|+\n", vec![(STRING, "i:|+"), (NEWLINE, "\n")]),
            ("+.INF\n", vec![(FLOAT, "+.INF"), (NEWLINE, "\n")]),
            (
                "{+, a}\n",
                vec![
                    (LEFT_BRACE, "{"),
                    (STRING, "+"),
                    (COMMA, ","),
                    (WHITESPACE, " "),
                    (STRING, "a"),
                    (RIGHT_BRACE, "}"),
                    (NEWLINE, "\n"),
                ],
            ),
            ("+,\n", vec![(STRING, "+,"), (NEWLINE, "\n")]),
            (
                "a: 1+2\n",
                vec![
                    (STRING, "a"),
                    (COLON, ":"),
                    (WHITESPACE, " "),
                    (STRING, "1+2"),
                    (NEWLINE, "\n"),
                ],
            ),
            ("i:|2+\n", vec![(STRING, "i:|2+"), (NEWLINE, "\n")]),
            (
                "x: a |2+\n",
                vec![
                    (STRING, "x"),
                    (COLON, ":"),
                    (WHITESPACE, " "),
                    (STRING, "a |2+"),
                    (NEWLINE, "\n"),
                ],
            ),
            (
                "k: |2+\n",
                vec![
                    (STRING, "k"),
                    (COLON, ":"),
                    (WHITESPACE, " "),
                    (PIPE, "|"),
                    (INT, "2"),
                    (PLUS, "+"),
                    (NEWLINE, "\n"),
                ],
            ),
            (
                "- |2+\n",
                vec![
                    (DASH, "-"),
                    (WHITESPACE, " "),
                    (PIPE, "|"),
                    (INT, "2"),
                    (PLUS, "+"),
                    (NEWLINE, "\n"),
                ],
            ),
        ];
        for (input, expected) in cases {
            assert_eq!(lex(input), expected, "lexing {input:?}");
        }
    }

    /// A header's chomping `+` stands alone after node properties and
    /// an explicit key's `?` too.
    #[test]
    fn test_chomping_plus_after_properties() {
        use SyntaxKind::*;
        for (input, before) in [
            ("? |2+\n", vec![(QUESTION, "?")]),
            ("!t |2+\n", vec![(TAG, "!t")]),
            ("&a |2+\n", vec![(ANCHOR, "&a")]),
        ] {
            let mut expected = before;
            expected.extend([
                (WHITESPACE, " "),
                (PIPE, "|"),
                (INT, "2"),
                (PLUS, "+"),
                (NEWLINE, "\n"),
            ]);
            assert_eq!(lex(input), expected, "lexing {input:?}");
        }
    }

    #[test]
    fn test_block_scalar_indicators() {
        // Test literal with chomping indicators
        let input1 = "key: |+ content";
        let tokens1 = lex(input1);

        assert!(tokens1
            .iter()
            .any(|(kind, text)| *kind == SyntaxKind::PIPE && *text == "|"));
        assert!(tokens1
            .iter()
            .any(|(kind, text)| *kind == SyntaxKind::PLUS && *text == "+"));

        // Test folded with chomping indicators
        let input2 = "key: >- content";
        let tokens2 = lex(input2);

        assert!(tokens2
            .iter()
            .any(|(kind, text)| *kind == SyntaxKind::GREATER && *text == ">"));
        assert!(tokens2
            .iter()
            .any(|(kind, text)| *kind == SyntaxKind::STRING && *text == "-"));

        // Test with explicit indentation
        let input3 = "key: |2+ content";
        let tokens3 = lex(input3);

        assert!(tokens3
            .iter()
            .any(|(kind, text)| *kind == SyntaxKind::PIPE && *text == "|"));
        assert!(tokens3
            .iter()
            .any(|(kind, text)| *kind == SyntaxKind::INT && *text == "2"));
        assert!(tokens3
            .iter()
            .any(|(kind, text)| *kind == SyntaxKind::PLUS && *text == "+"));
    }

    #[test]
    fn test_special_characters_in_block_content() {
        let input = "line with - and + and : characters";
        let tokens = lex(input);

        // The `-` and `+` are content, so the whole key is one plain scalar:
        // saphyr reads this as `line with - and + and` mapped to
        // `characters`.
        assert!(tokens
            .iter()
            .any(|(kind, text)| *kind == SyntaxKind::STRING && *text == "line with - and + and"));
        assert!(!tokens.iter().any(|(kind, _)| *kind == SyntaxKind::PLUS));

        // The mapping colon still stands alone.
        assert!(tokens
            .iter()
            .any(|(kind, text)| *kind == SyntaxKind::COLON && *text == ":"));

        // Round-trip: concatenating tokens reproduces input.
        let joined: String = tokens.iter().map(|(_, t)| *t).collect();
        assert_eq!(joined, input);
    }

    #[test]
    fn test_token_recognition() {
        let input = "key: |2+ \n  content with - and : and > chars\n  more content";
        let tokens = lex(input);

        // Print tokens for debugging
        println!("Tokens:");
        for (i, (kind, text)) in tokens.iter().enumerate() {
            println!("  {}: {:?} = {:?}", i, kind, text);
        }

        // Verify all expected token kinds are present (input: "key: |2+ \n  content with - and : and > chars\n  more content")
        let count = |k: SyntaxKind| tokens.iter().filter(|(kind, _)| *kind == k).count();
        // Two colons: one for the mapping ("key:"), one in the value content ("and :")
        assert_eq!(count(SyntaxKind::COLON), 2);
        assert_eq!(count(SyntaxKind::PIPE), 1); // "|"
        assert_eq!(count(SyntaxKind::INT), 1); // "2"
        assert_eq!(count(SyntaxKind::PLUS), 1); // "+"
                                                // The `>` here sits inside the block scalar's content, not at a
                                                // node start, so it is scalar text rather than a folded header.
        assert_eq!(count(SyntaxKind::GREATER), 0);
        assert_eq!(count(SyntaxKind::NEWLINE), 2); // after "|2+" line and after first content line
        assert_eq!(count(SyntaxKind::INDENT), 2); // "  " before each content line
                                                  // Multiple STRING tokens: "key", content words, and the hyphen
        assert!(count(SyntaxKind::STRING) >= 1, "expected STRING tokens");

        assert_eq!(
            tokens
                .iter()
                .filter(|(kind, text)| *kind == SyntaxKind::STRING && *text == "content with - and")
                .count(),
            1
        );
    }

    #[test]
    fn test_dash_handling() {
        // Test 1: Document start marker
        let input = "---\nkey: value";
        let tokens = lex(input);
        assert_eq!(tokens[0], (SyntaxKind::DOC_START, "---"));

        // Test 2: Document with just three dashes
        let input = "---";
        let tokens = lex(input);
        assert_eq!(tokens.len(), 1);
        assert_eq!(tokens[0], (SyntaxKind::DOC_START, "---"));

        // Test 3: Two dashes (not a document marker)
        let input = "--";
        let tokens = lex(input);
        assert_eq!(tokens, vec![(SyntaxKind::STRING, "--")]);

        // Test 4: Four dashes
        let input = "----";
        let tokens = lex(input);
        assert_eq!(tokens, vec![(SyntaxKind::STRING, "----")]);
    }

    #[test]
    fn test_dash_in_different_scalar_contexts() {
        // Test kebab-case identifiers
        let input = "package-name: my-awesome-package-v2";
        let tokens = lex(input);
        assert_eq!(tokens[0], (SyntaxKind::STRING, "package-name"));
        assert_eq!(tokens[1], (SyntaxKind::COLON, ":"));
        assert_eq!(tokens[2], (SyntaxKind::WHITESPACE, " "));
        assert_eq!(tokens[3], (SyntaxKind::STRING, "my-awesome-package-v2"));

        // Test UUID-like strings
        let input = "id: 123e4567-e89b-12d3-a456-426614174000";
        let tokens = lex(input);
        assert_eq!(tokens[0], (SyntaxKind::STRING, "id"));
        assert_eq!(
            tokens[3],
            (SyntaxKind::STRING, "123e4567-e89b-12d3-a456-426614174000")
        );

        // Test command-line arguments. The two words are one plain
        // scalar, so they lex as a single token rather than being
        // rejoined by the parser.
        let input = "args: --verbose --log-level=debug";
        let tokens = lex(input);
        assert_eq!(
            tokens[3],
            (SyntaxKind::STRING, "--verbose --log-level=debug")
        );

        // Test negative numbers
        let input = "temperature: -40";
        let tokens = lex(input);
        // Negative numbers are tokenized as INT tokens
        assert_eq!(
            tokens
                .iter()
                .filter(|(kind, text)| *kind == SyntaxKind::INT && *text == "-40")
                .count(),
            1
        );

        // Test ranges
        let input = "range: 1-10";
        let tokens = lex(input);
        assert_eq!(
            tokens
                .iter()
                .filter(|(kind, text)| *kind == SyntaxKind::STRING && *text == "1-10")
                .count(),
            1
        );
    }

    #[test]
    fn test_sequence_markers_with_indentation() {
        // Test basic sequence
        let input = "- item1\n- item2";
        let tokens = lex(input);
        assert_eq!(tokens[0], (SyntaxKind::DASH, "-"));
        assert_eq!(tokens[1], (SyntaxKind::WHITESPACE, " "));
        assert_eq!(tokens[2], (SyntaxKind::STRING, "item1"));

        // Test indented sequence
        let input = "  - item1\n  - item2";
        let tokens = lex(input);
        assert_eq!(tokens[0], (SyntaxKind::INDENT, "  "));
        assert_eq!(tokens[1], (SyntaxKind::DASH, "-"));

        // The indented lines continue the plain scalar `item1`, so their
        // dashes are content: the value is `item1 - nested1 - nested2`.
        let input = "- item1\n  - nested1\n  - nested2\n- item2";
        let tokens = lex(input);
        let dash_tokens: Vec<_> = tokens
            .iter()
            .filter(|(kind, _)| *kind == SyntaxKind::DASH)
            .collect();
        assert_eq!(dash_tokens.len(), 2);

        // Test sequence with hyphenated values
        let input = "- first-item\n- second-item";
        let tokens = lex(input);
        assert_eq!(tokens[0], (SyntaxKind::DASH, "-"));
        assert_eq!(tokens[2], (SyntaxKind::STRING, "first-item"));
        assert_eq!(tokens[4], (SyntaxKind::DASH, "-"));
        assert_eq!(tokens[6], (SyntaxKind::STRING, "second-item"));
    }

    #[test]
    fn test_dash_after_colon() {
        // Test hyphen immediately after colon
        // According to YAML spec, "key:-value" is a single plain scalar
        // because the colon is not followed by whitespace
        let input = "key:-value";
        let tokens = lex(input);
        assert_eq!(tokens[0], (SyntaxKind::STRING, "key:-value"));

        // Test with space - this creates a mapping
        let input = "key: -value";
        let tokens = lex(input);
        assert_eq!(tokens[0], (SyntaxKind::STRING, "key"));
        assert_eq!(tokens[1], (SyntaxKind::COLON, ":"));
        assert_eq!(tokens[2], (SyntaxKind::WHITESPACE, " "));
        assert_eq!(tokens[3], (SyntaxKind::STRING, "-value"));
    }

    #[test]
    fn test_yaml_spec_compliant_colon_handling() {
        // Test that colons are handled according to YAML spec:
        // - Colon followed by whitespace indicates mapping
        // - Colon not followed by whitespace is part of plain scalar

        // URLs should be single scalars (no space after colon)
        let input = "http://example.com:8080/path";
        let tokens = lex(input);
        assert_eq!(tokens.len(), 1);
        assert_eq!(
            tokens[0],
            (SyntaxKind::STRING, "http://example.com:8080/path")
        );

        // Timestamps should be single scalars
        let input = "2024:12:31:23:59:59";
        let tokens = lex(input);
        assert_eq!(tokens.len(), 1);
        assert_eq!(tokens[0], (SyntaxKind::STRING, "2024:12:31:23:59:59"));

        // Key-value pairs need space after colon
        let input = "key: value";
        let tokens = lex(input);
        assert_eq!(tokens[0], (SyntaxKind::STRING, "key"));
        assert_eq!(tokens[1], (SyntaxKind::COLON, ":"));
        assert_eq!(tokens[2], (SyntaxKind::WHITESPACE, " "));
        assert_eq!(tokens[3], (SyntaxKind::STRING, "value"));

        // Without space, it's a single scalar
        let input = "key:value";
        let tokens = lex(input);
        assert_eq!(tokens.len(), 1);
        assert_eq!(tokens[0], (SyntaxKind::STRING, "key:value"));

        // Multiple colons without spaces
        let input = "a:b:c:d";
        let tokens = lex(input);
        assert_eq!(tokens.len(), 1);
        assert_eq!(tokens[0], (SyntaxKind::STRING, "a:b:c:d"));
    }

    #[test]
    fn test_block_scalar_with_chomping() {
        // Helper to count tokens by kind
        let count_kind = |toks: &[(SyntaxKind, &str)], k: SyntaxKind| {
            toks.iter().filter(|(kind, _)| *kind == k).count()
        };

        // Test literal block scalar with strip chomping
        let input = "text: |-\n  content";
        let tokens = lex(input);
        assert_eq!(count_kind(&tokens, SyntaxKind::PIPE), 1);
        assert_eq!(
            tokens
                .iter()
                .filter(|(kind, text)| *kind == SyntaxKind::STRING && *text == "-")
                .count(),
            1
        );

        // Test literal block scalar with keep chomping
        let input = "text: |+\n  content";
        let tokens = lex(input);
        assert_eq!(count_kind(&tokens, SyntaxKind::PIPE), 1);
        assert_eq!(count_kind(&tokens, SyntaxKind::PLUS), 1);

        // Test folded block scalar with strip chomping
        let input = "text: >-\n  content";
        let tokens = lex(input);
        assert_eq!(count_kind(&tokens, SyntaxKind::GREATER), 1);
        assert_eq!(
            tokens
                .iter()
                .filter(|(kind, text)| *kind == SyntaxKind::STRING && *text == "-")
                .count(),
            1
        );

        // Test with explicit indentation and chomping
        let input = "text: |2-\n  content";
        let tokens = lex(input);
        assert_eq!(count_kind(&tokens, SyntaxKind::PIPE), 1);
        // The "2-" after pipe gets read as one token because hyphens in scalars are included
        let has_2_token = tokens.iter().any(|(kind, text)| {
            (*kind == SyntaxKind::STRING || *kind == SyntaxKind::INT) && text.contains("2")
        });
        assert!(has_2_token, "expected a token containing '2'");
    }

    #[test]
    fn test_dash_edge_cases() {
        // Test trailing hyphen
        let input = "value-";
        let tokens = lex(input);
        assert_eq!(tokens[0], (SyntaxKind::STRING, "value-"));

        // Test leading hyphen (not a sequence marker)
        let input = "-value";
        let tokens = lex(input);
        assert_eq!(tokens[0], (SyntaxKind::STRING, "-value"));

        // Test multiple consecutive hyphens in scalar
        let input = "key: a---b";
        let tokens = lex(input);
        assert_eq!(
            tokens
                .iter()
                .filter(|(kind, text)| *kind == SyntaxKind::STRING && *text == "a---b")
                .count(),
            1
        );

        // Test hyphen at end of line
        let input = "key: value-\nnext: item";
        let tokens = lex(input);
        assert!(tokens
            .iter()
            .any(|(kind, text)| *kind == SyntaxKind::STRING && *text == "value-"));

        // Test mix of dashes and underscores
        let input = "snake_case-with-dash_mix";
        let tokens = lex(input);
        assert_eq!(tokens[0], (SyntaxKind::STRING, "snake_case-with-dash_mix"));
    }

    #[test]
    fn test_whitespace_validation_tab_indentation() {
        // Test tab character validation
        let input_with_tabs = "key: value\n\tindented_key: indented_value";
        let (tokens, errors) = lex_with_validation(input_with_tabs);

        // Should have detected tab indentation error
        assert_eq!(errors.len(), 1);
        assert_eq!(errors[0].category, WhitespaceErrorCategory::TabIndentation);
        assert_eq!(
            errors[0].message,
            "Tab character used for indentation (forbidden in YAML)"
        );

        // But should still tokenize correctly
        assert!(tokens
            .iter()
            .any(|(kind, text)| *kind == SyntaxKind::INDENT && text.contains('\t')));
    }

    #[test]
    fn test_whitespace_validation_line_endings() {
        // Test mixed line ending detection
        let input_mixed = "line1\nline2\r\nline3\rline4";
        let config = ValidationConfig {
            enforce_consistent_line_endings: true,
            max_line_length: None,
        };
        let (tokens, errors) = lex_with_validation_config(input_mixed, &config);

        // Should detect mixed line endings
        assert!(errors
            .iter()
            .any(|e| e.category == WhitespaceErrorCategory::MixedLineEndings));

        // Should still tokenize all line endings
        let newlines: Vec<_> = tokens
            .iter()
            .filter(|(kind, _)| *kind == SyntaxKind::NEWLINE)
            .collect();
        assert_eq!(newlines.len(), 3); // Three line endings
        assert_eq!(newlines[0].1, "\n");
        assert_eq!(newlines[1].1, "\r\n");
        assert_eq!(newlines[2].1, "\r");
    }

    #[test]
    fn test_whitespace_validation_line_length() {
        // Test line length validation
        let long_line = format!("key: {}", "a".repeat(150));
        let config = ValidationConfig {
            enforce_consistent_line_endings: false,
            max_line_length: Some(120),
        };
        let (_, errors) = lex_with_validation_config(&long_line, &config);

        // Should detect line too long
        assert_eq!(errors.len(), 1);
        assert_eq!(errors[0].category, WhitespaceErrorCategory::LineTooLong);
        assert_eq!(errors[0].message, "Line too long (155 > 120 characters)");
    }

    #[test]
    fn test_whitespace_validation_disabled() {
        // Test with validation disabled
        let input_with_issues = "key: value\n\tindented: with_tabs\n";
        let config = ValidationConfig {
            enforce_consistent_line_endings: false,
            max_line_length: None,
        };
        let (tokens, errors) = lex_with_validation_config(input_with_issues, &config);

        // Should still detect tab indentation (always enforced in YAML)
        assert_eq!(errors.len(), 1);
        assert_eq!(errors[0].category, WhitespaceErrorCategory::TabIndentation);

        // Should tokenize normally
        assert!(!tokens.is_empty());
    }

    #[test]
    fn test_dash_in_flow_collections() {
        // Test dash in flow sequence
        let input = "[item-one, item-two]";
        let tokens = lex(input);
        assert_eq!(tokens[0], (SyntaxKind::LEFT_BRACKET, "["));
        assert_eq!(tokens[1], (SyntaxKind::STRING, "item-one"));
        assert_eq!(tokens[2], (SyntaxKind::COMMA, ","));
        assert_eq!(tokens[4], (SyntaxKind::STRING, "item-two"));
        assert_eq!(tokens[5], (SyntaxKind::RIGHT_BRACKET, "]"));

        // Test dash in flow mapping
        let input = "{kebab-key: kebab-value}";
        let tokens = lex(input);
        assert_eq!(tokens[0], (SyntaxKind::LEFT_BRACE, "{"));
        assert_eq!(tokens[1], (SyntaxKind::STRING, "kebab-key"));
        assert_eq!(tokens[2], (SyntaxKind::COLON, ":"));
        assert_eq!(tokens[4], (SyntaxKind::STRING, "kebab-value"));
        assert_eq!(tokens[5], (SyntaxKind::RIGHT_BRACE, "}"));
    }

    #[test]
    fn test_dash_with_quotes() {
        // Quoted strings should preserve everything inside as STRING tokens
        let input = r#"key: "- not a sequence marker""#;
        let tokens = lex(input);
        assert_eq!(
            tokens
                .iter()
                .filter(|(kind, text)| {
                    *kind == SyntaxKind::STRING && *text == "\"- not a sequence marker\""
                })
                .count(),
            1
        );

        let input = r#"key: '- also not a sequence marker'"#;
        let tokens = lex(input);
        assert_eq!(
            tokens
                .iter()
                .filter(|(kind, text)| {
                    *kind == SyntaxKind::STRING && *text == "'- also not a sequence marker'"
                })
                .count(),
            1
        );
    }

    #[test]
    fn test_dash_in_multiline_values() {
        // Test multiline with dashes: plain scalars absorb intra-line
        // whitespace, so the first line becomes one STRING token ending
        // in the trailing hyphen (hyphen-at-end-of-word is scalar content,
        // not a sequence marker mid-scalar).
        let input = "description: This is a multi-\n  line value with dashes";
        let tokens = lex(input);
        assert!(tokens
            .iter()
            .any(|(kind, text)| *kind == SyntaxKind::STRING && *text == "This is a multi-"));

        // A line continuing a plain scalar is content, dash included.
        let input = "text: value\n  - but this is not a sequence";
        let tokens = lex(input);
        assert_eq!(
            &tokens[tokens.len() - 2..],
            [
                (SyntaxKind::INDENT, "  "),
                (SyntaxKind::STRING, "- but this is not a sequence"),
            ]
        );
    }

    #[test]
    fn test_dash_special_yaml_values() {
        // Test that special YAML values with dashes work
        let input = "date: 2024-01-15";
        let tokens = lex(input);
        assert!(tokens
            .iter()
            .any(|(kind, text)| *kind == SyntaxKind::STRING && *text == "2024-01-15"));

        // Test ISO timestamp - gets tokenized as multiple parts due to hyphens
        let input = "timestamp: 2024-01-15T10:30:00-05:00";
        let tokens = lex(input);
        // The timestamp is split into multiple tokens but parses correctly
        assert!(tokens.iter().any(
            |(kind, text)| *kind == SyntaxKind::STRING && *text == "2024-01-15T10:30:00-05:00"
        ));

        // Test version strings
        let input = "version: 1.0.0-beta.1";
        let tokens = lex(input);
        assert!(tokens
            .iter()
            .any(|(kind, text)| *kind == SyntaxKind::STRING && *text == "1.0.0-beta.1"));
    }

    #[test]
    fn test_block_scalar_indentation_indicator_sets_body_column() {
        use SyntaxKind::*;
        // With `|2` the body sits at column 2, so the `y # z` and `[a` lines
        // belong to it even though the first line is indented further.
        assert_eq!(
            lex("- |2\n   x\n  y # z\n  [a\n"),
            vec![
                (DASH, "-"),
                (WHITESPACE, " "),
                (PIPE, "|"),
                (INT, "2"),
                (NEWLINE, "\n"),
                (INDENT, "   "),
                (STRING, "x"),
                (NEWLINE, "\n"),
                (INDENT, "  "),
                (STRING, "y"),
                (WHITESPACE, " "),
                (STRING, "# z"),
                (NEWLINE, "\n"),
                (INDENT, "  "),
                (STRING, "[a"),
                (NEWLINE, "\n"),
            ]
        );
        assert_eq!(
            lex("- |2\r\n   x\r\n  y # z\r\n  [a\r\n"),
            vec![
                (DASH, "-"),
                (WHITESPACE, " "),
                (PIPE, "|"),
                (INT, "2"),
                (NEWLINE, "\r\n"),
                (INDENT, "   "),
                (STRING, "x"),
                (NEWLINE, "\r\n"),
                (INDENT, "  "),
                (STRING, "y"),
                (WHITESPACE, " "),
                (STRING, "# z"),
                (NEWLINE, "\r\n"),
                (INDENT, "  "),
                (STRING, "[a"),
                (NEWLINE, "\r\n"),
            ]
        );
    }

    #[test]
    fn test_block_scalar_indentation_indicator_counts_from_parent() {
        use SyntaxKind::*;
        // The indicator counts from the key's column, 2 here, not from the
        // line's indentation or the dash.
        assert_eq!(
            lex("- k: >-2\n     x\n    [a\n"),
            vec![
                (DASH, "-"),
                (WHITESPACE, " "),
                (STRING, "k"),
                (COLON, ":"),
                (WHITESPACE, " "),
                (GREATER, ">"),
                (INT, "-2"),
                (NEWLINE, "\n"),
                (INDENT, "     "),
                (STRING, "x"),
                (NEWLINE, "\n"),
                (INDENT, "    "),
                (STRING, "[a"),
                (NEWLINE, "\n"),
            ]
        );
        // A header on its own line counts from the key on the line before.
        assert_eq!(
            lex("k:\n  |1\n   x\n # z\n"),
            vec![
                (STRING, "k"),
                (COLON, ":"),
                (NEWLINE, "\n"),
                (INDENT, "  "),
                (PIPE, "|"),
                (INT, "1"),
                (NEWLINE, "\n"),
                (INDENT, "   "),
                (STRING, "x"),
                (NEWLINE, "\n"),
                (INDENT, " "),
                (STRING, "# z"),
                (NEWLINE, "\n"),
            ]
        );
        // A line left of the indicated column ends the body.
        assert_eq!(
            lex("- k: |1\n   x\n  j: [a]\n"),
            vec![
                (DASH, "-"),
                (WHITESPACE, " "),
                (STRING, "k"),
                (COLON, ":"),
                (WHITESPACE, " "),
                (PIPE, "|"),
                (INT, "1"),
                (NEWLINE, "\n"),
                (INDENT, "   "),
                (STRING, "x"),
                (NEWLINE, "\n"),
                (INDENT, "  "),
                (STRING, "j"),
                (COLON, ":"),
                (WHITESPACE, " "),
                (LEFT_BRACKET, "["),
                (STRING, "a"),
                (RIGHT_BRACKET, "]"),
                (NEWLINE, "\n"),
            ]
        );
    }

    #[test]
    fn test_block_scalar_indentation_indicator_in_explicit_value() {
        use SyntaxKind::*;
        // The value of an explicit key counts from the mapping's column, 2
        // here, whether the header follows the `:` or sits on its own line.
        let head = [
            (STRING, "x"),
            (COLON, ":"),
            (NEWLINE, "\n"),
            (INDENT, "  "),
            (QUESTION, "?"),
            (WHITESPACE, " "),
            (STRING, "a"),
            (NEWLINE, "\n"),
            (INDENT, "  "),
            (COLON, ":"),
        ];
        let tail = [
            (INDENT, "     "),
            (STRING, "x"),
            (NEWLINE, "\n"),
            (INDENT, "  "),
            (LEFT_BRACKET, "["),
            (STRING, "a"),
            (RIGHT_BRACKET, "]"),
            (NEWLINE, "\n"),
        ];
        assert_eq!(
            lex("x:\n  ? a\n  : |1\n     x\n  [a]\n"),
            [
                &head[..],
                &[(WHITESPACE, " "), (PIPE, "|"), (INT, "1"), (NEWLINE, "\n")],
                &tail,
            ]
            .concat()
        );
        assert_eq!(
            lex("x:\n  ? a\n  :\n    |1\n     x\n  [a]\n"),
            [
                &head[..],
                &[
                    (NEWLINE, "\n"),
                    (INDENT, "    "),
                    (PIPE, "|"),
                    (INT, "1"),
                    (NEWLINE, "\n"),
                ],
                &tail,
            ]
            .concat()
        );
    }

    #[test]
    fn test_block_scalar_indentation_indicator_in_compact_explicit_value() {
        use SyntaxKind::*;
        // A compact mapping after an explicit `:` sits at its first key's
        // column, 2 here, and the indicator counts from there.
        let head = [
            (QUESTION, "?"),
            (WHITESPACE, " "),
            (STRING, "a"),
            (NEWLINE, "\n"),
            (COLON, ":"),
            (WHITESPACE, " "),
            (STRING, "b"),
            (COLON, ":"),
        ];
        let tail = [
            (INDENT, "     "),
            (STRING, "x"),
            (NEWLINE, "\n"),
            (INDENT, "  "),
            (STRING, "c"),
            (COLON, ":"),
            (WHITESPACE, " "),
            (LEFT_BRACKET, "["),
            (STRING, "d"),
            (RIGHT_BRACKET, "]"),
            (NEWLINE, "\n"),
        ];
        assert_eq!(
            lex("? a\n: b: |1\n     x\n  c: [d]\n"),
            [
                &head[..],
                &[(WHITESPACE, " "), (PIPE, "|"), (INT, "1"), (NEWLINE, "\n")],
                &tail,
            ]
            .concat()
        );
        assert_eq!(
            lex("? a\n: b:\n    |1\n     x\n  c: [d]\n"),
            [
                &head[..],
                &[
                    (NEWLINE, "\n"),
                    (INDENT, "    "),
                    (PIPE, "|"),
                    (INT, "1"),
                    (NEWLINE, "\n"),
                ],
                &tail,
            ]
            .concat()
        );
        // Inside a sequence entry the compact mapping is at column 4.
        assert_eq!(
            lex("- ? a\n  : b: |1\n       x\n    c: [d]\n"),
            vec![
                (DASH, "-"),
                (WHITESPACE, " "),
                (QUESTION, "?"),
                (WHITESPACE, " "),
                (STRING, "a"),
                (NEWLINE, "\n"),
                (INDENT, "  "),
                (COLON, ":"),
                (WHITESPACE, " "),
                (STRING, "b"),
                (COLON, ":"),
                (WHITESPACE, " "),
                (PIPE, "|"),
                (INT, "1"),
                (NEWLINE, "\n"),
                (INDENT, "       "),
                (STRING, "x"),
                (NEWLINE, "\n"),
                (INDENT, "    "),
                (STRING, "c"),
                (COLON, ":"),
                (WHITESPACE, " "),
                (LEFT_BRACKET, "["),
                (STRING, "d"),
                (RIGHT_BRACKET, "]"),
                (NEWLINE, "\n"),
            ]
        );
    }

    #[test]
    fn test_flow_indicators_in_block_scalar() {
        // Flow indicators should be allowed in block context scalars
        // This is valid YAML: the curly braces are part of the scalar value
        let input = "key: unix:///Users/${metadata.username}/path";
        let tokens = lex(input);
        assert_eq!(tokens.len(), 4);
        assert_eq!(tokens[0], (SyntaxKind::STRING, "key"));
        assert_eq!(tokens[1], (SyntaxKind::COLON, ":"));
        assert_eq!(tokens[2], (SyntaxKind::WHITESPACE, " "));
        assert_eq!(
            tokens[3],
            (
                SyntaxKind::STRING,
                "unix:///Users/${metadata.username}/path"
            )
        );
    }

    #[test]
    fn test_block_scalar_body_tab_is_not_indentation() {
        // The body is indented two, so the tab is content and the next line
        // is still body: its `"` must not start a quoted scalar.
        use SyntaxKind::*;
        assert_eq!(
            lex("- |-\n  \t[\n  ]\"\n- b\n"),
            vec![
                (DASH, "-"),
                (WHITESPACE, " "),
                (PIPE, "|"),
                (STRING, "-"),
                (NEWLINE, "\n"),
                (INDENT, "  \t"),
                (STRING, "["),
                (NEWLINE, "\n"),
                (INDENT, "  "),
                (STRING, "]\""),
                (NEWLINE, "\n"),
                (DASH, "-"),
                (WHITESPACE, " "),
                (STRING, "b"),
                (NEWLINE, "\n"),
            ]
        );
    }

    #[test]
    fn test_unterminated_verbatim_tag_ends_at_line_break() {
        use SyntaxKind::*;
        for line_break in ["\n", "\r\n", "\r"] {
            let input = format!("!<x{line_break}a");
            assert_eq!(
                lex(&input),
                vec![(TAG, "!<x"), (NEWLINE, line_break), (STRING, "a")]
            );
        }
    }
}