kataan 0.0.2

//! The lexer: ECMAScript source text → a stream of [`Token`]s.
//!
//! This is a hand-written, single-pass, allocation-light tokenizer. It is the
//! first stage of the pipeline and deliberately self-contained: it depends
//! only on `core`/`alloc` and produces tokens carrying [`Span`]s into the
//! original source.
//!
//! ## What it handles
//!
//! - All ECMAScript punctuators and assignment operators, optional chaining
//!   (`?.`) and nullish coalescing (`??`/`??=`).
//! - Keywords vs identifiers (Unicode identifier start/continue via a compact
//!   classifier), private names (`#field`).
//! - Numeric literals: decimal, hex/octal/binary, exponents, `BigInt` (`n`),
//!   and numeric separators (`1_000`).
//! - String literals with the full escape grammar, including line
//!   continuations.
//! - Template literals — including nested substitutions and nested templates —
//!   via an internal brace-kind stack, so the lexer is fully self-contained
//!   (no parser feedback needed for the common cases).
//! - The regex-vs-division ambiguity, resolved with the standard
//!   previous-significant-token heuristic.
//! - Line terminators and comments, recording for each token whether a line
//!   terminator preceded it (the signal the parser needs for Automatic
//!   Semicolon Insertion).
//!
//! ## What it defers
//!
//! Full template re-lexing driven by the parser (needed only for a few
//! pathological `}`-after-substitution cases the brace-stack cannot
//! disambiguate on its own) lands with the parser in Phase B.

mod token;

#[cfg(test)]
mod tests;

pub use token::{Keyword, Token, TokenKind};

use crate::common::Span;
use crate::error::{Error, Result};
use alloc::vec::Vec;

/// Tracks why we are inside a `{ … }`, so a `}` can be disambiguated between
/// "close a block/object" and "resume a template literal after `${ … }`".
#[derive(Clone, Copy, PartialEq, Eq)]
enum BraceKind {
    /// An ordinary `{` (block, object literal, destructuring, …).
    Normal,
    /// The `{` of a template substitution `${ … }`; its `}` resumes the
    /// template body.
    TemplateSubstitution,
}

/// A streaming ECMAScript tokenizer over a borrowed source string.
///
/// Drive it with [`Lexer::next_token`] until it yields [`TokenKind::Eof`], or
/// collect everything at once with [`Lexer::tokenize`].
pub struct Lexer<'src> {
    /// The full source text.
    source: &'src str,
    /// Remaining bytes, as raw bytes for fast ASCII dispatch. Always a valid
    /// UTF-8 boundary at `pos`.
    bytes: &'src [u8],
    /// Current byte offset into `source`.
    pos: usize,
    /// The kind of the previous significant (non-trivia) token, used to
    /// resolve the `/` regex-vs-division ambiguity. `None` at start of input.
    prev_significant: Option<TokenKind>,
    /// Open-brace stack for template-substitution tracking.
    brace_stack: Vec<BraceKind>,
}

impl<'src> Lexer<'src> {
    /// Creates a lexer over `source`.
    #[must_use]
    pub fn new(source: &'src str) -> Self {
        Self {
            source,
            bytes: source.as_bytes(),
            pos: 0,
            prev_significant: None,
            brace_stack: Vec::new(),
        }
    }

    /// The source text this lexer is scanning.
    #[inline]
    #[must_use]
    pub fn source(&self) -> &'src str {
        self.source
    }

    /// Tokenizes the entire input into a vector ending with an
    /// [`TokenKind::Eof`] token. Returns the first lexical [`Error`]
    /// encountered, if any.
    pub fn tokenize(mut self) -> Result<Vec<Token>> {
        let mut out = Vec::new();
        loop {
            let tok = self.next_token()?;
            let is_eof = tok.kind == TokenKind::Eof;
            out.push(tok);
            if is_eof {
                return Ok(out);
            }
        }
    }

    /// Produces the next token, consuming any leading whitespace/comments. The
    /// returned token's [`Token::newline_before`] records whether a line
    /// terminator was skipped before it (for ASI).
    pub fn next_token(&mut self) -> Result<Token> {
        let newline_before = self.skip_trivia();
        let start = self.pos;

        let Some(c) = self.peek() else {
            return Ok(self.make(TokenKind::Eof, start, newline_before));
        };

        let kind = match c {
            b'{' => {
                self.advance();
                self.brace_stack.push(BraceKind::Normal);
                TokenKind::LBrace
            }
            b'}' => {
                // A `}` that closes a template substitution resumes template
                // scanning rather than emitting a plain `}`.
                if matches!(
                    self.brace_stack.last(),
                    Some(BraceKind::TemplateSubstitution)
                ) {
                    self.brace_stack.pop();
                    return self.read_template_continuation(start, newline_before);
                }
                self.advance();
                self.brace_stack.pop();
                TokenKind::RBrace
            }
            b'(' => self.single(TokenKind::LParen),
            b')' => self.single(TokenKind::RParen),
            b'[' => self.single(TokenKind::LBracket),
            b']' => self.single(TokenKind::RBracket),
            b';' => self.single(TokenKind::Semicolon),
            b',' => self.single(TokenKind::Comma),
            b'~' => self.single(TokenKind::Tilde),
            b':' => self.single(TokenKind::Colon),
            b'?' => self.read_question(),
            // `.` may begin a number (`.5`), the spread `...`, or a member `.`.
            b'.' if matches!(self.peek_at(1), Some(b'0'..=b'9')) => self.read_number()?,
            b'.' => self.read_dot(),
            b'<' => self.read_lt(),
            b'>' => self.read_gt(),
            b'=' => self.read_eq(),
            b'!' => self.read_bang(),
            b'+' => self.read_plus(),
            b'-' => self.read_minus(),
            b'*' => self.read_star(),
            b'%' => self.read_percent(),
            b'&' => self.read_amp(),
            b'|' => self.read_pipe(),
            b'^' => self.read_caret(),
            b'/' => return self.read_slash(start, newline_before),
            b'"' | b'\'' => self.read_string(c)?,
            b'`' => return self.read_template_start(start, newline_before),
            b'#' => self.read_private_name()?,
            b'0'..=b'9' => self.read_number()?,
            _ => {
                if is_identifier_start_byte(c)
                    || (c >= 0x80 && self.peek_char().is_some_and(is_identifier_start_char))
                {
                    self.read_identifier_or_keyword()?
                } else {
                    // Consume the whole (possibly multi-byte) char before
                    // reporting, so the span is correct and `advance` isn't
                    // mid-codepoint.
                    let ch = self.peek_char().unwrap_or(c as char);
                    self.advance_char(ch);
                    return Err(Error::syntax(
                        alloc::format!("unexpected character {ch:?}"),
                        Span::new(start as u32, self.pos as u32),
                    ));
                }
            }
        };

        Ok(self.make(kind, start, newline_before))
    }

    // --- trivia ---------------------------------------------------------

    /// Skips whitespace and comments. Returns whether at least one line
    /// terminator was crossed.
    fn skip_trivia(&mut self) -> bool {
        let mut newline = false;
        loop {
            let Some(c) = self.peek() else { return newline };
            match c {
                // ASCII whitespace.
                b' ' | b'\t' | 0x0b | 0x0c => self.advance(),
                // Line terminators (LF, CR). CRLF counts once.
                b'\n' => {
                    newline = true;
                    self.advance();
                }
                b'\r' => {
                    newline = true;
                    self.advance();
                    if self.peek() == Some(b'\n') {
                        self.advance();
                    }
                }
                b'/' => match self.peek_at(1) {
                    Some(b'/') => self.skip_line_comment(),
                    Some(b'*') => newline |= self.skip_block_comment(),
                    _ => return newline,
                },
                // Non-ASCII whitespace / line terminators (NBSP, BOM, U+2028,
                // U+2029, the Zs category…). Decode one char to classify.
                _ if c >= 0x80 => {
                    let ch = self.peek_char().expect("non-empty");
                    if is_unicode_line_terminator(ch) {
                        newline = true;
                        self.advance_char(ch);
                    } else if is_unicode_whitespace(ch) {
                        self.advance_char(ch);
                    } else {
                        return newline;
                    }
                }
                _ => return newline,
            }
        }
    }

    fn skip_line_comment(&mut self) {
        // Consume `//` then everything up to (not including) a line terminator.
        self.advance();
        self.advance();
        while let Some(c) = self.peek() {
            if c == b'\n' || c == b'\r' {
                break;
            }
            if c >= 0x80 {
                let ch = self.peek_char().expect("non-empty");
                if is_unicode_line_terminator(ch) {
                    break;
                }
                self.advance_char(ch);
            } else {
                self.advance();
            }
        }
    }

    /// Skips a `/* … */` comment. Returns whether it contained a line
    /// terminator (which, per spec, makes it act as one for ASI).
    fn skip_block_comment(&mut self) -> bool {
        self.advance();
        self.advance();
        let mut newline = false;
        while let Some(c) = self.peek() {
            if c == b'*' && self.peek_at(1) == Some(b'/') {
                self.advance();
                self.advance();
                return newline;
            }
            if c == b'\n' || c == b'\r' {
                newline = true;
                self.advance();
            } else if c >= 0x80 {
                let ch = self.peek_char().expect("non-empty");
                if is_unicode_line_terminator(ch) {
                    newline = true;
                }
                self.advance_char(ch);
            } else {
                self.advance();
            }
        }
        newline
    }

    // --- multi-character punctuators ------------------------------------

    fn read_question(&mut self) -> TokenKind {
        self.advance();
        match self.peek() {
            // `?.` but only as optional chaining, not `?.5` (which is `?`
            // then `.5`). The spec carves out a digit after `?.`.
            Some(b'.') if !matches!(self.peek_at(1), Some(b'0'..=b'9')) => {
                self.advance();
                TokenKind::QuestionDot
            }
            Some(b'?') => {
                self.advance();
                if self.peek() == Some(b'=') {
                    self.advance();
                    TokenKind::QuestionQuestionEq
                } else {
                    TokenKind::QuestionQuestion
                }
            }
            _ => TokenKind::Question,
        }
    }

    /// `.` — member access or the `...` spread. The `.5` numeric case is
    /// routed to [`Self::read_number`] by the caller before reaching here.
    fn read_dot(&mut self) -> TokenKind {
        self.advance();
        if self.peek() == Some(b'.') && self.peek_at(1) == Some(b'.') {
            self.advance();
            self.advance();
            TokenKind::DotDotDot
        } else {
            TokenKind::Dot
        }
    }

    fn read_lt(&mut self) -> TokenKind {
        self.advance();
        match self.peek() {
            Some(b'=') => self.single(TokenKind::LtEq),
            Some(b'<') => {
                self.advance();
                if self.peek() == Some(b'=') {
                    self.single(TokenKind::ShlEq)
                } else {
                    TokenKind::Shl
                }
            }
            _ => TokenKind::Lt,
        }
    }

    fn read_gt(&mut self) -> TokenKind {
        self.advance();
        match self.peek() {
            Some(b'=') => self.single(TokenKind::GtEq),
            Some(b'>') => {
                self.advance();
                match self.peek() {
                    Some(b'=') => self.single(TokenKind::ShrEq),
                    Some(b'>') => {
                        self.advance();
                        if self.peek() == Some(b'=') {
                            self.single(TokenKind::UshrEq)
                        } else {
                            TokenKind::Ushr
                        }
                    }
                    _ => TokenKind::Shr,
                }
            }
            _ => TokenKind::Gt,
        }
    }

    fn read_eq(&mut self) -> TokenKind {
        self.advance();
        match self.peek() {
            Some(b'=') => {
                self.advance();
                if self.peek() == Some(b'=') {
                    self.single(TokenKind::EqEqEq)
                } else {
                    TokenKind::EqEq
                }
            }
            Some(b'>') => self.single(TokenKind::Arrow),
            _ => TokenKind::Eq,
        }
    }

    fn read_bang(&mut self) -> TokenKind {
        self.advance();
        if self.peek() == Some(b'=') {
            self.advance();
            if self.peek() == Some(b'=') {
                self.single(TokenKind::BangEqEq)
            } else {
                TokenKind::BangEq
            }
        } else {
            TokenKind::Bang
        }
    }

    fn read_plus(&mut self) -> TokenKind {
        self.advance();
        match self.peek() {
            Some(b'+') => self.single(TokenKind::PlusPlus),
            Some(b'=') => self.single(TokenKind::PlusEq),
            _ => TokenKind::Plus,
        }
    }

    fn read_minus(&mut self) -> TokenKind {
        self.advance();
        match self.peek() {
            Some(b'-') => self.single(TokenKind::MinusMinus),
            Some(b'=') => self.single(TokenKind::MinusEq),
            _ => TokenKind::Minus,
        }
    }

    fn read_star(&mut self) -> TokenKind {
        self.advance();
        match self.peek() {
            Some(b'*') => {
                self.advance();
                if self.peek() == Some(b'=') {
                    self.single(TokenKind::StarStarEq)
                } else {
                    TokenKind::StarStar
                }
            }
            Some(b'=') => self.single(TokenKind::StarEq),
            _ => TokenKind::Star,
        }
    }

    fn read_percent(&mut self) -> TokenKind {
        self.advance();
        if self.peek() == Some(b'=') {
            self.single(TokenKind::PercentEq)
        } else {
            TokenKind::Percent
        }
    }

    fn read_amp(&mut self) -> TokenKind {
        self.advance();
        match self.peek() {
            Some(b'&') => {
                self.advance();
                if self.peek() == Some(b'=') {
                    self.single(TokenKind::AmpAmpEq)
                } else {
                    TokenKind::AmpAmp
                }
            }
            Some(b'=') => self.single(TokenKind::AmpEq),
            _ => TokenKind::Amp,
        }
    }

    fn read_pipe(&mut self) -> TokenKind {
        self.advance();
        match self.peek() {
            Some(b'|') => {
                self.advance();
                if self.peek() == Some(b'=') {
                    self.single(TokenKind::PipePipeEq)
                } else {
                    TokenKind::PipePipe
                }
            }
            Some(b'=') => self.single(TokenKind::PipeEq),
            _ => TokenKind::Pipe,
        }
    }

    fn read_caret(&mut self) -> TokenKind {
        self.advance();
        if self.peek() == Some(b'=') {
            self.single(TokenKind::CaretEq)
        } else {
            TokenKind::Caret
        }
    }

    /// `/` — either a comment (already handled in trivia), a division
    /// operator, or the start of a regular-expression literal, decided by the
    /// previous significant token.
    fn read_slash(&mut self, start: usize, newline_before: bool) -> Result<Token> {
        if self.regex_allowed() {
            return self.read_regex(start, newline_before);
        }
        self.advance();
        let kind = if self.peek() == Some(b'=') {
            self.single(TokenKind::SlashEq)
        } else {
            TokenKind::Slash
        };
        Ok(self.make(kind, start, newline_before))
    }

    // --- literals -------------------------------------------------------

    fn read_string(&mut self, quote: u8) -> Result<TokenKind> {
        let start = self.pos;
        self.advance(); // opening quote
        loop {
            let Some(c) = self.peek() else {
                return Err(Error::syntax(
                    "unterminated string literal",
                    Span::new(start as u32, self.pos as u32),
                ));
            };
            match c {
                _ if c == quote => {
                    self.advance();
                    return Ok(TokenKind::String);
                }
                b'\\' => {
                    self.advance();
                    self.consume_escape_tail(start)?;
                }
                b'\n' | b'\r' => {
                    return Err(Error::syntax(
                        "unterminated string literal (line terminator in string)",
                        Span::new(start as u32, self.pos as u32),
                    ));
                }
                _ => self.advance_any(),
            }
        }
    }

    /// Consumes the character(s) after a `\` inside a string/template. We do
    /// not decode the value here (that is the parser's cooked-value step); we
    /// only consume the right number of source bytes so scanning stays in
    /// sync, while validating the few escapes that have a fixed shape.
    fn consume_escape_tail(&mut self, start: usize) -> Result<()> {
        let Some(c) = self.peek() else {
            return Err(Error::syntax(
                "unterminated escape sequence",
                Span::new(start as u32, self.pos as u32),
            ));
        };
        match c {
            // Line continuation: `\` followed by a line terminator.
            b'\n' => self.advance(),
            b'\r' => {
                self.advance();
                if self.peek() == Some(b'\n') {
                    self.advance();
                }
            }
            b'x' => {
                self.advance();
                for _ in 0..2 {
                    if !self.peek().is_some_and(|b| b.is_ascii_hexdigit()) {
                        return Err(Error::syntax(
                            "invalid hexadecimal escape sequence",
                            Span::new(start as u32, self.pos as u32),
                        ));
                    }
                    self.advance();
                }
            }
            b'u' => {
                self.advance();
                self.consume_unicode_escape(start)?;
            }
            _ => self.advance_any(),
        }
        Ok(())
    }

    /// Consumes the body of a `\u` escape: either `\uXXXX` or `\u{ … }`.
    fn consume_unicode_escape(&mut self, start: usize) -> Result<()> {
        if self.peek() == Some(b'{') {
            self.advance();
            let mut any = false;
            while self.peek().is_some_and(|b| b.is_ascii_hexdigit()) {
                any = true;
                self.advance();
            }
            if !any || self.peek() != Some(b'}') {
                return Err(Error::syntax(
                    "invalid Unicode code-point escape",
                    Span::new(start as u32, self.pos as u32),
                ));
            }
            self.advance(); // `}`
        } else {
            for _ in 0..4 {
                if !self.peek().is_some_and(|b| b.is_ascii_hexdigit()) {
                    return Err(Error::syntax(
                        "invalid Unicode escape sequence",
                        Span::new(start as u32, self.pos as u32),
                    ));
                }
                self.advance();
            }
        }
        Ok(())
    }

    /// Scans from a backtick: emits either a [`TokenKind::NoSubstitutionTemplate`]
    /// (no `${`) or a [`TokenKind::TemplateHead`] (up to and including the first
    /// `${`), pushing a substitution marker so the matching `}` resumes here.
    fn read_template_start(&mut self, start: usize, newline_before: bool) -> Result<Token> {
        self.advance(); // opening backtick
        let kind = self.scan_template_body(start)?;
        if kind == TokenKind::TemplateHead {
            self.brace_stack.push(BraceKind::TemplateSubstitution);
        }
        Ok(self.make(kind, start, newline_before))
    }

    /// Scans from the `}` that closes a substitution: emits either a
    /// [`TokenKind::TemplateMiddle`] (another `${` follows) or a
    /// [`TokenKind::TemplateTail`] (closing backtick).
    fn read_template_continuation(&mut self, start: usize, newline_before: bool) -> Result<Token> {
        self.advance(); // the `}`
        let kind = match self.scan_template_body(start)? {
            TokenKind::NoSubstitutionTemplate => TokenKind::TemplateTail,
            TokenKind::TemplateHead => {
                self.brace_stack.push(BraceKind::TemplateSubstitution);
                TokenKind::TemplateMiddle
            }
            other => other,
        };
        Ok(self.make(kind, start, newline_before))
    }

    /// Shared template-body scanner. Assumes the introducer (backtick or `}`)
    /// has been consumed. Returns [`TokenKind::NoSubstitutionTemplate`] if it
    /// reached a closing backtick, or [`TokenKind::TemplateHead`] if it
    /// reached a `${`.
    fn scan_template_body(&mut self, start: usize) -> Result<TokenKind> {
        loop {
            let Some(c) = self.peek() else {
                return Err(Error::syntax(
                    "unterminated template literal",
                    Span::new(start as u32, self.pos as u32),
                ));
            };
            match c {
                b'`' => {
                    self.advance();
                    return Ok(TokenKind::NoSubstitutionTemplate);
                }
                b'$' if self.peek_at(1) == Some(b'{') => {
                    self.advance();
                    self.advance();
                    return Ok(TokenKind::TemplateHead);
                }
                b'\\' => {
                    self.advance();
                    // A `\` escapes the next char (incl. backtick and `$`); we
                    // just consume one unit so scanning stays in sync.
                    self.advance_any();
                }
                _ => self.advance_any(),
            }
        }
    }

    /// A regular-expression literal `/pattern/flags`. Handles character
    /// classes (`[...]`, inside which `/` is literal) and escapes.
    fn read_regex(&mut self, start: usize, newline_before: bool) -> Result<Token> {
        self.advance(); // opening `/`
        let mut in_class = false;
        loop {
            let Some(c) = self.peek() else {
                return Err(Error::syntax(
                    "unterminated regular expression literal",
                    Span::new(start as u32, self.pos as u32),
                ));
            };
            match c {
                b'\n' | b'\r' => {
                    return Err(Error::syntax(
                        "unterminated regular expression literal (line terminator)",
                        Span::new(start as u32, self.pos as u32),
                    ));
                }
                b'\\' => {
                    self.advance();
                    if self.peek().is_some_and(|b| b == b'\n' || b == b'\r') {
                        return Err(Error::syntax(
                            "unterminated regular expression literal",
                            Span::new(start as u32, self.pos as u32),
                        ));
                    }
                    self.advance_any();
                }
                b'[' => {
                    in_class = true;
                    self.advance();
                }
                b']' => {
                    in_class = false;
                    self.advance();
                }
                b'/' if !in_class => {
                    self.advance();
                    break;
                }
                _ => self.advance_any(),
            }
        }
        // Flags: identifier-continue characters immediately after the closing
        // slash.
        while let Some(c) = self.peek() {
            if c < 0x80 {
                if is_identifier_part_byte(c) {
                    self.advance();
                } else {
                    break;
                }
            } else {
                let ch = self.peek_char().expect("non-empty");
                if is_identifier_part_char(ch) {
                    self.advance_char(ch);
                } else {
                    break;
                }
            }
        }
        Ok(self.make(TokenKind::Regex, start, newline_before))
    }

    fn read_private_name(&mut self) -> Result<TokenKind> {
        let start = self.pos;
        self.advance(); // `#`
        match self.peek() {
            Some(c)
                if is_identifier_start_byte(c)
                    || (c >= 0x80 && self.peek_char().is_some_and(is_identifier_start_char)) =>
            {
                self.read_identifier_tail();
                Ok(TokenKind::PrivateName)
            }
            _ => Err(Error::syntax(
                "expected an identifier after `#`",
                Span::new(start as u32, self.pos as u32),
            )),
        }
    }

    fn read_number(&mut self) -> Result<TokenKind> {
        let start = self.pos;
        let first = self.peek().expect("called with a digit or dot");

        if first == b'0' {
            match self.peek_at(1) {
                Some(b'x' | b'X') => return self.read_radix_number(16, start),
                Some(b'o' | b'O') => return self.read_radix_number(8, start),
                Some(b'b' | b'B') => return self.read_radix_number(2, start),
                _ => {}
            }
        }

        // Integer part (decimal), allowing numeric separators.
        if first == b'.' {
            self.advance(); // `.`
            self.read_decimal_digits()?;
            self.read_exponent()?;
            return Ok(TokenKind::Number);
        }

        self.read_decimal_digits()?;

        // BigInt suffix is only valid for an integer with no fraction/exponent.
        if self.peek() == Some(b'n') {
            self.advance();
            return Ok(TokenKind::BigInt);
        }

        let mut is_float = false;
        if self.peek() == Some(b'.') {
            is_float = true;
            self.advance();
            // Fractional digits are optional (`1.`).
            if self.peek().is_some_and(|b| b.is_ascii_digit()) {
                self.read_decimal_digits()?;
            }
        }
        if matches!(self.peek(), Some(b'e' | b'E')) {
            is_float = true;
            self.read_exponent()?;
        }

        let _ = is_float; // both fold to TokenKind::Number for now
        self.reject_identifier_after_number(start)?;
        Ok(TokenKind::Number)
    }

    fn read_radix_number(&mut self, radix: u32, start: usize) -> Result<TokenKind> {
        self.advance(); // `0`
        self.advance(); // radix marker
        let mut any = false;
        let mut last_was_sep = false;
        while let Some(c) = self.peek() {
            if c == b'_' {
                if !any || last_was_sep {
                    return Err(self.sep_error(start));
                }
                last_was_sep = true;
                self.advance();
            } else if (c as char).is_digit(radix) {
                any = true;
                last_was_sep = false;
                self.advance();
            } else {
                break;
            }
        }
        if !any || last_was_sep {
            return Err(Error::syntax(
                "missing digits in numeric literal",
                Span::new(start as u32, self.pos as u32),
            ));
        }
        if self.peek() == Some(b'n') {
            self.advance();
            return Ok(TokenKind::BigInt);
        }
        self.reject_identifier_after_number(start)?;
        Ok(TokenKind::Number)
    }

    fn read_decimal_digits(&mut self) -> Result<()> {
        let start = self.pos;
        let mut last_was_sep = false;
        let mut any = false;
        while let Some(c) = self.peek() {
            if c == b'_' {
                if !any || last_was_sep {
                    return Err(self.sep_error(start));
                }
                last_was_sep = true;
                self.advance();
            } else if c.is_ascii_digit() {
                any = true;
                last_was_sep = false;
                self.advance();
            } else {
                break;
            }
        }
        if last_was_sep {
            return Err(self.sep_error(start));
        }
        Ok(())
    }

    fn read_exponent(&mut self) -> Result<()> {
        if !matches!(self.peek(), Some(b'e' | b'E')) {
            return Ok(());
        }
        let start = self.pos;
        self.advance(); // `e`
        if matches!(self.peek(), Some(b'+' | b'-')) {
            self.advance();
        }
        if !self.peek().is_some_and(|b| b.is_ascii_digit()) {
            return Err(Error::syntax(
                "missing exponent in numeric literal",
                Span::new(start as u32, self.pos as u32),
            ));
        }
        self.read_decimal_digits()
    }

    /// Per spec, an identifier may not immediately follow a numeric literal
    /// (`3in` is an error, not `3 in`).
    fn reject_identifier_after_number(&mut self, start: usize) -> Result<()> {
        if let Some(c) = self.peek()
            && (is_identifier_start_byte(c)
                || (c >= 0x80 && self.peek_char().is_some_and(is_identifier_start_char)))
        {
            return Err(Error::syntax(
                "identifier directly after numeric literal",
                Span::new(start as u32, self.pos as u32),
            ));
        }
        Ok(())
    }

    fn sep_error(&self, start: usize) -> Error {
        Error::syntax(
            "misplaced numeric separator `_`",
            Span::new(start as u32, self.pos as u32),
        )
    }

    // --- identifiers & keywords -----------------------------------------

    fn read_identifier_or_keyword(&mut self) -> Result<TokenKind> {
        let start = self.pos;
        self.read_identifier_tail();
        let text = &self.source[start..self.pos];
        Ok(match Keyword::from_str(text) {
            Some(kw) => TokenKind::Keyword(kw),
            None => TokenKind::Identifier,
        })
    }

    /// Consumes identifier-continue characters from the current position.
    fn read_identifier_tail(&mut self) {
        // The start char may already be consumed by the caller, or not; this
        // routine just eats all identifier-part chars from here.
        while let Some(c) = self.peek() {
            if c < 0x80 {
                if is_identifier_part_byte(c) {
                    self.advance();
                } else {
                    break;
                }
            } else {
                let ch = self.peek_char().expect("non-empty");
                if is_identifier_part_char(ch) {
                    self.advance_char(ch);
                } else {
                    break;
                }
            }
        }
    }

    // --- regex-vs-division heuristic ------------------------------------

    /// Whether a `/` at the current position should begin a regex literal,
    /// based on the previous significant token. This is the standard heuristic
    /// used by hand-written JS lexers.
    fn regex_allowed(&self) -> bool {
        match self.prev_significant {
            // Start of input → expression position.
            None => true,
            Some(kind) => match kind {
                // After a value-producing token, `/` is division.
                TokenKind::Identifier
                | TokenKind::PrivateName
                | TokenKind::Number
                | TokenKind::BigInt
                | TokenKind::String
                | TokenKind::Regex
                | TokenKind::NoSubstitutionTemplate
                | TokenKind::TemplateTail
                | TokenKind::RParen
                | TokenKind::RBracket
                | TokenKind::RBrace
                | TokenKind::PlusPlus
                | TokenKind::MinusMinus => false,
                // Keywords that produce/precede a value vs. those that precede
                // an expression.
                TokenKind::Keyword(kw) => kw.before_expression(),
                // Everything else (operators, `(`, `,`, `=`, `return`, …) is an
                // expression position.
                _ => true,
            },
        }
    }

    // --- low-level cursor ------------------------------------------------

    #[inline]
    fn peek(&self) -> Option<u8> {
        self.bytes.get(self.pos).copied()
    }

    #[inline]
    fn peek_at(&self, n: usize) -> Option<u8> {
        self.bytes.get(self.pos + n).copied()
    }

    /// Decodes the full Unicode scalar at `pos` (for the non-ASCII paths).
    #[inline]
    fn peek_char(&self) -> Option<char> {
        self.source[self.pos..].chars().next()
    }

    /// Advances one ASCII byte. Must not be called on a multi-byte lead byte.
    #[inline]
    fn advance(&mut self) {
        debug_assert!(self.bytes.get(self.pos).is_some_and(|b| *b < 0x80));
        self.pos += 1;
    }

    /// Advances over whatever is at `pos`, whether ASCII or a multi-byte char.
    #[inline]
    fn advance_any(&mut self) {
        match self.peek() {
            Some(c) if c < 0x80 => self.pos += 1,
            Some(_) => {
                let ch = self.peek_char().expect("non-empty");
                self.pos += ch.len_utf8();
            }
            None => {}
        }
    }

    /// Advances over a known decoded char.
    #[inline]
    fn advance_char(&mut self, ch: char) {
        self.pos += ch.len_utf8();
    }

    /// Consumes a single ASCII byte and returns `kind` — the common
    /// "two-character operator, second char matched" tail.
    #[inline]
    fn single(&mut self, kind: TokenKind) -> TokenKind {
        self.advance();
        kind
    }

    /// Finalizes a token spanning `[start, self.pos)`, updating the
    /// previous-significant-token state for the regex heuristic.
    fn make(&mut self, kind: TokenKind, start: usize, newline_before: bool) -> Token {
        if kind != TokenKind::Eof {
            self.prev_significant = Some(kind);
        }
        Token {
            kind,
            span: Span::new(start as u32, self.pos as u32),
            newline_before,
        }
    }
}

// --- identifier classification ------------------------------------------

/// ASCII identifier-start bytes (`$`, `_`, `A–Z`, `a–z`). Non-ASCII is handled
/// separately via [`is_identifier_part_char`].
#[inline]
fn is_identifier_start_byte(c: u8) -> bool {
    c == b'$' || c == b'_' || c.is_ascii_alphabetic()
}

/// ASCII identifier-continue bytes (start set plus digits).
#[inline]
fn is_identifier_part_byte(c: u8) -> bool {
    is_identifier_start_byte(c) || c.is_ascii_digit()
}

/// Whether a non-ASCII char may *start* an identifier (`ID_Start`: letters and
/// letter-numbers). With the `intl` feature this uses the Unicode property
/// tables; otherwise it falls back to a pragmatic `is_alphabetic` approximation.
#[inline]
pub(crate) fn is_identifier_start_char(ch: char) -> bool {
    if ch.is_ascii() {
        return is_identifier_start_byte(ch as u8);
    }
    #[cfg(feature = "intl")]
    {
        use intl::unicode::category::GeneralCategory as Gc;
        let gc = intl::unicode::general_category(ch);
        gc.is_letter() || gc == Gc::LetterNumber
    }
    #[cfg(not(feature = "intl"))]
    {
        ch.is_alphabetic()
    }
}

/// Whether a non-ASCII char may continue an identifier (`ID_Continue`:
/// `ID_Start` plus marks, decimal digits, connector punctuation, and ZWNJ/ZWJ).
/// With the `intl` feature this uses the Unicode property tables; otherwise a
/// pragmatic `is_alphanumeric` approximation. ASCII is routed through the byte
/// classifiers.
#[inline]
fn is_identifier_part_char(ch: char) -> bool {
    if ch.is_ascii() {
        return is_identifier_part_byte(ch as u8);
    }
    if ch == '\u{200C}' || ch == '\u{200D}' {
        return true; // ZWNJ / ZWJ
    }
    #[cfg(feature = "intl")]
    {
        use intl::unicode::category::GeneralCategory as Gc;
        let gc = intl::unicode::general_category(ch);
        gc.is_letter()
            || gc.is_mark()
            || matches!(
                gc,
                Gc::LetterNumber | Gc::DecimalNumber | Gc::ConnectorPunctuation
            )
    }
    #[cfg(not(feature = "intl"))]
    {
        ch.is_alphanumeric()
    }
}

/// Whether a char is ECMAScript whitespace (the `WhiteSpace` production):
/// TAB, VT, FF, SP, NBSP, ZWNBSP/BOM, and the Unicode `Zs` category.
#[inline]
fn is_unicode_whitespace(ch: char) -> bool {
    matches!(ch, '\u{00A0}' | '\u{FEFF}') || ch.is_whitespace() && !is_unicode_line_terminator(ch)
}

/// Whether a char is an ECMAScript `LineTerminator`: LF, CR, LS, PS.
#[inline]
fn is_unicode_line_terminator(ch: char) -> bool {
    matches!(ch, '\n' | '\r' | '\u{2028}' | '\u{2029}')
}