tools/eu4_clausewitz/src/lexer.mjs

// Byte-exact, position-tracking tokenizer for Paradox Clausewitz syntax
// (EU4 history/common/events/decisions/missions files).
//
// The lexer NEVER regenerates text — it only locates things. Every token
// keeps the exact source slice it came from (`raw`), and concatenating every
// token's `raw` in order reproduces the input byte-for-byte. This is the
// load-bearing property tested by test/roundtrip.test.mjs: a past tool that
// decoded+re-encoded a file silently reformatted thousands of lines (indent
// style, comment position, blank lines) while leaving values unchanged, and
// that must be structurally impossible here.
//
// Grammar notes (measured against real vanilla EU4 data — see the task
// writeup this package was built from):
//  - Only `=` occurs as an operator. `>=`, `<=`, `!=`, `><`, `hsv{...}`, and
//    `@variable` are confirmed absent from EU4 and are deliberately not
//    special-cased; any of those characters appearing in source text just
//    fall into an ordinary bare token or standalone character, which still
//    round-trips correctly even though the lexer attaches no special
//    meaning to it.
//  - No `\`-escaped quotes were found in any real string token (backslashes
//    that appear near quotes in vanilla data are inside `#comment` text,
//    never inside an actual `"..."` value), so strings are lexed as
//    `"` ... next unescaped `"` with no escape processing.
//  - A `#` inside a quoted string is not a comment start, and a `"` inside a
//    comment is not a string start. Handling this requires one real state
//    machine (in-string / in-comment / code) rather than independently
//    regexing for quotes and comments — naive independent scans
//    desynchronize permanently after the first line where the two collide
//    (see common/countries/*.txt: `monarch_names = { # ... "nori"` and
//    Silesia.txt's `"#LIST#|leader_names|Hoppe"`, which is a string that
//    itself starts with `#`).

/** @typedef {'comment'|'string'|'bare'|'lbrace'|'rbrace'|'operator'|'whitespace'} TokenKind */

/**
 * @typedef {Object} Token
 * @property {TokenKind} kind
 * @property {string} value  - semantic content (comment text sans '#', string sans quotes, etc.)
 * @property {string} raw    - exact source slice, quotes/braces/# included
 * @property {number} index  - 0-based character offset into the input text
 * @property {number} line   - 1-based line number of the token's first character
 * @property {number} col    - 1-based column of the token's first character
 * @property {number} length - raw.length
 */

const WHITESPACE = new Set([' ', '\t', '\r', '\n']);
const SINGLE_CHAR_KIND = new Map([
  ['{', 'lbrace'],
  ['}', 'rbrace'],
  ['=', 'operator'],
]);
// Characters that end a bare token: whitespace, structural punctuation, and
// the two characters that switch lexer state (# and ").
const BARE_STOP = new Set([' ', '\t', '\r', '\n', '{', '}', '=', '"', '#']);

/**
 * Tokenize Clausewitz source text into a position-tracked token stream.
 *
 * @param {string} text
 * @returns {Token[]}
 */
export function tokenize(text) {
  const tokens = [];
  const length = text.length;

  let index = 0;
  let line = 1;
  let col = 1;

  // Advance the line/col cursor over `slice` (which may contain newlines),
  // shared by every branch below so position bookkeeping lives in one place.
  function advance(slice) {
    for (let i = 0; i < slice.length; i += 1) {
      if (slice[i] === '\n') {
        line += 1;
        col = 1;
      } else {
        col += 1;
      }
    }
  }

  function push(kind, value, raw) {
    tokens.push({ kind, value, raw, index, line, col, length: raw.length });
    index += raw.length;
    advance(raw);
  }

  while (index < length) {
    const char = text[index];

    if (WHITESPACE.has(char)) {
      let end = index + 1;
      while (end < length && WHITESPACE.has(text[end])) end += 1;
      const raw = text.slice(index, end);
      push('whitespace', raw, raw);
      continue;
    }

    if (char === '#') {
      // Unquoted '#' to end of line (exclusive of the line terminator, so
      // the terminator itself is tokenized separately as whitespace).
      let end = index + 1;
      while (end < length && text[end] !== '\n' && text[end] !== '\r') end += 1;
      const raw = text.slice(index, end);
      push('comment', raw.slice(1), raw);
      continue;
    }

    if (char === '"') {
      // Quoted string: '"' ... next '"' (or EOF, for a truncated/malformed
      // file — we still must not throw, since the round-trip test asserts
      // over every real vanilla+mod file and a single malformed one must
      // not crash the whole run; it simply won't find a closing quote).
      let end = index + 1;
      while (end < length && text[end] !== '"') end += 1;
      const closed = end < length;
      const stringEnd = closed ? end + 1 : end;
      const raw = text.slice(index, stringEnd);
      const value = closed ? text.slice(index + 1, end) : text.slice(index + 1, end);
      push('string', value, raw);
      continue;
    }

    const single = SINGLE_CHAR_KIND.get(char);
    if (single) {
      push(single, char, char);
      continue;
    }

    // Bare token: run until the next character that starts a new kind.
    let end = index + 1;
    while (end < length && !BARE_STOP.has(text[end])) end += 1;
    const raw = text.slice(index, end);
    push('bare', raw, raw);
  }

  return tokens;
}