/* =====================================================================
   Inline markup — a deliberately small syntax, chosen over contenteditable
   because it round-trips exactly and cannot carry pasted styling.

     `code`            → <code>
     [[Ctrl]]          → <kbd>
     **strong**        → <strong>
     *emphasis*        → <em>
     [text](url)       → <a href>
     {{part:id}}       → "Part 5", resolved from the live numbering

   parse() and serialise() are inverses; the round-trip test relies on it.
   ===================================================================== */

const ESC = { '&': '&amp;', '<': '&lt;', '>': '&gt;' };
const escape = (s) => s.replace(/[&<>]/g, (c) => ESC[c]);

/** markup → HTML. `refs` maps a part id to its number. */
export function parse(text, refs = new Map()) {
  const slots = [];
  const stash = (html) => ` ${slots.push(html) - 1} `;

  let s = text;

  // code first: nothing inside a code span is markup
  s = s.replace(/`([^`]+)`/g, (_, body) => stash(`<code>${escape(body)}</code>`));
  s = s.replace(/\[\[([^\]]+)\]\]/g, (_, body) => stash(`<kbd>${escape(body)}</kbd>`));
  s = s.replace(/\{\{part:([A-Za-z0-9_-]+)\}\}/g, (_, id) => {
    const n = refs.get(id);
    return stash(n ? `Part ${n}` : '<span class="ref-missing">?</span>');
  });
  s = s.replace(/\[([^\]]+)\]\(([^)\s]+)\)/g, (_, label, href) =>
    stash(`<a href="${escape(href)}">${escape(label)}</a>`));

  s = escape(s);
  s = s.replace(/\*\*([^*]+)\*\*/g, '<strong>$1</strong>');
  s = s.replace(/(^|[^*])\*([^*]+)\*/g, '$1<em>$2</em>');
  s = s.replace(/\n/g, '<br>');          // inverse of the sentinel in serialise()

  return s.replace(/ (\d+) /g, (_, i) => slots[Number(i)]);
}

const UNESC = { '&amp;': '&', '&lt;': '<', '&gt;': '>', '&quot;': '"', '&#39;': "'" };
const unescape = (s) => s.replace(/&(amp|lt|gt|quot|#39);/g, (m) => UNESC[m]);

/** HTML → markup, for importing an existing sheet. */
export function serialise(html) {
  let s = html.replace(/&nbsp;|&#160;/g, '\u00A0');
  s = s.replace(/<code>(.*?)<\/code>/gs, (_, b) => '`' + unescape(b) + '`');
  s = s.replace(/<kbd>(.*?)<\/kbd>/gs, (_, b) => '[[' + unescape(b) + ']]');
  s = s.replace(/<a [^>]*href="([^"]*)"[^>]*>(.*?)<\/a>/gs, (_, h, b) => `[${unescape(b)}](${h})`);
  s = s.replace(/<strong>(.*?)<\/strong>/gs, '**$1**');
  s = s.replace(/<em>(.*?)<\/em>/gs, '*$1*');
  // a sentinel, so the whitespace collapse below cannot swallow the break
  s = s.replace(/<br\s*\/?>/g, '\u0000');
  s = s.replace(/<[^>]+>/g, '');               // anything left is decoration
  // collapse ordinary whitespace but never a non-breaking space, which is
  // meaningful typography the author put there on purpose
  return unescape(s)
    .replace(/[^\S\u00A0]+/g, ' ')
    .replace(/ ?\u0000 ?/g, '\n')
    .trim();
}

/** Part ids referenced by a document, so the UI can flag broken links. */
export function refsUsed(text) {
  return [...text.matchAll(/\{\{part:([A-Za-z0-9_-]+)\}\}/g)].map((m) => m[1]);
}