import {
  createBlock,
  getBlockType,
  parse,
  rawHandler,
} from '@wordpress/blocks';

import {
  FAQ_BLOCK,
  mergeNewFaqs,
  type BlockLike,
  type FaqPair,
} from './faq-merge';

/** The FAQ the agent writes, as it wraps it. */
const FAQ_SECTION =
  /<div\s[^>]*class="[^"]*\bsection-faq\b[^"]*"[^>]*>[\s\S]*?<\/div>/gi;

/** Block markup carries its own comments; pasted HTML does not. */
const BLOCK_MARKUP = /<!--\s+\/?wp:/;

/** A picture alone in a paragraph, linked or not, as the agent places one. */
const LONE_IMAGE =
  /<p\b[^>]*>\s*(<a\b[^>]*>\s*<img\b[^>]*>\s*<\/a>|<img\b[^>]*>)\s*<\/p>/gi;

/**
 * Pictures alone in a paragraph as figures: the raw handler made such a
 * paragraph an image block with an empty paragraph after it.
 *
 * @param {string} html - Pasted or agent-written HTML.
 * @returns {string} The same HTML, each lone picture in a figure.
 */
export const loneImagesAsFigures = (html: string): string =>
  html.replace(LONE_IMAGE, '<figure>$1</figure>');

/** The question headings of an agent-written FAQ and the answers under each. */
const faqPairs = (faq: HTMLElement): FaqPair[] => {
  const pairs: FaqPair[] = [];
  for (const child of Array.from(faq.children)) {
    if (/^H[3-4]$/.test(child.tagName)) {
      pairs.push({ question: child.textContent?.trim() ?? '', answer: '' });
    } else if (pairs.length > 0 && !/^H[1-6]$/.test(child.tagName)) {
      const last = pairs[pairs.length - 1];
      last.answer += (last.answer ? '<br>' : '') + child.innerHTML.trim();
    }
  }
  return pairs.filter(pair => pair.question !== '' && pair.answer !== '');
};

/**
 * Blocks for a chunk of content: block markup is parsed as itself, so the
 * post keeps the blocks it had, and anything else goes through the raw
 * handler the way pasted HTML does.
 */
const chunkBlocks = (html: string): unknown[] => {
  if (html.trim() === '') {
    return [];
  }
  return BLOCK_MARKUP.test(html)
    ? (parse(html) as unknown[])
    : (rawHandler({ HTML: loneImagesAsFigures(html) }) as unknown[]);
};

/**
 * Blocks for HTML the agent wrote or the editor already holds: a FAQ it
 * wrapped in div.section-faq becomes the plugin's FAQ block (which prints
 * FAQPage schema), its heading a heading before it, and the rest is parsed
 * as block markup or handled as raw HTML. Without the FAQ block registered
 * the div stays as it is.
 */
export const blocksFromHtml = (html: string): unknown[] => {
  const hasFaqBlock = Boolean(getBlockType(FAQ_BLOCK));
  const blocks: BlockLike[] = [];
  const made = new Set<BlockLike>();
  let at = 0;
  for (const match of html.matchAll(FAQ_SECTION)) {
    const start = match.index ?? 0;
    const container = document.createElement('div');
    container.innerHTML = match[0];
    const section = container.firstElementChild as HTMLElement | null;
    const pairs = section && hasFaqBlock ? faqPairs(section) : [];
    if (pairs.length === 0) {
      continue;
    }
    blocks.push(...(chunkBlocks(html.slice(at, start)) as BlockLike[]));
    // The FAQ block prints the questions, not a heading of its own, so the
    // one the agent wrote becomes a heading block before it.
    const heading = section?.querySelector('h2')?.textContent?.trim() ?? '';
    if (heading !== '') {
      const block = createBlock('core/heading', {
        content: heading,
        level: 2,
      }) as BlockLike;
      made.add(block);
      blocks.push(block);
    }
    const faq = createBlock(FAQ_BLOCK, { questions: pairs }) as BlockLike;
    made.add(faq);
    blocks.push(faq);
    at = start + match[0].length;
  }
  blocks.push(...(chunkBlocks(html.slice(at)) as BlockLike[]));
  return mergeNewFaqs(blocks, made);
};

/**
 * The content with the table of contents block after its first paragraph,
 * where readers look for it, or at the top when there is none.
 *
 * @param {string} content - Block markup.
 * @returns {string} The content with the block.
 */
export const withToc = (content: string): string => {
  const toc = '<!-- wp:recomaze/toc /-->';
  const end = '<!-- /wp:paragraph -->';
  const at = content.indexOf(end);
  return at < 0
    ? `${toc}\n\n${content}`
    : `${content.slice(0, at + end.length)}\n\n${toc}${content.slice(at + end.length)}`;
};