dorfteich/packages/shared/src/editor-schema/markdown.ts

import MarkdownIt from 'markdown-it';
import Token from 'markdown-it/lib/token.mjs';
import { Mark, Node } from 'prosemirror-model';
import { MarkdownParser, MarkdownSerializer, MarkdownSerializerState } from 'prosemirror-markdown';

import { editorSchema } from './schema';

/**
 * markdown-it's default preset already tokenizes GFM tables and
 * strikethrough; task lists are GFM-only and not tokenized by markdown-it
 * itself, so {@link transformTokens} rewrites the raw token stream before
 * `prosemirror-markdown` turns it into a document (ADR 0004).
 */
function findMatchingClose(tokens: Token[], openIndex: number): number {
  let depth = 1;
  for (let i = openIndex + 1; i < tokens.length; i += 1) {
    const type = tokens[i]?.type ?? '';
    if (type.endsWith('_open')) depth += 1;
    else if (type.endsWith('_close')) depth -= 1;
    if (depth === 0) return i;
  }
  throw new Error(`Unbalanced markdown-it token stream at index ${openIndex}`);
}

/** Indices of `type` tokens directly inside `[start, end)`, not in a nested container. */
function directChildOpens(tokens: Token[], start: number, end: number, type: string): number[] {
  const result: number[] = [];
  let depth = 0;
  for (let i = start; i < end; i += 1) {
    const tok = tokens[i];
    if (!tok) continue;
    if (depth === 0 && tok.type === type) result.push(i);
    if (tok.type.endsWith('_open')) depth += 1;
    else if (tok.type.endsWith('_close')) depth -= 1;
  }
  return result;
}

const TASK_MARKER = /^\[([ xX])\]\s+/;

/**
 * Reads (and strips) the `[ ] `/`[x] ` prefix from a list item's first
 * paragraph. Returns `null` when the item has no checkbox marker, which
 * also signals "this list is not a task list" to the caller.
 */
function taskCheckedFor(tokens: Token[], itemOpen: number): boolean | null {
  const paragraphOpen = tokens[itemOpen + 1];
  if (paragraphOpen?.type !== 'paragraph_open') return null;
  const inline = tokens[itemOpen + 2];
  if (inline?.type !== 'inline') return null;
  const firstChild = inline.children?.[0];
  if (!firstChild || firstChild.type !== 'text') return null;
  const match = TASK_MARKER.exec(firstChild.content);
  if (!match) return null;
  firstChild.content = firstChild.content.slice(match[0].length);
  return (match[1] ?? '').toLowerCase() === 'x';
}

function retype(token: Token, type: string): Token {
  const clone = new Token(type, token.tag, token.nesting);
  Object.assign(clone, token, { type });
  return clone;
}

/**
 * Rewrites the markdown-it token stream so it matches the editor schema:
 * - a bullet list where every item carries a `[ ]`/`[x]` marker becomes a
 *   `task_list` of `task_item`s (mixed lists are left as plain bullet
 *   lists — GFM does not define their meaning either);
 * - table cell content (a bare `inline` token in markdown-it) is wrapped
 *   in a synthetic paragraph, matching `table_cell`/`table_header`'s
 *   `block+` content.
 */
function transformTokens(tokens: Token[]): Token[] {
  const out: Token[] = [];
  let i = 0;
  while (i < tokens.length) {
    const tok = tokens[i];
    if (!tok) {
      i += 1;
      continue;
    }

    if (tok.type === 'bullet_list_open') {
      const close = findMatchingClose(tokens, i);
      const itemOpens = directChildOpens(tokens, i + 1, close, 'list_item_open');
      const checks = itemOpens.map((idx) => taskCheckedFor(tokens, idx));
      const isTaskList = checks.length > 0 && checks.every((c) => c !== null);
      if (isTaskList) {
        out.push(retype(tok, 'task_list_open'));
        let itemPos = 0;
        for (let j = i + 1; j < close; j += 1) {
          const t = tokens[j];
          if (!t) continue;
          if (t.type === 'list_item_open') {
            const retyped = retype(t, 'task_item_open');
            retyped.attrSet('checked', String(checks[itemPos] === true));
            itemPos += 1;
            out.push(retyped);
          } else if (t.type === 'list_item_close') {
            out.push(retype(t, 'task_item_close'));
          } else {
            out.push(t);
          }
        }
        const closeTok = tokens[close];
        if (closeTok) out.push(retype(closeTok, 'task_list_close'));
        i = close + 1;
        continue;
      }
    }

    if (tok.type === 'th_open' || tok.type === 'td_open') {
      const close = findMatchingClose(tokens, i);
      out.push(tok);
      out.push(Object.assign(new Token('paragraph_open', 'p', 1), { hidden: true }));
      for (let j = i + 1; j < close; j += 1) {
        const t = tokens[j];
        if (t) out.push(t);
      }
      out.push(Object.assign(new Token('paragraph_close', 'p', -1), { hidden: true }));
      const closeTok = tokens[close];
      if (closeTok) out.push(closeTok);
      i = close + 1;
      continue;
    }

    out.push(tok);
    i += 1;
  }
  return out;
}

function createTokenizer(): MarkdownIt {
  const md = new MarkdownIt('default', { html: false });
  const rawParse = md.parse.bind(md);
  md.parse = (src, env) => transformTokens(rawParse(src, env));
  return md;
}

const markdownParser = new MarkdownParser(editorSchema, createTokenizer(), {
  blockquote: { block: 'blockquote' },
  paragraph: { block: 'paragraph' },
  list_item: { block: 'list_item' },
  task_item: {
    block: 'task_item',
    getAttrs: (tok) => ({ checked: tok.attrGet('checked') === 'true' }),
  },
  bullet_list: { block: 'bullet_list' },
  task_list: { block: 'task_list' },
  ordered_list: {
    block: 'ordered_list',
    getAttrs: (tok) => ({ order: Number(tok.attrGet('start')) || 1 }),
  },
  heading: {
    block: 'heading',
    // The schema only defines levels 1-4; deeper markdown headings clamp
    // down rather than fail the parse.
    getAttrs: (tok) => ({ level: Math.min(4, Number(tok.tag.slice(1)) || 1) }),
  },
  code_block: { block: 'code_block', noCloseToken: true },
  fence: { block: 'code_block', noCloseToken: true },
  hr: { node: 'horizontal_rule' },
  image: {
    node: 'image',
    // The "src" slot carries the opaque fileId; the API's markdown import/
    // export endpoint (#30) is responsible for resolving it to a servable
    // URL and back — that boundary is not this package's concern.
    getAttrs: (tok) => ({
      fileId: tok.attrGet('src') ?? '',
      alt: tok.children?.[0]?.content ?? '',
      width: null,
    }),
  },
  hardbreak: { node: 'hard_break' },
  em: { mark: 'italic' },
  strong: { mark: 'bold' },
  s: { mark: 'strikethrough' },
  link: { mark: 'link', getAttrs: (tok) => ({ href: tok.attrGet('href') ?? '' }) },
  code_inline: { mark: 'code', noCloseToken: true },
  table: { block: 'table' },
  thead: { ignore: true },
  tbody: { ignore: true },
  tr: { block: 'table_row' },
  th: { block: 'table_header' },
  td: { block: 'table_cell' },
});

/** Parses Markdown into a document of {@link editorSchema}. */
export function markdownToDoc(markdown: string): Node {
  return markdownParser.parse(markdown);
}

function renderTable(state: MarkdownSerializerState, node: Node): void {
  const rows: string[][] = [];
  node.forEach((row) => {
    const cells: string[] = [];
    row.forEach((cell) => {
      cells.push(cell.textContent.replace(/\|/g, '\\|').replace(/\r?\n/g, ' ').trim());
    });
    rows.push(cells);
  });
  const headerRow = rows[0];
  if (!headerRow) {
    state.closeBlock(node);
    return;
  }
  const lines = [
    `| ${headerRow.join(' | ')} |`,
    `| ${headerRow.map(() => '---').join(' | ')} |`,
    ...rows.slice(1).map((row) => `| ${row.join(' | ')} |`),
  ];
  state.write(lines.join('\n'));
  state.closeBlock(node);
}

/** Serializes a document of {@link editorSchema} back to Markdown. */
const markdownSerializer = new MarkdownSerializer(
  {
    blockquote(state, node) {
      state.wrapBlock('> ', null, node, () => state.renderContent(node));
    },
    code_block(state, node) {
      const backticks = node.textContent.match(/`{3,}/gm);
      const fence = backticks ? `${[...backticks].sort().slice(-1)[0]}\`` : '```';
      state.write(`${fence}\n`);
      state.text(node.textContent, false);
      state.write('\n');
      state.write(fence);
      state.closeBlock(node);
    },
    heading(state, node) {
      state.write(`${state.repeat('#', node.attrs.level as number)} `);
      state.renderInline(node, false);
      state.closeBlock(node);
    },
    horizontal_rule(state, node) {
      state.write('---');
      state.closeBlock(node);
    },
    bullet_list(state, node) {
      state.renderList(node, '  ', () => '- ');
    },
    task_list(state, node) {
      state.renderList(node, '  ', (i) => `- [${node.child(i).attrs.checked ? 'x' : ' '}] `);
    },
    ordered_list(state, node) {
      const start = (node.attrs.order as number) || 1;
      const maxWidth = String(start + node.childCount - 1).length;
      const space = state.repeat(' ', maxWidth + 2);
      state.renderList(node, space, (i) => {
        const numeral = String(start + i);
        return state.repeat(' ', maxWidth - numeral.length) + numeral + '. ';
      });
    },
    list_item(state, node) {
      state.renderContent(node);
    },
    task_item(state, node) {
      state.renderContent(node);
    },
    paragraph(state, node) {
      state.renderInline(node);
      state.closeBlock(node);
    },
    image(state, node) {
      const alt = state.esc((node.attrs.alt as string) || '');
      const fileId = (node.attrs.fileId as string).replace(/[()]/g, '\\$&');
      state.write(`![${alt}](${fileId})`);
    },
    hard_break(state, node, parent, index) {
      for (let i = index + 1; i < parent.childCount; i += 1) {
        if (parent.child(i).type !== node.type) {
          state.write('\\\n');
          return;
        }
      }
    },
    text(state, node) {
      state.text(node.text ?? '');
    },
    table: renderTable,
  },
  {
    italic: { open: '*', close: '*', mixable: true, expelEnclosingWhitespace: true },
    bold: { open: '**', close: '**', mixable: true, expelEnclosingWhitespace: true },
    strikethrough: { open: '~~', close: '~~', mixable: true, expelEnclosingWhitespace: true },
    link: {
      open: '[',
      close: (_state, mark: Mark) => `](${(mark.attrs.href as string).replace(/[()]/g, '\\$&')})`,
      mixable: true,
    },
    code: {
      open: (_state, _mark, parent, index) => backticksFor(parent.child(index), -1),
      close: (_state, _mark, parent, index) => backticksFor(parent.child(index - 1), 1),
      escape: false,
    },
  },
);

function backticksFor(node: Node, side: -1 | 1): string {
  const text = node.isText ? (node.text ?? '') : '';
  const matches = text.match(/`+/g);
  const len = matches ? Math.max(...matches.map((m) => m.length)) : 0;
  let result = len > 0 && side > 0 ? ' `' : '`';
  result += '`'.repeat(len);
  if (len > 0 && side < 0) result += ' ';
  return result;
}

export function docToMarkdown(doc: Node): string {
  // The schema has no `tight`/`loose` list attribute (out of scope for v1)
  // — every list renders without blank lines between items.
  return markdownSerializer.serialize(doc, { tightLists: true });
}