collabSupport/utils/markdownToAdf.js

// src/utils/markdownToAdf.js
//
// Narrow, hand-rolled markdown → Atlassian Document Format converter.
//
// Grammar supported (deliberately minimal, matches what our chat
// renderers actually emit — see commands/voiceStatus.js and
// commands/avStatus.js):
//
//   **text**              → { type:'text', text, marks:[{type:'strong'}] }
//   *text*                → { type:'text', text, marks:[{type:'em'}] }
//   [label](url)          → { type:'text', text:label, marks:[{type:'link',attrs:{href:url}}] }
//   \n\n (blank line)     → paragraph split
//   \n (single line)      → { type:'hardBreak' } inside the paragraph
//   anything else         → plain { type:'text', text } — emojis, arrows,
//                            bullets, unicode symbols all pass through.
//
// Explicitly NOT supported: headings, bulletLists, fenced code, tables,
// nested marks, block quotes, images. Our renderers don't produce any
// of that; if they start to, we'll extend the grammar then rather than
// pull in a full CommonMark library for a narrow use case.
//
// The parser is deliberately forgiving with malformed input. An unclosed
// `**` becomes plain text — never an exception. That's the right call
// for a converter that runs unattended on live ticket data.

const LINK_RE   = /\[([^\]\n]+?)\]\(([^)\n]+?)\)/;
const STRONG_RE = /\*\*([^*\n][^*\n]*?)\*\*/;
const EM_RE     = /\*([^*\n]+?)\*/;

// Attempt each pattern at the beginning of the remaining slice and
// return the earliest match. Ties resolve by pattern order (link,
// strong, em) — which happens to be the safe direction (link brackets
// can't be confused with `*`, and `**` must be tried before `*`).
function findEarliestMatch(text) {
  const candidates = [
    { re: LINK_RE,   kind: 'link' },
    { re: STRONG_RE, kind: 'strong' },
    { re: EM_RE,     kind: 'em' },
  ];
  let best = null;
  for (const c of candidates) {
    const m = c.re.exec(text);
    if (m && (best === null || m.index < best.match.index)) {
      best = { kind: c.kind, match: m };
      // Can't early-exit — a link at index 5 beats a strong at index 0
      // is impossible (best keeps track), but a link at index 5 CAN
      // beat a strong at index 10.
    }
  }
  return best;
}

/**
 * Tokenize a single line (no `\n` in the input) into a sequence of
 * ADF inline nodes. Returns [] for empty input.
 */
export function tokenizeLine(line) {
  if (!line) return [];
  const nodes = [];
  let cursor = 0;
  while (cursor < line.length) {
    const rest = line.slice(cursor);
    const found = findEarliestMatch(rest);
    if (!found) {
      // No more markup — the rest is plain text.
      pushText(nodes, line.slice(cursor));
      break;
    }
    // Emit whatever plain text sits before the match.
    if (found.match.index > 0) {
      pushText(nodes, rest.slice(0, found.match.index));
    }
    // Emit the marked node.
    if (found.kind === 'link') {
      const [, label, url] = found.match;
      nodes.push({
        type: 'text',
        text: label,
        marks: [{ type: 'link', attrs: { href: url } }],
      });
    } else {
      // strong or em — both extract group 1 as inner text.
      const inner = found.match[1];
      nodes.push({
        type: 'text',
        text: inner,
        marks: [{ type: found.kind }],
      });
    }
    cursor += found.match.index + found.match[0].length;
  }
  return nodes;
}

// Append a text node while collapsing adjacent runs of plain text.
// ADF permits multiple sibling text nodes, but a single collapsed node
// keeps the output tidy for humans skimming a rendered ADF payload.
function pushText(nodes, text) {
  if (!text) return;
  const last = nodes[nodes.length - 1];
  if (last && last.type === 'text' && !last.marks) {
    last.text += text;
  } else {
    nodes.push({ type: 'text', text });
  }
}

/**
 * Convert a markdown string to an array of ADF block nodes (paragraphs
 * containing inline nodes and hardBreaks). Suitable for splicing into
 * a larger ADF document's `content` array — e.g. below a fixed header
 * paragraph in the Jira poller's `buildAdfComment`.
 *
 * Blank lines separate paragraphs. Non-blank lines within a paragraph
 * are joined with `hardBreak` nodes so line-oriented output like the
 * chat renderers survives visually intact.
 *
 * @param {string} markdown
 * @returns {Array<object>} ADF content nodes (each a `paragraph`).
 */
export function markdownToAdfContent(markdown) {
  if (typeof markdown !== 'string' || markdown.trim() === '') return [];

  // Split into paragraph groups on blank-line boundaries. Preserve
  // relative order — leading/trailing blank lines just yield empty
  // groups that we drop.
  const groups = markdown
    .split(/\n\s*\n/)
    .map((g) => g.replace(/\n+$/, ''))
    .filter((g) => g.length > 0);

  const paragraphs = [];
  for (const group of groups) {
    const lines = group.split('\n');
    const content = [];
    lines.forEach((line, i) => {
      const inline = tokenizeLine(line);
      if (inline.length > 0) content.push(...inline);
      if (i < lines.length - 1) {
        // Soft line break inside a paragraph. ADF's hardBreak renders
        // as an in-paragraph line break in the Jira viewer.
        content.push({ type: 'hardBreak' });
      }
    });
    // Skip paragraphs that ended up empty (e.g. a group that was just
    // whitespace lines). An empty ADF paragraph is legal but noisy.
    if (content.length === 0) continue;
    paragraphs.push({ type: 'paragraph', content });
  }

  return paragraphs;
}

/**
 * Full-document convenience: wrap `markdownToAdfContent` in a valid
 * ADF `doc`. Callers that want to embed the paragraphs inside a
 * larger custom document (like the poller's header + body pattern)
 * should call `markdownToAdfContent` directly.
 *
 * @param {string} markdown
 * @returns {object} ADF document `{ version:1, type:'doc', content }`.
 */
export function markdownToAdf(markdown) {
  return {
    version: 1,
    type: 'doc',
    content: markdownToAdfContent(markdown),
  };
}