BPOS-MVP/apps/api/src/app/file-doc.ts

/**
 * 造好的资料文件**读回来给前端渲染** —— 进字节,出 `FileDoc`。
 *
 * 为什么绕这一圈,而不是直接把 Markdown 原文丢给前端按格式摆:
 * **那样界面渲的就不是那份文件了**。人下载到的是 .docx,屏上看到的是另一条路
 * 拼出来的样子,两边哪天不一致谁也发现不了。这里读的就是 `composeFile` 出的那一份
 * 字节:docx 解出段落、xlsx 解出行列 —— 渲得出来,才证明那份文件是真的。
 *
 * 纯计算:进字节、出结构,不碰库不碰网。
 */
import type { FileDoc } from '@contracts';
import { markdownHtmlDoc, type FileFormat } from './file-compose.ts';
import { readZip, sharedStrings, sheetToRows, unescapeXml } from './upload-extract.ts';

/** docx 的一段:`<w:p>` 里所有 `<w:t>` 连起来就是这一段的文字 */
function docxParagraphs(xml: string): Array<{ style: string; text: string }> {
  const out: Array<{ style: string; text: string }> = [];
  for (const p of xml.matchAll(/<w:p\b[^>]*>([\s\S]*?)<\/w:p>/g)) {
    const inner = p[1] ?? '';
    const style = /<w:pStyle\s+w:val="([^"]+)"/.exec(inner)?.[1] ?? '';
    const text = [...inner.matchAll(/<w:t\b[^>]*>([\s\S]*?)<\/w:t>/g)].map((t) => unescapeXml(t[1] ?? '')).join('');
    out.push({ style, text });
  }
  return out;
}

/** 查看面认得的块 —— 与 `FileDoc` 的 word 分支同形 */
export type WordBlock = { t: 'h1' | 'h2' | 'p'; text: string } | { t: 'ul'; items: string[] };

/**
 * docx → 查看面的块。表格在 `<w:tbl>` 里,而查看面(`FileViewer` 的 word 分支)
 * 只认 h1/h2/p/ul —— 表格按行摊成 `p`(用「 | 」连列),**宁可朴素也不丢内容**。
 */
export function docxBlocks(xml: string): WordBlock[] {
  const blocks: WordBlock[] = [];
  let bullets: string[] = [];
  const flush = (): void => {
    if (bullets.length === 0) return;
    blocks.push({ t: 'ul', items: bullets });
    bullets = [];
  };

  // 表格与段落按它们在文档里的先后走:一次扫过,谁先出现算谁
  for (const node of xml.matchAll(/<w:tbl\b[\s\S]*?<\/w:tbl>|<w:p\b[^>]*>[\s\S]*?<\/w:p>/g)) {
    const chunk = node[0];
    if (chunk.startsWith('<w:tbl')) {
      flush();
      for (const tr of chunk.matchAll(/<w:tr\b[^>]*>([\s\S]*?)<\/w:tr>/g)) {
        const cells = [...(tr[1] ?? '').matchAll(/<w:tc\b[^>]*>([\s\S]*?)<\/w:tc>/g)].map((tc) =>
          [...(tc[1] ?? '').matchAll(/<w:t\b[^>]*>([\s\S]*?)<\/w:t>/g)].map((t) => unescapeXml(t[1] ?? '')).join(''),
        );
        if (cells.some((c) => c !== '')) blocks.push({ t: 'p', text: cells.join(' | ') });
      }
      continue;
    }
    const [para] = docxParagraphs(chunk);
    if (para === undefined) continue;
    const { style, text } = para;
    if (text === '') continue;
    if (style === 'ListParagraph') {
      bullets.push(text.replace(/^[•\d]+[.、]?\s*/, ''));
      continue;
    }
    flush();
    if (style === 'Title' || style === 'Heading1') blocks.push({ t: 'h1', text });
    else if (style === 'Heading2') blocks.push({ t: 'h2', text });
    else blocks.push({ t: 'p', text });
  }
  flush();
  return blocks;
}

/** xlsx → 行列(第一张表)。读的是**真包**:解不开就如实说,不编一张表出来 */
export function xlsxRows(bytes: Uint8Array): { rows: string[][]; sheet: string } | undefined {
  const zip = readZip(bytes);
  if (zip.size === 0) return undefined;
  const dec = new TextDecoder('utf-8');
  const sharedRaw = zip.get('xl/sharedStrings.xml');
  const shared = sharedRaw === undefined ? [] : sharedStrings(dec.decode(sharedRaw));
  const first = [...zip.keys()].filter((k) => /^xl\/worksheets\/sheet\d+\.xml$/.test(k)).sort()[0];
  if (first === undefined) return undefined;
  const raw = zip.get(first);
  if (raw === undefined) return undefined;
  const wbRaw = zip.get('xl/workbook.xml');
  const sheet =
    wbRaw === undefined ? 'Sheet1' : unescapeXml(/<sheet\s+name="([^"]*)"/.exec(dec.decode(wbRaw))?.[1] ?? 'Sheet1');
  return { rows: sheetToRows(dec.decode(raw), shared), sheet };
}

/**
 * 字节 → `FileDoc`。**读不出来就返回 undefined** —— 上层据此如实说
 * 「这份文件打不开」,不摆一个空壳假装渲染过(界面不撒谎)。
 */
export function toFileDoc(input: {
  readonly name: string;
  readonly title: string;
  readonly format: FileFormat;
  readonly bytes: Uint8Array;
}): FileDoc | undefined {
  const { name, title, format, bytes } = input;
  const dec = new TextDecoder('utf-8');

  if (format === 'html') return { kind: 'html', name, html: dec.decode(bytes) };
  if (format === 'md') {
    const md = dec.decode(bytes);
    // 原文与渲好的一起交:查看面摆渲染版,而「下载」拿到的仍是这段原文
    return { kind: 'markdown', name, md, html: markdownHtmlDoc(title, md) };
  }
  if (format === 'xlsx') {
    const sheet = xlsxRows(bytes);
    if (sheet === undefined) return undefined;
    return { kind: 'excel', name, sheet: sheet.sheet, rows: sheet.rows };
  }

  const zip = readZip(bytes);
  const doc = zip.get('word/document.xml');
  if (doc === undefined) return undefined;
  const blocks = docxBlocks(dec.decode(doc));
  if (blocks.length === 0) return undefined;
  return { kind: 'word', name, blocks };
}