@xiaobuyu/nodesign 0.0.1

package/server/lib/docx/dump-styles.js

@@ -0,0 +1,287 @@
+/**
+ * dump-styles.js — 外来 docx 的 styles.xml/theme/sectPr → token JSON（尽力而为）。
+ *
+ * 覆盖规则（字段闭合原则的读端）：
+ *   - 一个 pPr/rPr 子元素要么**完整**映射进 token 字段，要么**整条**进
+ *     extraXml 透传（绝不半吊子拆一半，防止重建时双重应用）；
+ *   - 整条 w:style 装不进模型（表格样式等）→ rawStyles 原文透传；
+ *   - 主题字体引用（w:asciiTheme 等）就地解析成实际字体名。
+ *
+ * 另附 applyStyleEdit：对外来文档做「定点样式手术」——只改目标 style 的
+ * 目标属性，其他字节不动（编辑外来 docx 的推荐路线）。
+ */
+
+import { readFileSync } from 'node:fs';
+import { parseXml, serialize, serializeNode, elem } from './xml.js';
+import { sortChildren } from './order.js';
+import { readZip, entryData, replaceEntry, writeZip } from './rawzip.js';
+import { buildRPr } from './build.js';
+
+/* ── theme ── */
+function parseTheme(zip) {
+  const buf = entryData(zip, 'word/theme/theme1.xml');
+  if (!buf) return null;
+  const doc = parseXml(buf.toString());
+  const out = { major: {}, minor: {} };
+  for (const [key, tag] of [['major', 'a:majorFont'], ['minor', 'a:minorFont']]) {
+    const font = doc.root.find(tag)[0];
+    if (!font) continue;
+    out[key].latin = font.firstChild('a:latin')?.attr('typeface') || null;
+    out[key].ea = font.firstChild('a:ea')?.attr('typeface') || null;
+    out[key].cs = font.firstChild('a:cs')?.attr('typeface') || null;
+  }
+  return out;
+}
+
+function resolveThemeFont(theme, ref) {
+  if (!theme || !ref) return null;
+  const m = { majorHAnsi: theme.major.latin, majorAscii: theme.major.latin, majorEastAsia: theme.major.ea, majorBidi: theme.major.cs,
+    minorHAnsi: theme.minor.latin, minorAscii: theme.minor.latin, minorEastAsia: theme.minor.ea, minorBidi: theme.minor.cs };
+  return m[ref] ?? null;
+}
+
+/* ── rPr → run ── */
+function dumpRPr(rPr, theme) {
+  const run = {};
+  const extra = [];
+  if (!rPr) return { run, extra };
+  for (const c of rPr.childElems()) {
+    const v = c.attr('w:val');
+    switch (c.name) {
+      case 'w:rFonts': {
+        const f = {};
+        const ascii = c.attr('w:ascii') ?? resolveThemeFont(theme, c.attr('w:asciiTheme'));
+        const hAnsi = c.attr('w:hAnsi') ?? resolveThemeFont(theme, c.attr('w:hAnsiTheme'));
+        const ea = c.attr('w:eastAsia') ?? resolveThemeFont(theme, c.attr('w:eastAsiaTheme'));
+        const cs = c.attr('w:cs') ?? resolveThemeFont(theme, c.attr('w:cstheme'));
+        if (ascii) f.ascii = ascii;
+        if (hAnsi) f.hAnsi = hAnsi;
+        if (ea) f.eastAsia = ea;
+        if (cs) f.cs = cs;
+        const known = ['w:ascii', 'w:hAnsi', 'w:eastAsia', 'w:cs', 'w:asciiTheme', 'w:hAnsiTheme', 'w:eastAsiaTheme', 'w:cstheme', 'w:hint'];
+        if (c.attrs.every(([k]) => known.includes(k))) run.font = f; else extra.push(serializeNode(c));
+        break;
+      }
+      case 'w:b': run.bold = v !== '0' && v !== 'false'; break;
+      case 'w:i': run.italic = v !== '0' && v !== 'false'; break;
+      case 'w:caps': run.caps = v !== '0' && v !== 'false'; break;
+      case 'w:smallCaps': run.smallCaps = v !== '0' && v !== 'false'; break;
+      case 'w:strike': run.strike = v !== '0' && v !== 'false'; break;
+      case 'w:color':
+        if (c.attrs.length === 1 && v && v !== 'auto') run.color = v;
+        else if (v === 'auto' && c.attrs.length === 1) { /* auto = 继承，丢弃等价 */ }
+        else extra.push(serializeNode(c));
+        break;
+      case 'w:sz': run.sizePt = Number(v) / 2; break;
+      case 'w:szCs': break;   // 由 build 端从 sizePt 再生
+      case 'w:u': run.underline = v; break;
+      case 'w:vertAlign': run.vertAlign = v; break;
+      case 'w:em': run.em = v; break;
+      case 'w:kern': run.kernPt = Number(v) / 2; break;
+      case 'w:spacing': run.spacingTwip = Number(v); break;
+      case 'w:highlight': run.highlight = v; break;
+      case 'w:bCs': case 'w:iCs': break;  // build 端跟随 b/i 语义，可再生；不单独建模
+      default: extra.push(serializeNode(c));
+    }
+  }
+  return { run, extra };
+}
+
+/* ── pPr → para ── */
+const CJK_MAP = { 'w:kinsoku': 'kinsoku', 'w:wordWrap': 'wordWrap', 'w:overflowPunct': 'overflowPunct',
+  'w:autoSpaceDE': 'autoSpaceDE', 'w:autoSpaceDN': 'autoSpaceDN', 'w:adjustRightInd': 'adjustRightInd',
+  'w:snapToGrid': 'snapToGrid' };
+
+function dumpPPr(pPr, theme) {
+  const para = {};
+  const extra = [];
+  let runFromParaMark = null;
+  if (!pPr) return { para, extra, runFromParaMark };
+  for (const c of pPr.childElems()) {
+    const v = c.attr('w:val');
+    const on = v !== '0' && v !== 'false';
+    switch (c.name) {
+      case 'w:jc': para.align = v; break;
+      case 'w:outlineLvl': para.outlineLevel = Number(v); break;
+      case 'w:keepNext': para.keepNext = on; break;
+      case 'w:keepLines': para.keepLines = on; break;
+      case 'w:pageBreakBefore': para.pageBreakBefore = on; break;
+      case 'w:widowControl': para.widowControl = on; break;
+      case 'w:contextualSpacing': para.contextualSpacing = on; break;
+      case 'w:ind': {
+        const i = {};
+        const map = { 'w:left': 'leftTwip', 'w:start': 'leftTwip', 'w:leftChars': 'leftChars',
+          'w:right': 'rightTwip', 'w:end': 'rightTwip', 'w:rightChars': 'rightChars',
+          'w:hanging': 'hangingTwip', 'w:hangingChars': 'hangingChars',
+          'w:firstLine': 'firstLineTwip', 'w:firstLineChars': 'firstLineChars' };
+        let ok = true;
+        for (const [k, val] of c.attrs) {
+          if (map[k]) i[map[k]] = Number(val); else ok = false;
+        }
+        if (ok) para.indent = i; else extra.push(serializeNode(c));
+        break;
+      }
+      case 'w:spacing': {
+        const s = {};
+        let ok = true;
+        for (const [k, val] of c.attrs) {
+          if (k === 'w:before') s.beforePt = Number(val) / 20;
+          else if (k === 'w:after') s.afterPt = Number(val) / 20;
+          else if (k === 'w:beforeLines') s.beforeLines = Number(val);
+          else if (k === 'w:afterLines') s.afterLines = Number(val);
+          else if (k === 'w:line') s._line = Number(val);
+          else if (k === 'w:lineRule') s._rule = val;
+          else if (k === 'w:beforeAutospacing' || k === 'w:afterAutospacing') { if (val !== '0' && val !== 'false') ok = false; }
+          else ok = false;
+        }
+        if (ok) {
+          if (s._line != null) {
+            const rule = s._rule ?? 'auto';
+            if (rule === 'auto') { s.line = s._line / 240; s.lineRule = 'multiple'; }
+            else { s.line = s._line / 20; s.lineRule = rule; }
+          }
+          delete s._line; delete s._rule;
+          para.spacing = s;
+        } else extra.push(serializeNode(c));
+        break;
+      }
+      case 'w:shd':
+        if (c.attr('w:val') === 'clear' && c.attr('w:fill') && c.attr('w:fill') !== 'auto') para.shading = c.attr('w:fill');
+        else extra.push(serializeNode(c));
+        break;
+      case 'w:rPr': {   // 段落标记的 run 属性：作为该风格 run 的一部分
+        runFromParaMark = c;
+        break;
+      }
+      default:
+        if (CJK_MAP[c.name]) { para.cjk = para.cjk ?? {}; para.cjk[CJK_MAP[c.name]] = on; }
+        else extra.push(serializeNode(c));
+    }
+  }
+  return { para, extra, runFromParaMark };
+}
+
+/**
+ * @param {Buffer|string} input docx buffer 或路径
+ * @returns token JSON（provenance:'imported'）+ 覆盖率统计
+ */
+export function dumpStyles(input) {
+  const buf = Buffer.isBuffer(input) ? input : readFileSync(input);
+  const zip = readZip(buf);
+  const theme = parseTheme(zip);
+  const stylesXml = entryData(zip, 'word/styles.xml');
+  if (!stylesXml) throw new Error('no styles.xml');
+  const doc = parseXml(stylesXml.toString());
+
+  const tokens = { v: 1, provenance: 'imported', styles: {}, rawStyles: [] };
+  const stat = { styles: 0, raw: 0, extraFragments: 0 };
+
+  // docDefaults
+  const dd = doc.root.firstChild('w:docDefaults');
+  if (dd) {
+    const rDef = dd.firstChild('w:rPrDefault')?.firstChild('w:rPr');
+    const { run, extra } = dumpRPr(rDef, theme);
+    const pDef = dd.firstChild('w:pPrDefault')?.firstChild('w:pPr');
+    const { para, extra: pExtra } = dumpPPr(pDef, theme);
+    tokens.base = {};
+    if (run.font) { tokens.fonts = { body: run.font }; tokens.base.font = 'body'; }
+    if (run.sizePt != null) tokens.base.sizePt = run.sizePt;
+    if (run.color) tokens.base.color = run.color;
+    if (run.kernPt != null) tokens.base.kernPt = run.kernPt;
+    if (para.spacing) tokens.base.spacing = para.spacing;
+    if (para.cjk) tokens.base.cjk = para.cjk;
+    // 语言
+    const lang = rDef?.firstChild('w:lang');
+    if (lang) tokens.lang = { latin: lang.attr('w:val') ?? 'en-US', eastAsia: lang.attr('w:eastAsia') ?? 'zh-CN' };
+    const leftovers = [...extra, ...pExtra];
+    if (leftovers.length) { tokens.base.extraXml = leftovers; stat.extraFragments += leftovers.length; }
+  }
+
+  // theme 快照（重建 theme part 或字体回填用）
+  if (theme) tokens.theme = theme;
+
+  for (const st of doc.root.childElems('w:style')) {
+    const type = st.attr('w:type');
+    const id = st.attr('w:styleId');
+    if (!id) continue;
+    if (type !== 'paragraph' && type !== 'character') {
+      tokens.rawStyles.push(serializeNode(st));
+      stat.raw += 1;
+      continue;
+    }
+    const s = { type, name: st.firstChild('w:name')?.attr('w:val') ?? id };
+    if (st.firstChild('w:basedOn')) s.basedOn = st.firstChild('w:basedOn').attr('w:val');
+    if (st.firstChild('w:next')) s.next = st.firstChild('w:next').attr('w:val');
+    if (st.firstChild('w:link')) s.link = st.firstChild('w:link').attr('w:val');
+    if (st.firstChild('w:qFormat')) s.qFormat = true;
+    if (st.firstChild('w:uiPriority')) s.uiPriority = Number(st.firstChild('w:uiPriority').attr('w:val'));
+    const { run, extra: rExtra } = dumpRPr(st.firstChild('w:rPr'), theme);
+    const { para, extra: pExtra, runFromParaMark } = dumpPPr(st.firstChild('w:pPr'), theme);
+    if (Object.keys(run).length) s.run = run;
+    if (Object.keys(para).length) s.para = para;
+    const known = ['w:name', 'w:basedOn', 'w:next', 'w:link', 'w:qFormat', 'w:uiPriority', 'w:rPr', 'w:pPr',
+      'w:semiHidden', 'w:unhideWhenUsed', 'w:rsid', 'w:autoRedefine', 'w:hidden', 'w:locked', 'w:aliases'];
+    const otherKids = st.childElems().filter((c) => !known.includes(c.name)).map((c) => serializeNode(c));
+    const leftovers = [...rExtra, ...pExtra, ...otherKids];
+    if (runFromParaMark) leftovers.push(`<w:pPr-rPr>${serializeNode(runFromParaMark)}</w:pPr-rPr>`);
+    if (leftovers.length) { s.extraXml = leftovers; stat.extraFragments += leftovers.length; }
+    tokens.styles[id] = s;
+    stat.styles += 1;
+  }
+
+  // sectPr（版面）
+  const docPart = entryData(zip, 'word/document.xml');
+  if (docPart) {
+    const d = parseXml(docPart.toString());
+    const sect = d.root.firstChild('w:body')?.childElems('w:sectPr')[0];
+    if (sect) {
+      const pgSz = sect.firstChild('w:pgSz');
+      const pgMar = sect.firstChild('w:pgMar');
+      const grid = sect.firstChild('w:docGrid');
+      tokens.page = {};
+      if (pgSz) tokens.page.size = { wTwip: Number(pgSz.attr('w:w')), hTwip: Number(pgSz.attr('w:h')) };
+      if (pgSz?.attr('w:orient') === 'landscape') tokens.page.landscape = true;
+      if (pgMar) {
+        tokens.page.marginsTwip = Object.fromEntries(['top', 'right', 'bottom', 'left', 'header', 'footer', 'gutter']
+          .map((k) => [k, Number(pgMar.attr(`w:${k}`))]).filter(([, v]) => Number.isFinite(v)));
+      }
+      if (grid) {
+        tokens.page.docGrid = { type: grid.attr('w:type') ?? 'default', linePitchTwip: Number(grid.attr('w:linePitch')) };
+        if (grid.attr('w:charSpace')) tokens.page.docGrid.charSpace = Number(grid.attr('w:charSpace'));
+      }
+    }
+  }
+  return { tokens, stat };
+}
+
+/**
+ * 定点样式手术：只改一个 style 的 run 字段（sizePt/color/bold/font…），
+ * styles.xml 其余部分与其他 entry 逐字节不动。
+ * @returns {Buffer} 新 docx
+ */
+export function applyStyleEdit(docxBuf, styleId, patch) {
+  const zip = readZip(docxBuf);
+  const stylesXml = entryData(zip, 'word/styles.xml').toString();
+  const doc = parseXml(stylesXml);
+  const st = doc.root.childElems('w:style').find((s) => s.attr('w:styleId') === styleId);
+  if (!st) throw new Error(`style not found: ${styleId}`);
+  let rPr = st.firstChild('w:rPr');
+  if (!rPr) {
+    rPr = elem('w:rPr');
+    st.append(rPr);            // 位置交给 sortChildren
+    sortChildren(st, { strictUnknown: false });
+  }
+  if (patch.run) {
+    // 用 build 端同一套逻辑生成要写入的元素，再逐个 upsert
+    const fake = { fonts: {} };
+    const built = buildRPr(fake, patch.run) ?? elem('w:rPr');
+    for (const nel of built.childElems()) {
+      const old = rPr.firstChild(nel.name);
+      if (old) rPr.remove(old);
+      rPr.append(nel);
+    }
+    sortChildren(rPr, { strictUnknown: false });
+  }
+  replaceEntry(zip, 'word/styles.xml', serialize(doc));
+  return writeZip(zip);
+}

package/server/lib/docx/fonts/nodesign-cjk.conf

@@ -0,0 +1,164 @@
+<?xml version="1.0"?>
+<!DOCTYPE fontconfig SYSTEM "fonts.dtd">
+<!--
+  nodesign-cjk.conf : 中文 Windows 字体名 → 服务器上真实存在的替身。
+  用法：渲染进程设 FONTCONFIG_FILE=<本文件绝对路径>（per-run，不污染系统）。
+  判据是「产物优先于仪器」：docx 里写的是 Windows 字体名（用户机器上是真的），
+  这份映射只服务 QC 渲染；渲染观感与用户 Word 的已知偏差写进 SKILL。
+
+  映射决策（2026-08-19 换代：雅黑档与仿宋族换真身位替身）：
+    宋体族   -> Noto Serif CJK SC   （衬线对衬线；行高系数 ≈1.24）
+    黑体族   -> Noto Sans CJK SC    （SimHei 是方正黑，Noto Sans 最近）
+    雅黑/等线/苹方 -> MiSans        （⭐行高系数 1.326 ≈ 真微软雅黑 1.32 ——
+                                     multiple 行距下预览页数与 Windows Word 基本对齐，
+                                     Noto Sans 是 1.24 差 6%，2 页简历 Word 变 4 页案的病根之一）
+    楷体族   -> LXGW WenKai         （真楷体）
+    仿宋族   -> Zhuque Fangsong (technical preview)  （朱雀仿宋，OFL —— 真仿宋，公文观感归位）
+    小标宋   -> Noto Serif CJK SC
+  字体文件在 /usr/local/share/fonts/{misans,zhuque}/（系统依赖，换机器要重装）。
+-->
+<fontconfig>
+  <include ignore_missing="no">/etc/fonts/fonts.conf</include>
+
+  <!-- ── 宋体族 → Noto Serif CJK SC ── -->
+  <match target="pattern">
+    <test name="family"><string>SimSun</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>Noto Serif CJK SC</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>宋体</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>Noto Serif CJK SC</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>NSimSun</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>Noto Serif CJK SC</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>新宋体</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>Noto Serif CJK SC</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>Songti SC</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>Noto Serif CJK SC</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>STSong</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>Noto Serif CJK SC</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>华文宋体</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>Noto Serif CJK SC</string></edit>
+  </match>
+
+  <!-- ── 小标宋（公文红头/标题） → Noto Serif CJK SC ── -->
+  <match target="pattern">
+    <test name="family"><string>方正小标宋简体</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>Noto Serif CJK SC</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>FZXiaoBiaoSong-B05S</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>Noto Serif CJK SC</string></edit>
+  </match>
+
+  <!-- ── 黑体族 → Noto Sans CJK SC ── -->
+  <match target="pattern">
+    <test name="family"><string>SimHei</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>Noto Sans CJK SC</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>黑体</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>Noto Sans CJK SC</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>Microsoft YaHei</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>MiSans</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>微软雅黑</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>MiSans</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>Microsoft YaHei UI</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>MiSans</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>DengXian</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>MiSans</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>等线</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>MiSans</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>PingFang SC</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>MiSans</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>苹方</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>MiSans</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>Heiti SC</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>Noto Sans CJK SC</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>华文黑体</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>Noto Sans CJK SC</string></edit>
+  </match>
+
+  <!-- ── 楷体族 → LXGW WenKai（真楷体） ── -->
+  <match target="pattern">
+    <test name="family"><string>KaiTi</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>LXGW WenKai</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>楷体</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>LXGW WenKai</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>楷体_GB2312</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>LXGW WenKai</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>KaiTi_GB2312</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>LXGW WenKai</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>Kaiti SC</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>LXGW WenKai</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>STKaiti</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>LXGW WenKai</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>华文楷体</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>LXGW WenKai</string></edit>
+  </match>
+
+  <!-- ── 仿宋族 → Noto Serif CJK SC（已知偏差，无真仿宋替身） ── -->
+  <match target="pattern">
+    <test name="family"><string>FangSong</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>Zhuque Fangsong (technical preview)</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>仿宋</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>Zhuque Fangsong (technical preview)</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>仿宋_GB2312</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>Zhuque Fangsong (technical preview)</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>FangSong_GB2312</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>Zhuque Fangsong (technical preview)</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>STFangsong</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>Zhuque Fangsong (technical preview)</string></edit>
+  </match>
+  <match target="pattern">
+    <test name="family"><string>华文仿宋</string></test>
+    <edit name="family" mode="prepend" binding="strong"><string>Zhuque Fangsong (technical preview)</string></edit>
+  </match>
+</fontconfig>

package/server/lib/docx/merge-runs.js

@@ -0,0 +1,129 @@
+/**
+ * merge-runs.js — 把 Word 打碎的相邻同格式 run 合回去，让「眼睛看得见的短语」
+ * 在 XML 里重新成为连续字符串（编辑外来 docx 的第一道工序）。
+ *
+ * Word 打碎 run 的三大来源：rsid（修订会话指纹）、proofErr（拼写检查标记）、
+ * 格式微差。合并规则（保守优先，宁可少合不可错合）：
+ *   - 只在同一父元素下合并**相邻**的、只含 w:t 的 run；
+ *   - 中间夹 w:proofErr 视为透明并把标记丢弃（Word 会自己重算拼写）；
+ *   - 夹 bookmark/comment/ins/del/任何别的东西 = 硬边界，不越过；
+ *   - 含 br/tab/drawing/fldChar 等非纯文本的 run = 硬边界；
+ *   - 格式签名 = rPr 规范化（剥 rsid 系属性、按元素名排序后序列化）逐字节相等；
+ *   - 合并结果 w:t 首尾有空白时补 xml:space="preserve"。
+ * 递归进 w:p / w:hyperlink / w:ins / w:del / w:sdtContent / w:smartTag / w:tc 等容器。
+ *
+ * 返回 {merged, dropped}：合并掉的 run 数 / 丢弃的 proofErr 数。
+ */
+
+import { serializeNode, textNode } from './xml.js';
+
+const RSID_ATTR = /^w:rsid/;
+const CONTAINERS = new Set(['w:body', 'w:p', 'w:hyperlink', 'w:ins', 'w:del', 'w:sdt',
+  'w:sdtContent', 'w:smartTag', 'w:tc', 'w:tr', 'w:tbl', 'w:hdr', 'w:ftr',
+  'w:footnote', 'w:endnote', 'w:comment', 'w:fldSimple']);
+
+/** run 是否为「纯文本 run」（只含 rPr 和 w:t） */
+function isTextOnlyRun(r) {
+  if (r.name !== 'w:r') return false;
+  for (const c of r.children) {
+    if (c.type !== 'elem') continue;
+    if (c.name === 'w:rPr' || c.name === 'w:t') continue;
+    return false;
+  }
+  return true;
+}
+
+/** rPr 的格式签名：剥 rsid、子元素按名排序（同名保序）、规范序列化 */
+function runSig(r) {
+  const rPr = r.firstChild('w:rPr');
+  if (!rPr) return '';
+  const parts = [];
+  for (const c of rPr.children) {
+    if (c.type !== 'elem') continue;
+    const attrs = c.attrs.filter(([k]) => !RSID_ATTR.test(k));
+    parts.push(`${c.name}|${attrs.map(([k, v]) => `${k}=${v}`).sort().join(',')}`
+      + (c.children.length ? `|${serializeNode(c)}` : ''));
+  }
+  return parts.sort().join(';');
+}
+
+function runText(r) {
+  let s = '';
+  for (const c of r.children) {
+    if (c.type === 'elem' && c.name === 'w:t') s += c.text();
+  }
+  return s;
+}
+
+function setRunText(r, text) {
+  const t = r.firstChild('w:t');
+  if (!t) return;
+  // 重建 w:t（脏节点走规范序列化）
+  const tn = textNode(text);
+  tn.parent = t;
+  t.children = [tn];
+  if (/^\s|\s$/.test(text)) {
+    if (!t.attr('xml:space')) t.attrs.push(['xml:space', 'preserve']);
+  } else {
+    const i = t.attrs.findIndex(([k]) => k === 'xml:space');
+    if (i >= 0) t.attrs.splice(i, 1);
+  }
+  t.markDirty();
+}
+
+export function mergeRuns(root) {
+  const stats = { merged: 0, dropped: 0 };
+  walk(root, stats);
+  return stats;
+}
+
+function walk(node, stats) {
+  if (node.type !== 'elem') return;
+  if (CONTAINERS.has(node.name)) mergeInContainer(node, stats);
+  for (const c of node.children) walk(c, stats);
+}
+
+function mergeInContainer(parent, stats) {
+  const kids = parent.children;
+  let changed = false;
+  let i = 0;
+  while (i < kids.length) {
+    const a = kids[i];
+    if (!(a.type === 'elem' && isTextOnlyRun(a))) { i += 1; continue; }
+    const sigA = runSig(a);
+    let acc = null;   // 合并累积文本（惰性启动）
+    let j = i + 1;
+    const absorbed = [];   // 将被移除的节点（run 与 proofErr）
+    while (j < kids.length) {
+      const b = kids[j];
+      if (b.type === 'elem' && b.name === 'w:proofErr') {
+        absorbed.push(b); j += 1; continue;      // 透明，越过并丢弃
+      }
+      if (b.type === 'elem' && isTextOnlyRun(b) && runSig(b) === sigA) {
+        if (acc == null) acc = runText(a);
+        acc += runText(b);
+        absorbed.push(b);
+        stats.merged += 1;
+        j += 1;
+        continue;
+      }
+      break;
+    }
+    if (acc != null) {
+      setRunText(a, acc);
+      for (const n of absorbed) {
+        const k = kids.indexOf(n);
+        if (n.name === 'w:proofErr') stats.dropped += 1;
+        kids.splice(k, 1);
+      }
+      parent.markDirty();
+      changed = true;
+      // a 现在是合并后的 run；从 a 之后继续（可能还有下一簇）
+      i = kids.indexOf(a) + 1;
+    } else {
+      // 没合成任何东西：被 absorb 的 proofErr 留在原位（不动干净区域）
+      i += 1;
+    }
+  }
+  return changed;
+}