From 52bb8858c8fd812415af94de83d962da7fac71c1 Mon Sep 17 00:00:00 2001
From: Leon <leon@langcore.cn>
Date: Mon, 8 Jun 2026 15:00:31 +0800
Subject: [PATCH] =?UTF-8?q?fix:=20=E7=BF=BB=E8=AF=91=E5=89=8D=E5=90=88?=
 =?UTF-8?q?=E5=B9=B6=E7=9B=B8=E9=82=BB=E5=90=8C=E6=A0=BC=E5=BC=8FRun?=
 =?UTF-8?q?=EF=BC=8C=E8=A7=A3=E5=86=B3Word=E5=BE=AE=E8=A7=82Run=E7=A2=8E?=
 =?UTF-8?q?=E7=89=87=E5=AF=BC=E8=87=B4=E7=9A=84=E6=A0=BC=E5=BC=8F=E6=B7=B7?=
 =?UTF-8?q?=E4=B9=B1?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- 新增_run_format_key(): 生成Run格式签名(bold/italic/size/color/font)
- 新增_merge_adjacent_runs(): 合并相邻同格式Run，减少Run碎片
- P91从42个Run合并为4个(2格式交替)，P92从50个合并为1个
- 合并后翻译比例分配不再丢失内容和格式

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
---
 .../ai_translator/docx_translator.py          | 53 ++++++++++++++++---
 1 file changed, 47 insertions(+), 6 deletions(-)

diff --git a/docutranslate/translator/ai_translator/docx_translator.py b/docutranslate/translator/ai_translator/docx_translator.py
index 006c9a5..dbb8787 100644
--- a/docutranslate/translator/ai_translator/docx_translator.py
+++ b/docutranslate/translator/ai_translator/docx_translator.py
@@ -91,11 +91,52 @@ class DocxTranslator(AiTranslator):
         self.insert_mode = config.insert_mode
         self.separator = config.separator
 
+    @staticmethod
+    def _run_format_key(run: Run):
+        """生成 Run 的格式签名，用于合并相同格式的 Run。"""
+        return (
+            run.bold,
+            run.italic,
+            run.underline,
+            run.font.size if run.font.size else None,
+            str(run.font.color.rgb) if run.font.color and run.font.color.rgb else None,
+            run.font.name or None,
+        )
+
+    def _merge_adjacent_runs(self, runs: List[Run]) -> List[Run]:
+        """
+        合并相邻的、格式完全相同的 Run，同时累积文本到第一个 Run。
+        解决 Word 因修订历史/变更追踪产生的微观 Run 碎片问题（单个字符一个 Run）。
+        """
+        if len(runs) <= 1:
+            return runs
+
+        merged = []
+        group_start = 0
+        for i in range(1, len(runs)):
+            if self._run_format_key(runs[i]) != self._run_format_key(runs[group_start]):
+                # Format boundary: finalize the current group
+                if i - group_start > 1:
+                    # Merge: accumulate all text into first run, delete the rest
+                    runs[group_start].text = "".join(r.text for r in runs[group_start:i])
+                    for r in runs[group_start + 1:i]:
+                        self._remove_run_element(r)
+                merged.append(runs[group_start])
+                group_start = i
+
+        # Final group
+        if len(runs) - group_start > 1:
+            runs[group_start].text = "".join(r.text for r in runs[group_start:])
+            for r in runs[group_start + 1:]:
+                self._remove_run_element(r)
+        merged.append(runs[group_start])
+
+        return merged
+
     def _process_paragraph(self, para: Paragraph, elements: List[Dict[str, Any]], texts: List[str],
                          top_level_para: Paragraph = None):
         """
-        简化版段落处理：将段落内所有文本 Run 收集为一个翻译单元，完整保留格式结构。
-        不再按格式变化切分 segment，翻译质量更好、格式保留 100%。
+        段落级翻译处理：收集所有文本 Run → 合并相邻同格式 Run → 整段翻译 → 按比例分配。
         """
         if top_level_para is None:
             top_level_para = para
@@ -103,14 +144,14 @@ class DocxTranslator(AiTranslator):
         text_runs = []
         for run in para.runs:
             if is_image_run(run):
-                continue  # 跳过图片
-            if not run.text.strip():
-                # 保留带格式的空 Run（如下划线空格），但不加入文本
                 continue
-
+            if not run.text.strip():
+                continue
             text_runs.append(run)
 
         if text_runs:
+            # Merge adjacent runs with identical formatting to reduce fragmentation
+            text_runs = self._merge_adjacent_runs(text_runs)
             full_text = "".join(r.text for r in text_runs)
             if full_text.strip():
                 elements.append({