Function resolve_conflict

mainprogress/pdf_metadata_extractor.py:367–424

(p)

                raw_save_path = os.path.join(initial_data_dir, raw_filename)
                
                b64_img = create_concat_image_b64(doc, p, p, save_path=img_save_path)
                if not b64_img:
                    return p, False, None
                    
                prompt = f"""这是一张 PDF 页面的图片，物理页码为 {p}。
请判断这一页是否是目录。目录的严格定义为：这张图中是否能提取出多个标题 - 页码对。
【输出要求】：
1. 仅输出 JSON 格式，不要包含任何 markdown 标记。
2. 如果是目录，输出：{{"is_toc": true}}
3. 如果不是目录，输出：{{"is_toc": false}}"""

                try:
                    completion = await client.chat.completions.create(
                        model=model,
                        messages=[
                            {
                                "role": "user",
                                "content": [
                                    {"type": "text", "text": prompt},
                                    {"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{b64_img}"}},
                                ]
                            }
                        ],
                        extra_body={"enable_thinking": False},
                        temperature=0,
                    )
                    raw_content = completion.choices[0].message.content.strip()
                    
                    with open(raw_save_path, 'w', encoding='utf-8') as f:
                        json.dump({"page": p, "raw_response": raw_content}, f, ensure_ascii=False, indent=2)
                        
                    clean_text = re.sub(r'^```(?:json)?\s*', '', raw_content, flags=re.MULTILINE)
                    clean_text = re.sub(r'\s*```$', '', clean_text, flags=re.MULTILINE)
                    data = json.loads(clean_text)
                    is_toc = data.get("is_toc", False)
                    
                    detail_filename = f"single_vote_detail_{p}.json"
                    detail_save_path = os.path.join(initial_data_dir, detail_filename)
                    detail_data = {
                        "page": p,
                        "parsed_result": {"is_toc": is_toc},
                        "raw_response": raw_content,
                        "image_saved_as": img_filename
                    }
                    with open(detail_save_path, 'w', encoding='utf-8') as f:
                        json.dump(detail_data, f, ensure_ascii=False, indent=2)
                        
                    return p, is_toc, raw_content
                except Exception as e:
                    logger.error(f"单页投票失败 (页码 {p}): {e}")
                    return p, False, None

        conflict_tasks = [resolve_conflict(p) for p in conflict_pages]
        conflict_results = await asyncio.gather(*conflict_tasks)
        
        for p, is_toc, _ in conflict_results:
            if is_toc:
                final_toc_pages.append(p)