Function parse_csv_response

mainprogress/determine_toc_levels.py:184–240

将模型返回的 CSV 文本解析为目标 JSON 结构。 目标结构：[{"text": "...", "number": int, "level": int}, ...]

(csv_text: str, source_file: str)

    try:
        reader = csv.DictReader(StringIO(cleaned_text))
        # 检查必要的列
        if not {'title', 'page_number', 'level'}.issubset(set(reader.fieldnames or [])):
            raise ValueError(f"CSV 缺少必要列 (title, page_number, level)。实际列：{reader.fieldnames}")
            
        for row in reader:
            try:
                title = row['title'].strip()
                if not title:
                    continue
                
                # 尝试转换页码
                page_num_str = row['page_number'].strip()
                page_num = int(page_num_str) if page_num_str else None
                
                # 尝试转换层级
                level_str = row['level'].strip()
                level = int(level_str) if level_str else None
                
                if page_num is None or level is None:
                    write_log(f"警告：跳过无效行 (页码或层级非整数)：{row} in {source_file}")
                    continue

                result_data.append({
                    "text": title,
                    "number": page_num,
                    "level": level
                })
            except (ValueError, KeyError) as e:
                write_log(f"解析单行失败：{row}, 错误：{e}")
                continue
                
    except Exception as e:
        write_log(f"CSV 整体解析失败 {source_file}: {str(e)}")
        raise e
        
    return result_data

async def process_first_page(img_file: Path, csv_file: Path, output_path: Path) -> bool:
    """
    专门处理第一张图片，获取 CSV 格式的响应，并缓存为 Few-shot 示例。
    """
    write_log(f"正在处理首图作为示例：{img_file.name}")
    
    if not csv_file.exists():
        write_log(f"缺少对应的 CSV 文件，跳过首图处理：{img_file.name}")
        return False

    csv_content = csv_file.read_text(encoding='utf-8')
    
    content_list = [
        {"type": "text", "text": "当前页图片（需处理，作为后续页面的参考示例）："},
        {"type": "text", "text": f"{PROMPT_TEXT}\n\n当前页提取的原始 CSV 数据如下：\n{csv_content}"},
        {"type": "image_url", "image_url": {"url": get_encoded_image(img_file)}}
    ]

    messages = [{"role": "user", "content": content_list}]

    try: