gardel/Mofox-Core
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

refactor(chat): 重构图片在聊天记录中的处理与表示方式

Browse Source 
为了简化LLM的上下文并提高代码可维护性，对聊天记录中图片的处理方式进行了彻底重构。

旧系统使用 [图片1] 等占位符，并在消息头部附加一个独立的图片描述映射块。这种方式结构复杂，容易造成上下文分离。

新系统将图片描述直接内联到消息文本中，格式为 `[图片：一只猫]`，使聊天记录对LLM更加自然和易于理解。

主要变更：
- **消息构建 (`chat_message_builder`):** 在构建可读消息时，异步将数据库中的 `[picid:...]` 标签直接替换为完整的 `[图片：描述]`。
- **废弃映射:** 移除了独立的图片映射信息块 (`build_pic_mapping_info` 函数)，所有信息都在消息内联。
- **图片处理 (`utils_image`):** `process_image` 流程现在同步返回完整的描述字符串，并增强了VLM调用的重试逻辑和缓存机制，提高了健壮性。
- **消息存储 (`storage`):** 在消息存入数据库前，将 `[图片：描述]` 转换为 `[picid:...]`，以保持存储规范化。
- **修复:** 增加了多处空值检查，提高了代码的稳定性。这不得之前稳定好用多了😋😋😋
This commit is contained in:
tt-P607
2025-10-21 20:14:58 +08:00
committed by Windpicker-owo
parent 3accbabd3a
commit cba200520b
4 changed files with 198 additions and 291 deletions
Download Patch File Download Diff File Expand all files Collapse all files

167
src/chat/utils/chat_message_builder.py
View File

@@ -547,6 +547,36 @@ async def _build_readable_messages_internal(
if pic_id_mapping is None:
pic_id_mapping = {}
current_pic_counter = pic_counter
# --- 异步图片ID处理器 (修复核心问题) ---
async def process_pic_ids(content: str) -> str:
"""异步处理内容中的图片ID，将其直接替换为[图片：描述]格式"""
pic_pattern = r"\[picid:([^\]]+)\]"
matches = list(re.finditer(pic_pattern, content))
if not matches:
return content
new_content = ""
last_end = 0
for match in matches:
new_content += content[last_end : match.start()]
pic_id = match.group(1)
description = "[图片内容未知]"
try:
async with get_db_session() as session:
result = await session.execute(select(Images.description).where(Images.image_id == pic_id))
desc_scalar = result.scalar_one_or_none()
if desc_scalar and desc_scalar.strip():
description = f"[图片：{desc_scalar}]"
else:
description = "[图片内容未知]"
except Exception as e:
logger.debug(f"[chat_message_builder] 查询图片 {pic_id} 描述失败: {e}")
description = "[图片内容未知]"
new_content += description
last_end = match.end()
new_content += content[last_end:]
return new_content
# 创建时间戳到消息ID的映射，用于在消息前添加[id]标识符
timestamp_to_id = {}
@@ -557,25 +587,6 @@ async def _build_readable_messages_internal(
if timestamp is not None:
timestamp_to_id[timestamp] = item.get("id", "")
def process_pic_ids(content: str) -> str:
"""处理内容中的图片ID，将其替换为[图片x]格式"""
nonlocal current_pic_counter
# 匹配 [picid:xxxxx] 格式
pic_pattern = r"\[picid:([^\]]+)\]"
def replace_pic_id(match):
nonlocal current_pic_counter
pic_id = match.group(1)
if pic_id not in pic_id_mapping:
pic_id_mapping[pic_id] = f"图片{current_pic_counter}"
current_pic_counter += 1
return f"[{pic_id_mapping[pic_id]}]"
return re.sub(pic_pattern, replace_pic_id, content)
# 1 & 2: 获取发送者信息并提取消息组件
for msg in messages:
# 检查是否是动作记录
@@ -583,8 +594,8 @@ async def _build_readable_messages_internal(
is_action = True
timestamp: float = msg.get("time") # type: ignore
content = msg.get("display_message", "")
# 对于动作记录，也处理图片ID
content = process_pic_ids(content)
if show_pic:
content = await process_pic_ids(content)
message_details_raw.append((timestamp, global_config.bot.nickname, content, is_action))
continue
@@ -619,7 +630,7 @@ async def _build_readable_messages_internal(
# 处理图片ID
if show_pic:
content = process_pic_ids(content)
content = await process_pic_ids(content)
# 检查必要信息是否存在
if not all([platform, user_id, timestamp is not None]):
@@ -808,43 +819,12 @@ async def _build_readable_messages_internal(
current_pic_counter,
)
async def build_pic_mapping_info(pic_id_mapping: dict[str, str]) -> str:
# sourcery skip: use-contextlib-suppress
"""
构建图片映射信息字符串，显示图片的具体描述内容
Args:
pic_id_mapping: 图片ID到显示名称的映射字典
Returns:
格式化的映射信息字符串
此函数已废弃，因为图片描述现在被内联处理。
保留此函数以确保向后兼容性，但它将始终返回一个空字符串。
"""
if not pic_id_mapping:
return ""
mapping_lines = []
# 按图片编号排序
sorted_items = sorted(pic_id_mapping.items(), key=lambda x: int(x[1].replace("图片", "")))
for pic_id, display_name in sorted_items:
# 从数据库中获取图片描述
description = "[图片内容未知]" # 默认描述
try:
async with get_db_session() as session:
result = await session.execute(select(Images).where(Images.image_id == pic_id))
image = result.scalar_one_or_none()
if image and hasattr(image, "description") and image.description:
description = image.description
except Exception as e:
# 如果查询失败，保持默认描述
logger.debug(f"[chat_message_builder] 查询图片描述失败: {e}")
pass
mapping_lines.append(f"[{display_name}] 的内容：{description}")
return "\n".join(mapping_lines)
return ""
def build_readable_actions(actions: list[dict[str, Any]]) -> str:
@@ -932,13 +912,9 @@ async def build_readable_messages_with_list(
将消息列表转换为可读的文本格式，并返回原始(时间戳, 昵称, 内容)列表。
允许通过参数控制格式化行为。
"""
formatted_string, details_list, pic_id_mapping, _ = await _build_readable_messages_internal(
formatted_string, details_list, _, _ = await _build_readable_messages_internal(
messages, replace_bot_name, merge_messages, timestamp_mode, truncate
)
if pic_mapping_info := await build_pic_mapping_info(pic_id_mapping):
formatted_string = f"{pic_mapping_info}\n\n{formatted_string}"
return formatted_string, details_list
@@ -970,12 +946,6 @@ async def build_readable_messages_with_id(
message_id_list=message_id_list,
)
# 如果存在图片映射信息，附加之
if pic_mapping_info := await build_pic_mapping_info({}):
# 如果当前没有图片映射则不附加
if pic_mapping_info:
formatted_string = f"{pic_mapping_info}\n\n{formatted_string}"
return formatted_string, message_id_list
@@ -1078,7 +1048,7 @@ async def build_readable_messages(
if read_mark <= 0:
# 没有有效的 read_mark，直接格式化所有消息
formatted_string, _, pic_id_mapping, _ = await _build_readable_messages_internal(
formatted_string, _, _, _ = await _build_readable_messages_internal(
copy_messages,
replace_bot_name,
merge_messages,
@@ -1088,12 +1058,7 @@ async def build_readable_messages(
message_id_list=message_id_list,
)
# 生成图片映射信息并添加到最前面
pic_mapping_info = await build_pic_mapping_info(pic_id_mapping)
if pic_mapping_info:
return f"{pic_mapping_info}\n\n{formatted_string}"
else:
return formatted_string
return formatted_string
else:
# 按 read_mark 分割消息
messages_before_mark = [msg for msg in copy_messages if msg.get("time", 0) <= read_mark]
@@ -1128,23 +1093,15 @@ async def build_readable_messages(
)
read_mark_line = "\n--- 以上消息是你已经看过，请关注以下未读的新消息---\n"
# 生成图片映射信息
if pic_id_mapping:
pic_mapping_info = f"图片信息：\n{await build_pic_mapping_info(pic_id_mapping)}\n聊天记录信息：\n"
else:
pic_mapping_info = "聊天记录信息：\n"
# 组合结果
result_parts = []
if pic_mapping_info:
result_parts.extend((pic_mapping_info, "\n"))
if formatted_before and formatted_after:
result_parts.extend([formatted_before, read_mark_line, formatted_after])
elif formatted_before:
result_parts.extend([formatted_before, read_mark_line])
elif formatted_after:
result_parts.extend([read_mark_line, formatted_after])
result_parts.extend([read_mark_line.strip(), formatted_after])
else:
result_parts.append(read_mark_line.strip())
@@ -1164,28 +1121,9 @@ async def build_anonymous_messages(messages: list[dict[str, Any]]) -> str:
current_char = ord("A")
output_lines = []
# 图片ID映射字典
pic_id_mapping = {}
pic_counter = 1
def process_pic_ids(content: str) -> str:
"""处理内容中的图片ID，将其替换为[图片x]格式"""
nonlocal pic_counter
# 匹配 [picid:xxxxx] 格式
pic_pattern = r"\[picid:([^\]]+)\]"
def replace_pic_id(match):
nonlocal pic_counter
pic_id = match.group(1)
if pic_id not in pic_id_mapping:
pic_id_mapping[pic_id] = f"图片{pic_counter}"
pic_counter += 1
return f"[{pic_id_mapping[pic_id]}]"
return re.sub(pic_pattern, replace_pic_id, content)
# This function builds anonymous messages, so we don't need full descriptions.
# The existing placeholder logic is sufficient.
# However, to maintain consistency, we will adapt it slightly.
def get_anon_name(platform, user_id):
# print(f"get_anon_name: platform:{platform}, user_id:{user_id}")
@@ -1222,8 +1160,8 @@ async def build_anonymous_messages(messages: list[dict[str, Any]]) -> str:
if "ⁿ" in content:
content = content.replace("ⁿ", "")
# 处理图片ID
content = process_pic_ids(content)
# For anonymous messages, we just replace with a placeholder.
content = re.sub(r"\[picid:([^\]]+)\]", "[图片]", content)
# if not all([platform, user_id, timestamp is not None]):
# continue
@@ -1252,15 +1190,8 @@ async def build_anonymous_messages(messages: list[dict[str, Any]]) -> str:
except Exception:
continue
# 在最前面添加图片映射信息
final_output_lines = []
pic_mapping_info = await build_pic_mapping_info(pic_id_mapping)
if pic_mapping_info:
final_output_lines.append(pic_mapping_info)
final_output_lines.append("\n\n")
final_output_lines.extend(output_lines)
formatted_string = "".join(final_output_lines).strip()
# Since we are not generating a pic_mapping_info block, just join and return.
formatted_string = "".join(output_lines).strip()
return formatted_string
@@ -1295,4 +1226,4 @@ async def get_person_id_list(messages: list[dict[str, Any]]) -> list[str]:
if person_id := get_person_id(platform, user_id):
person_ids_set.add(person_id)
return list(person_ids_set) # 将集合转换为列表返回
return list(person_ids_set)