jhzhang/rag
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(rag): 修复表格图片替换和跨页合并三个bug

Browse Source 
1. chat_routes._replace_table_image_placeholders:
   - 处理 ![图片](图片URL) LLM 格式（去掉多余!和假URL）
   - 支持 [N张图片] 多张占位符（html_table_to_markdown 生成）
   - 修复 !! 双感叹号导致 markdown 图片语法失效

2. manager._merge_cross_page_tables:
   - 表头跳过逻辑改为对比第一行而非 find_all('th')
   - 修复合并后表格中间出现重复表头行的问题
This commit is contained in:
lacerate551
2026-06-21 00:40:12 +08:00
parent eb177e11e5
commit f0e5426b4d
2 changed files with 65 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
knowledge/manager.py
View File

@@ -39,6 +39,7 @@ from pathlib import Path
import logging
import chromadb
from bs4 import BeautifulSoup
# 从 base.py 导入基础类和常量
from .base import (
@@ -550,8 +551,39 @@ class KnowledgeBaseManager(
curr_html = getattr(current, 'table_html', '') or ''
next_html = getattr(next_chunk, 'table_html', '') or ''
if curr_html and next_html:
# 合并两个表格的 HTML
current.table_html = curr_html + '\n' + next_html
# 正确合并两个表格的 HTML：
# 将第二个表格的 <tr> 行追加到第一个表格中
# （而非简单拼接两个 <table>，否则 html_table_to_markdown
# 的 soup.find('table') 只能找到第一个表格）
try:
soup1 = BeautifulSoup(curr_html, 'html.parser')
soup2 = BeautifulSoup(next_html, 'html.parser')
table1 = soup1.find('table')
table2 = soup2.find('table')
if table1 and table2:
# 从第二个表格提取数据行
next_rows = table2.find_all('tr')
# 跳过与第一个表格表头重复的行
# 对比第一行而非所有 th（find_all('th') 会匹配
# 整个表格的 th，无法与单行做列表比较）
first_row_t1 = table1.find('tr')
if first_row_t1 and next_rows:
row1_texts = [c.get_text(strip=True) for c in first_row_t1.find_all(['th', 'td'])]
row2_texts = [c.get_text(strip=True) for c in next_rows[0].find_all(['th', 'td'])]
if row1_texts and row2_texts and row1_texts == row2_texts:
next_rows = next_rows[1:]
logger.debug("跨页表格合并: 跳过了重复的表头行")
for row in next_rows:
table1.append(row)
current.table_html = str(soup1)
logger.debug(f"跨页表格 HTML 合并成功: 追加了 {len(next_rows)} 行")
else:
current.table_html = curr_html + '\n' + next_html
except Exception as e:
logger.warning(f"跨页表格 HTML 合并异常: {e}，回退到简单拼接")
current.table_html = curr_html + '\n' + next_html
elif not curr_html and next_html:
current.table_html = next_html
# 合并 image_path 和嵌入图片到 images
curr_img = getattr(current, 'image_path', None)