Files
Gitcloned-BettaFish/ReportEngine/renderers/pdf_renderer.py
2025-11-20 00:34:01 +08:00

711 lines
24 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
"""
PDF渲染器 - 使用WeasyPrint从HTML生成PDF
支持完整的CSS样式和中文字体
"""
from __future__ import annotations
import base64
import copy
from pathlib import Path
from typing import Any, Dict
from datetime import datetime
from loguru import logger
try:
from weasyprint import HTML, CSS
from weasyprint.text.fonts import FontConfiguration
WEASYPRINT_AVAILABLE = True
except (ImportError, OSError) as e:
WEASYPRINT_AVAILABLE = False
# 判断错误类型以提供更友好的提示
if isinstance(e, OSError):
logger.warning(
"PDF 导出依赖缺失(系统库未安装或环境变量未设置),"
"PDF 导出功能将不可用。其他功能不受影响。"
)
else:
logger.warning("WeasyPrint未安装PDF导出功能将不可用")
except Exception as e:
WEASYPRINT_AVAILABLE = False
logger.warning(f"WeasyPrint 加载失败: {e}PDF导出功能将不可用")
from .html_renderer import HTMLRenderer
from .pdf_layout_optimizer import PDFLayoutOptimizer, PDFLayoutConfig
from .chart_to_svg import create_chart_converter
from .math_to_svg import MathToSVG
class PDFRenderer:
"""
基于WeasyPrint的PDF渲染器
- 直接从HTML生成PDF保留所有CSS样式
- 完美支持中文字体
- 自动处理分页和布局
"""
def __init__(
self,
config: Dict[str, Any] | None = None,
layout_optimizer: PDFLayoutOptimizer | None = None
):
"""
初始化PDF渲染器
参数:
config: 渲染器配置
layout_optimizer: PDF布局优化器可选
"""
self.config = config or {}
self.html_renderer = HTMLRenderer(config)
self.layout_optimizer = layout_optimizer or PDFLayoutOptimizer()
if not WEASYPRINT_AVAILABLE:
raise RuntimeError("WeasyPrint未安装请运行: pip install weasyprint")
# 初始化图表转换器
try:
font_path = self._get_font_path()
self.chart_converter = create_chart_converter(font_path=str(font_path))
logger.info("图表SVG转换器初始化成功")
except Exception as e:
logger.warning(f"图表SVG转换器初始化失败: {e},将使用表格降级")
# 初始化数学公式转换器
try:
self.math_converter = MathToSVG(font_size=16, color='black')
logger.info("数学公式SVG转换器初始化成功")
except Exception as e:
logger.warning(f"数学公式SVG转换器初始化失败: {e},公式将显示为文本")
self.math_converter = None
@staticmethod
def _get_font_path() -> Path:
"""获取字体文件路径"""
# 优先使用完整字体以确保字符覆盖
fonts_dir = Path(__file__).parent / "assets" / "fonts"
# 检查完整字体
full_font = fonts_dir / "SourceHanSerifSC-Medium.otf"
if full_font.exists():
logger.info(f"使用完整字体: {full_font}")
return full_font
# 检查TTF子集字体
subset_ttf = fonts_dir / "SourceHanSerifSC-Medium-Subset.ttf"
if subset_ttf.exists():
logger.info(f"使用TTF子集字体: {subset_ttf}")
return subset_ttf
# 检查OTF子集字体
subset_otf = fonts_dir / "SourceHanSerifSC-Medium-Subset.otf"
if subset_otf.exists():
logger.info(f"使用OTF子集字体: {subset_otf}")
return subset_otf
raise FileNotFoundError(f"未找到字体文件,请检查 {fonts_dir} 目录")
def _preprocess_charts(self, document_ir: Dict[str, Any]) -> Dict[str, Any]:
"""
预处理图表:验证和修复所有图表数据
这个方法确保在转换为SVG之前所有图表数据都是有效的。
使用与HTMLRenderer相同的验证和修复逻辑保证PDF和HTML的一致性。
参数:
document_ir: Document IR数据
返回:
Dict[str, Any]: 修复后的Document IR深拷贝
"""
# 深拷贝以避免修改原始IR
ir_copy = copy.deepcopy(document_ir)
repair_stats = {
'total': 0,
'repaired': 0,
'failed': 0
}
def repair_widgets_in_blocks(blocks: list) -> None:
"""递归修复blocks中的所有widget"""
for block in blocks:
if not isinstance(block, dict):
continue
# 处理widget类型
if block.get('type') == 'widget':
widget_type = block.get('widgetType', '')
if widget_type.startswith('chart.js'):
repair_stats['total'] += 1
# 使用HTMLRenderer的验证器和修复器
validation = self.html_renderer.chart_validator.validate(block)
if not validation.is_valid:
logger.debug(f"图表 {block.get('widgetId')} 需要修复: {validation.errors}")
# 尝试修复
repair_result = self.html_renderer.chart_repairer.repair(block, validation)
if repair_result.success and repair_result.repaired_block:
# 更新block内容在副本中
block.update(repair_result.repaired_block)
repair_stats['repaired'] += 1
logger.debug(
f"图表 {block.get('widgetId')} 已修复 "
f"(方法: {repair_result.method})"
)
else:
repair_stats['failed'] += 1
logger.warning(
f"图表 {block.get('widgetId')} 修复失败,将使用原始数据"
)
# 递归处理嵌套的blocks
nested_blocks = block.get('blocks')
if isinstance(nested_blocks, list):
repair_widgets_in_blocks(nested_blocks)
# 处理列表项
if block.get('type') == 'list':
items = block.get('items', [])
for item in items:
if isinstance(item, list):
repair_widgets_in_blocks(item)
# 处理表格单元格
if block.get('type') == 'table':
rows = block.get('rows', [])
for row in rows:
cells = row.get('cells', [])
for cell in cells:
cell_blocks = cell.get('blocks', [])
if isinstance(cell_blocks, list):
repair_widgets_in_blocks(cell_blocks)
# 处理所有章节
chapters = ir_copy.get('chapters', [])
for chapter in chapters:
blocks = chapter.get('blocks', [])
repair_widgets_in_blocks(blocks)
# 输出统计信息
if repair_stats['total'] > 0:
logger.info(
f"PDF图表预处理完成: "
f"总计 {repair_stats['total']} 个图表, "
f"修复 {repair_stats['repaired']} 个, "
f"失败 {repair_stats['failed']}"
)
return ir_copy
def _convert_charts_to_svg(self, document_ir: Dict[str, Any]) -> Dict[str, str]:
"""
将document_ir中的所有图表转换为SVG
参数:
document_ir: Document IR数据
返回:
Dict[str, str]: widgetId到SVG字符串的映射
"""
svg_map = {}
if not hasattr(self, 'chart_converter') or not self.chart_converter:
logger.warning("图表转换器未初始化,跳过图表转换")
return svg_map
# 遍历所有章节
chapters = document_ir.get('chapters', [])
for chapter in chapters:
blocks = chapter.get('blocks', [])
self._extract_and_convert_widgets(blocks, svg_map)
logger.info(f"成功转换 {len(svg_map)} 个图表为SVG")
return svg_map
def _extract_and_convert_widgets(
self,
blocks: list,
svg_map: Dict[str, str]
) -> None:
"""
递归遍历blocks找到所有widget并转换为SVG
参数:
blocks: block列表
svg_map: 用于存储转换结果的字典
"""
for block in blocks:
if not isinstance(block, dict):
continue
block_type = block.get('type')
# 处理widget类型
if block_type == 'widget':
widget_id = block.get('widgetId')
widget_type = block.get('widgetType', '')
# 只处理chart.js类型的widget
if widget_id and widget_type.startswith('chart.js'):
try:
svg_content = self.chart_converter.convert_widget_to_svg(
block,
width=800,
height=500,
dpi=100
)
if svg_content:
svg_map[widget_id] = svg_content
logger.debug(f"图表 {widget_id} 转换为SVG成功")
else:
logger.warning(f"图表 {widget_id} 转换为SVG失败")
except Exception as e:
logger.error(f"转换图表 {widget_id} 时出错: {e}")
# 递归处理嵌套的blocks
nested_blocks = block.get('blocks')
if isinstance(nested_blocks, list):
self._extract_and_convert_widgets(nested_blocks, svg_map)
# 处理列表项
if block_type == 'list':
items = block.get('items', [])
for item in items:
if isinstance(item, list):
self._extract_and_convert_widgets(item, svg_map)
# 处理表格单元格
if block_type == 'table':
rows = block.get('rows', [])
for row in rows:
cells = row.get('cells', [])
for cell in cells:
cell_blocks = cell.get('blocks', [])
if isinstance(cell_blocks, list):
self._extract_and_convert_widgets(cell_blocks, svg_map)
def _convert_math_to_svg(self, document_ir: Dict[str, Any]) -> Dict[str, str]:
"""
将document_ir中的所有数学公式转换为SVG
参数:
document_ir: Document IR数据
返回:
Dict[str, str]: 公式块ID到SVG字符串的映射
"""
svg_map = {}
if not hasattr(self, 'math_converter') or not self.math_converter:
logger.warning("数学公式转换器未初始化,跳过公式转换")
return svg_map
# 遍历所有章节
chapters = document_ir.get('chapters', [])
for chapter in chapters:
blocks = chapter.get('blocks', [])
self._extract_and_convert_math_blocks(blocks, svg_map)
logger.info(f"成功转换 {len(svg_map)} 个数学公式为SVG")
return svg_map
def _extract_and_convert_math_blocks(
self,
blocks: list,
svg_map: Dict[str, str],
block_counter: list = None
) -> None:
"""
递归遍历blocks找到所有math块并转换为SVG
参数:
blocks: block列表
svg_map: 用于存储转换结果的字典
block_counter: 用于生成唯一ID的计数器
"""
if block_counter is None:
block_counter = [0]
for block in blocks:
if not isinstance(block, dict):
continue
block_type = block.get('type')
# 处理math类型
if block_type == 'math':
latex = block.get('latex', '').strip()
if latex:
block_counter[0] += 1
math_id = f"math-block-{block_counter[0]}"
try:
svg_content = self.math_converter.convert_display_to_svg(latex)
if svg_content:
svg_map[math_id] = svg_content
# 将ID添加到block中以便后续注入时识别
block['mathId'] = math_id
logger.debug(f"公式 {math_id} 转换为SVG成功")
else:
logger.warning(f"公式 {math_id} 转换为SVG失败: {latex[:50]}...")
except Exception as e:
logger.error(f"转换公式 {latex[:50]}... 时出错: {e}")
# 递归处理嵌套的blocks
nested_blocks = block.get('blocks')
if isinstance(nested_blocks, list):
self._extract_and_convert_math_blocks(nested_blocks, svg_map, block_counter)
# 处理列表项
if block_type == 'list':
items = block.get('items', [])
for item in items:
if isinstance(item, list):
self._extract_and_convert_math_blocks(item, svg_map, block_counter)
# 处理表格单元格
if block_type == 'table':
rows = block.get('rows', [])
for row in rows:
cells = row.get('cells', [])
for cell in cells:
cell_blocks = cell.get('blocks', [])
if isinstance(cell_blocks, list):
self._extract_and_convert_math_blocks(cell_blocks, svg_map, block_counter)
# 处理callout内部的blocks
if block_type == 'callout':
callout_blocks = block.get('blocks', [])
if isinstance(callout_blocks, list):
self._extract_and_convert_math_blocks(callout_blocks, svg_map, block_counter)
def _inject_svg_into_html(self, html: str, svg_map: Dict[str, str]) -> str:
"""
将SVG内容直接注入到HTML中不使用JavaScript
参数:
html: 原始HTML内容
svg_map: widgetId到SVG内容的映射
返回:
str: 注入SVG后的HTML
"""
if not svg_map:
return html
import re
# 为每个widgetId查找对应的canvas并替换为SVG
for widget_id, svg_content in svg_map.items():
# 清理SVG内容移除XML声明因为SVG将嵌入HTML
svg_content = re.sub(r'<\?xml[^>]+\?>', '', svg_content)
svg_content = re.sub(r'<!DOCTYPE[^>]+>', '', svg_content)
svg_content = svg_content.strip()
# 创建SVG容器HTML
svg_html = f'<div class="chart-svg-container">{svg_content}</div>'
# 查找包含此widgetId的配置脚本
# 格式: <script type="application/json" id="chart-config-N">{"widgetId":"widget_id",...}</script>
config_pattern = rf'<script[^>]+id="([^"]+)"[^>]*>\s*\{{[^}}]*"widgetId"\s*:\s*"{re.escape(widget_id)}"[^}}]*\}}'
match = re.search(config_pattern, html, re.DOTALL)
if match:
config_id = match.group(1)
# 查找对应的canvas元素
# 格式: <canvas id="chart-N" data-config-id="chart-config-N"></canvas>
canvas_pattern = rf'<canvas[^>]+data-config-id="{re.escape(config_id)}"[^>]*></canvas>'
# 【修复】替换canvas为SVG使用lambda避免反斜杠转义问题
html = re.sub(canvas_pattern, lambda m: svg_html, html)
logger.debug(f"已替换图表 {widget_id} 的canvas为SVG")
else:
logger.warning(f"未找到图表 {widget_id} 对应的配置脚本")
return html
def _inject_math_svg_into_html(self, html: str, svg_map: Dict[str, str]) -> str:
"""
将数学公式SVG内容注入到HTML中
参数:
html: 原始HTML内容
svg_map: 公式ID到SVG内容的映射
返回:
str: 注入SVG后的HTML
"""
if not svg_map:
return html
import re
# 为每个math block查找对应的div并替换为SVG
for math_id, svg_content in svg_map.items():
# 清理SVG内容移除XML声明因为SVG将嵌入HTML
svg_content = re.sub(r'<\?xml[^>]+\?>', '', svg_content)
svg_content = re.sub(r'<!DOCTYPE[^>]+>', '', svg_content)
svg_content = svg_content.strip()
# 创建SVG容器HTML
svg_html = f'<div class="math-svg-container">{svg_content}</div>'
# 查找对应的math-block div
# 格式: <div class="math-block">$$ latex $$</div>
# 我们需要找到包含特定LaTeX内容的div
# 但由于我们在转换时已经给block添加了mathId我们可以用另一种方式
# 方案在HTML渲染器中为math-block添加data-math-id属性
# 但这需要修改HTMLRenderer暂时我们使用更简单的方法
# 按顺序替换所有math-block
# 暂时使用简单的替换方案
# 找到第一个math-block div并替换
math_block_pattern = r'<div class="math-block">\$\$[^$]*\$\$</div>'
# 【修复】转义svg_html中的反斜杠避免re.sub将其解释为转义序列
# 使用re.escape处理替换字符串中的特殊字符
escaped_svg_html = svg_html.replace('\\', r'\\')
html = re.sub(math_block_pattern, lambda m: svg_html, html, count=1)
logger.debug(f"已替换公式 {math_id} 为SVG")
return html
def _get_pdf_html(
self,
document_ir: Dict[str, Any],
optimize_layout: bool = True
) -> str:
"""
生成适用于PDF的HTML内容
- 移除交互式元素(按钮、导航等)
- 添加PDF专用样式
- 嵌入字体文件
- 应用布局优化
- 将图表转换为SVG矢量图形
参数:
document_ir: Document IR数据
optimize_layout: 是否启用布局优化
返回:
str: 优化后的HTML内容
"""
# 如果启用布局优化,先分析文档并生成优化配置
if optimize_layout:
logger.info("启用PDF布局优化...")
layout_config = self.layout_optimizer.optimize_for_document(document_ir)
# 保存优化日志
log_dir = Path('logs/pdf_layouts')
log_dir.mkdir(parents=True, exist_ok=True)
log_file = log_dir / f"layout_{datetime.now().strftime('%Y%m%d_%H%M%S')}.json"
# 保存配置和优化日志
optimization_log = self.layout_optimizer._log_optimization(
self.layout_optimizer._analyze_document(document_ir),
layout_config
)
self.layout_optimizer.config = layout_config
self.layout_optimizer.save_config(log_file, optimization_log)
else:
layout_config = self.layout_optimizer.config
# 关键修复:先预处理图表,确保数据有效
logger.info("预处理图表数据...")
preprocessed_ir = self._preprocess_charts(document_ir)
# 转换图表为SVG使用预处理后的IR
logger.info("开始转换图表为SVG矢量图形...")
svg_map = self._convert_charts_to_svg(preprocessed_ir)
# 转换数学公式为SVG
logger.info("开始转换数学公式为SVG矢量图形...")
math_svg_map = self._convert_math_to_svg(preprocessed_ir)
# 使用HTML渲染器生成基础HTML使用原始IR因为HTMLRenderer会自己修复
# 注意这里仍使用原始document_ir因为HTMLRenderer内部会进行相同的修复
# 这确保了HTML和SVG使用相同的修复逻辑
html = self.html_renderer.render(document_ir)
# 注入图表SVG
if svg_map:
html = self._inject_svg_into_html(html, svg_map)
logger.info(f"已注入 {len(svg_map)} 个SVG图表")
# 注入数学公式SVG
if math_svg_map:
html = self._inject_math_svg_into_html(html, math_svg_map)
logger.info(f"已注入 {len(math_svg_map)} 个SVG公式")
# 获取字体路径并转换为base64用于嵌入
font_path = self._get_font_path()
font_data = font_path.read_bytes()
font_base64 = base64.b64encode(font_data).decode('ascii')
# 判断字体格式
font_format = 'opentype' if font_path.suffix == '.otf' else 'truetype'
# 生成优化后的CSS
optimized_css = self.layout_optimizer.generate_pdf_css()
# 添加PDF专用CSS
pdf_css = f"""
<style>
/* PDF专用字体嵌入 */
@font-face {{
font-family: 'SourceHanSerif';
src: url(data:font/{font_format};base64,{font_base64}) format('{font_format}');
font-weight: normal;
font-style: normal;
}}
/* 强制所有文本使用思源宋体 */
body, h1, h2, h3, h4, h5, h6, p, li, td, th, div, span {{
font-family: 'SourceHanSerif', serif !important;
}}
/* PDF专用样式调整 */
.report-header {{
display: none !important;
}}
.no-print {{
display: none !important;
}}
body {{
background: white !important;
}}
/* SVG图表容器样式 */
.chart-svg-container {{
width: 100%;
height: auto;
display: flex;
justify-content: center;
align-items: center;
}}
.chart-svg-container svg {{
max-width: 100%;
height: auto;
}}
/* 数学公式SVG容器样式 */
.math-svg-container {{
width: 100%;
height: auto;
display: flex;
justify-content: center;
align-items: center;
margin: 20px 0;
}}
.math-svg-container svg {{
max-width: 100%;
height: auto;
}}
/* 隐藏原始的math-block因为已被SVG替换 */
.math-block {{
display: none !important;
}}
/* 隐藏fallback表格因为现在使用SVG */
.chart-fallback {{
display: none !important;
}}
/* 确保chart-container显示用于放置SVG */
.chart-container {{
display: block !important;
min-height: 400px;
}}
{optimized_css}
</style>
"""
# 在</head>前插入PDF专用CSS
html = html.replace('</head>', f'{pdf_css}\n</head>')
return html
def render_to_pdf(
self,
document_ir: Dict[str, Any],
output_path: str | Path,
optimize_layout: bool = True
) -> Path:
"""
将Document IR渲染为PDF文件
参数:
document_ir: Document IR数据
output_path: PDF输出路径
optimize_layout: 是否启用布局优化默认True
返回:
Path: 生成的PDF文件路径
"""
output_path = Path(output_path)
logger.info(f"开始生成PDF: {output_path}")
# 生成HTML内容
html_content = self._get_pdf_html(document_ir, optimize_layout)
# 配置字体
font_config = FontConfiguration()
# 从HTML字符串创建WeasyPrint HTML对象
html_doc = HTML(string=html_content, base_url=str(Path.cwd()))
# 生成PDF
try:
html_doc.write_pdf(
output_path,
font_config=font_config,
presentational_hints=True # 保留HTML的呈现提示
)
logger.info(f"✓ PDF生成成功: {output_path}")
return output_path
except Exception as e:
logger.error(f"PDF生成失败: {e}")
raise
def render_to_bytes(
self,
document_ir: Dict[str, Any],
optimize_layout: bool = True
) -> bytes:
"""
将Document IR渲染为PDF字节流
参数:
document_ir: Document IR数据
optimize_layout: 是否启用布局优化默认True
返回:
bytes: PDF文件的字节内容
"""
html_content = self._get_pdf_html(document_ir, optimize_layout)
font_config = FontConfiguration()
html_doc = HTML(string=html_content, base_url=str(Path.cwd()))
return html_doc.write_pdf(
font_config=font_config,
presentational_hints=True
)
__all__ = ["PDFRenderer"]