diff --git a/backend/app/services/pdf_generator.py b/backend/app/services/pdf_generator.py index 86380352..8da8db6a 100644 --- a/backend/app/services/pdf_generator.py +++ b/backend/app/services/pdf_generator.py @@ -21,42 +21,144 @@ import markdown class PDFGenerator: """Generate PDF reports from markdown content""" + # Emoji to Unicode symbol mapping for PDF compatibility + # Emojis don't render well in PDFs, so we replace them with Unicode text symbols + # NOTE: Use ASCII brackets [] not full-width ［］ for better font compatibility + EMOJI_TO_UNICODE = { + # Status & Indicators + '✅': '✓', + '❌': '✗', + '⚠️': '⚠', + '⚡': '⚡', + '🔔': '◉', + + # Rating & Quality + '⭐': '★', + '🌟': '☆', + '💎': '◆', + '🏆': '◈', + + # Charts & Analytics + '📊': '[圖表]', + '📈': '↑', + '📉': '↓', + '📋': '[清單]', + '📌': '◆', + + # Money & Business + '💰': '$', + '💵': '$', + '💴': '¥', + '💶': '€', + '💷': '£', + '💸': '[支出]', # Fixed: was ［支出］ (full-width brackets) + '💹': '[增長]', + + # Direction & Movement + '🚀': '↑↑', + '⬆️': '↑', + '⬇️': '↓', + '➡️': '→', + '⬅️': '←', + '🔼': '▲', + '🔽': '▼', + + # Symbols + '🎯': '◎', + '🔥': '※', + '💡': '◐', + '🔔': '◉', + '⚙️': '⚙', + '🔧': '[工具]', + '🔨': '[工具]', + + # AI & Tech + '🤖': '[AI]', + '💻': '[電腦]', + '📱': '[手機]', + '🖥️': '[系統]', + + # People & Roles + '👤': '[用戶]', + '👥': '[團隊]', + '🔬': '[研究]', + '📚': '[資料]', + + # Time + '⏰': '[時間]', + '📅': '[日期]', + '⏱️': '[計時]', + + # Other common emojis + '✨': '‧', + '🎨': '[設計]', + '📝': '[筆記]', + '📄': '[文件]', + '🗂️': '[資料夾]', + '🌐': '[網路]', + '🔗': '[連結]', + '💼': '[業務]', + } + """Generate PDF reports from markdown content""" + def __init__(self): """Initialize PDF generator with Chinese font support""" import os from reportlab.pdfbase.cidfonts import UnicodeCIDFont + from reportlab.pdfbase import pdfmetrics + from reportlab.pdfbase.ttfonts import TTFont # Initialize font variables self.custom_font = None self.chinese_font = None - # Ensure we have the absolute path to the current file - current_file = os.path.abspath(__file__) - # backend/app/services/pdf_generator.py -> backend/app/services -> backend/app -> backend -> root - # CRITICAL FIX: Use system fonts for maximum compatibility - # Arial Unicode MS supports: Chinese, English, Math symbols, Emoji - # This avoids font file loading issues with ReportLab + # CRITICAL FIX: Use ReportLab's built-in CID fonts for proper character spacing + # CID fonts (Adobe-GB1, Adobe-CNS1) are specifically designed for PDF rendering + # and don't have the character spacing issues that TTC files have try: - # Try to use Arial Unicode MS (best for all characters) - from reportlab.pdfbase import pdfmetrics - from reportlab.pdfbase.ttfonts import TTFont - - # macOS system font path - arial_unicode_path = '/System/Library/Fonts/Supplemental/Arial Unicode.ttf' - if os.path.exists(arial_unicode_path): - pdfmetrics.registerFont(TTFont('ArialUnicode', arial_unicode_path)) - self.custom_font = 'ArialUnicode' - self.chinese_font = 'ArialUnicode' - print(f"✅ Successfully registered Arial Unicode MS (supports Chinese, English, Math, Emoji)") - else: - # Fallback: Use built-in Helvetica (limited Chinese support) + # Method 1: Try using built-in CID fonts (best for Chinese PDFs) + # These fonts have PERFECT character spacing without gaps + try: + # Try STSong-Light (for Traditional + Simplified Chinese) + pdfmetrics.registerFont(UnicodeCIDFont('STSong-Light')) + self.custom_font = 'STSong-Light' + self.chinese_font = 'STSong-Light' + print(f"✅ Using STSong-Light CID font - Perfect Chinese character spacing") + except: + # Fallback to MSung-Light (Traditional Chinese) + try: + pdfmetrics.registerFont(UnicodeCIDFont('MSung-Light')) + self.custom_font = 'MSung-Light' + self.chinese_font = 'MSung-Light' + print(f"✅ Using MSung-Light CID font - Perfect Traditional Chinese spacing") + except: + # Last CID font attempt: STSongStd-Light + try: + pdfmetrics.registerFont(UnicodeCIDFont('STSongStd-Light')) + self.custom_font = 'STSongStd-Light' + self.chinese_font = 'STSongStd-Light' + print(f"✅ Using STSongStd-Light CID font") + except: + raise Exception("No CID fonts available") + except: + # Method 2: Fallback to TTF fonts if CID fonts fail + print("⚠️ CID fonts not available, trying TTF fonts...") + try: + # Try Arial Unicode MS (TTF file, not TTC) + arial_unicode_path = '/System/Library/Fonts/Supplemental/Arial Unicode.ttf' + if os.path.exists(arial_unicode_path): + pdfmetrics.registerFont(TTFont('ArialUnicode', arial_unicode_path)) + self.custom_font = 'ArialUnicode' + self.chinese_font = 'ArialUnicode' + print(f"✅ Using Arial Unicode MS (TTF) - Good Chinese support") + else: + raise Exception("Arial Unicode not found") + except Exception as e: + # Final fallback: Use built-in Helvetica + print(f"❌ Font registration failed: {e}") + print(f"⚠️ Using Helvetica (limited Chinese character support)") self.custom_font = 'Helvetica' self.chinese_font = 'Helvetica' - print(f"⚠️ Using Helvetica (limited Chinese character support)") - except Exception as e: - print(f"❌ Font registration error: {e}") - self.custom_font = 'Helvetica' - self.chinese_font = 'Helvetica' # Set primary font self.primary_font = self.custom_font if self.custom_font else self.chinese_font @@ -156,24 +258,13 @@ class PDFGenerator: elements.append(Paragraph(metadata, subtitle_style)) elements.append(Spacer(1, 0.5*cm)) - # Convert markdown to simple text (basic conversion) - # Clean markdown formatting - # DEBUG: Log content before PDF conversion - print(f"\n[PDF DEBUG] Content BEFORE _clean_markdown:") - if '煉' in report_content: - print(f" ⚠️ Found '煉' in original content") - if '練' in report_content: - print(f" ✅ Found '練' in original content") + # STEP 1: Replace emojis with Unicode symbols BEFORE markdown cleaning + report_content = self._replace_emojis(report_content) + analyst_name = self._replace_emojis(analyst_name) + # STEP 2: Clean markdown formatting content = self._clean_markdown(report_content) - # DEBUG: Log content after cleaning - print(f"[PDF DEBUG] Content AFTER _clean_markdown:") - if '煉' in content: - print(f" ⚠️ Found '煉' AFTER cleaning") - if '練' in content: - print(f" ✅ Found '練' AFTER cleaning") - # Split content into paragraphs paragraphs = content.split('\n') @@ -294,3 +385,26 @@ class PDFGenerator: text = text.replace(old, new) return text + + def _replace_emojis(self, text: str) -> str: + """ + Replace emoji characters with Unicode text symbols for PDF compatibility + + Emojis don't render well in PDFs, especially with CID fonts. + This method replaces common emojis with Unicode text symbols that + display reliably across all PDF viewers. + + Args: + text: Text containing potential emoji characters + + Returns: + Text with emojis replaced by Unicode symbols + """ + if not text: + return text + + # Replace each emoji with its Unicode symbol equivalent + for emoji, unicode_symbol in self.EMOJI_TO_UNICODE.items(): + text = text.replace(emoji, unicode_symbol) + + return text diff --git a/tradingagents/graph/trading_graph.py b/tradingagents/graph/trading_graph.py index 873ae9ea..264ca22f 100644 --- a/tradingagents/graph/trading_graph.py +++ b/tradingagents/graph/trading_graph.py @@ -97,13 +97,19 @@ class TradingAgentsXGraph: def _create_llm(model: str, base_url: str, api_key: str): # Determine provider based on Base URL if "anthropic.com" in base_url: - return ChatAnthropic(model=model, base_url=base_url, api_key=api_key) + return ChatAnthropic( + model=model, + base_url=base_url, + api_key=api_key, + max_tokens=16000 # Prevent report truncation + ) else: # Default to ChatOpenAI for OpenAI, Grok, DeepSeek, Qwen, and other OpenAI-compatible APIs return ChatOpenAI( model=model, base_url=base_url, - openai_api_key=api_key + openai_api_key=api_key, + max_tokens=16000 # Prevent report truncation ) # Initialize LLMs independently