VNEWS

Running

App Files Files Community

bep40 commited on Jul 22

Commit

e62d653

verified ·

1 Parent(s): 259d6f1

Upload app_v2_entry.py

Browse files

Files changed (1) hide show

app_v2_entry.py +12 -1078

app_v2_entry.py CHANGED Viewed

@@ -1,4 +1,4 @@
-"""VNEWS v2 Entry Point - with fast bongda proxy + rewrite endpoints + multilingual TTS"""
 import sys, os
 from main import app, HEADERS, BONGDA_HEADERS, fetch_bongda_api, HL_LEAGUES
@@ -12,6 +12,13 @@ try:
 except Exception as e:
     print(f"[WARN] ai_patch import failed: {e}")
 from fastapi.responses import HTMLResponse, JSONResponse, FileResponse, Response
 from fastapi.staticfiles import StaticFiles
 from starlette.routing import Mount
@@ -26,17 +33,15 @@ import asyncio
 HL_LEAGUES['friendly'] = {"path": "giai-khac/friendly", "name": "Giao hữu", "emoji": "🤝"}
 STATIC_DIR = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'static')
-SPACE = "https://bep40-vnews.hf.space"  # SEO URL base for share links
 app.router.routes=[r for r in app.router.routes if not(getattr(r,'path',None)=='/' and hasattr(r,'methods') and 'GET' in getattr(r,'methods',set()))]
 app.routes[:]=[r for r in app.routes if not isinstance(r, Mount)]
 app.router.routes=[r for r in app.router.routes if not isinstance(r, Mount)]
 def _clean(s): return re.sub(r"\s+"," ",html_lib.unescape(str(s or""))).strip()
-# Cache for match details (5 min TTL)
 _match_cache = {}
-# === FAST BONGDA PROXY ENDPOINT ===
 def _get_match_detail(event_id, slug=None):
     headers = {"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36", "Accept": "text/html", "Referer": "https://bongda.com.vn/"}
     if slug:
@@ -508,18 +513,13 @@ def _st():return JSONResponse({'persistent':os.path.isdir('/data') and os.access
 # ===== SHARE HELPERS: render content pages for shared links =====
 def _render_slides_page(post, safe_title, safe_img, safe_url):
     slides = post.get('slides', [])
-    # Get image from post.img or first slide's image
     if not safe_img and slides and slides[0].get('image'):
         safe_img = slides[0].get('image', '')
-    # Use text for description if available
     description = _clean((post.get('text') or '')[:200]) or "Tin tức tóm tắt, AI rewrite, World Cup 2026"
-    # Build canonical URL preserving original query format if url was provided
     if safe_url and safe_url != '/':
         canonical_url = f"{SPACE}/s?url={quote(safe_url)}&title={quote(safe_title[:100])}"
     else:
         canonical_url = f"{SPACE}/s?post_id={post.get('id') or ''}"
     h = f'''<!DOCTYPE html>
 <html lang="vi">
 <head>
@@ -552,15 +552,11 @@ def _render_slides_page(post, safe_title, safe_img, safe_url):
 def _render_video_page(post, safe_title, safe_img, safe_url):
     video_url = post.get('video', '')
-    # Use text for description if available
     description = _clean((post.get('text') or '')[:200]) or "Tin tức tóm tắt, AI rewrite, World Cup 2026"
-    # Build canonical URL preserving original query format if url was provided
     if safe_url and safe_url != '/':
         canonical_url = f"{SPACE}/s?url={quote(safe_url)}&title={quote(safe_title[:100])}"
     else:
         canonical_url = f"{SPACE}/s?post_id={post.get('id') or ''}"
     h = f'''<!DOCTYPE html>
 <html lang="vi">
 <head>
@@ -592,17 +588,12 @@ video{{width:100%;height:100%;max-height:100vh;object-fit:contain;background:#00
 @app.get('/s/{slug}')
 async def _sh_slug(slug: str, request: Request, url: str = '', title: str = '', img: str = ''):
-    """SEO-friendly share endpoint with slug in URL path.
-    Shows slide content when slug matches a wall post ID, otherwise redirects.
-    """
     safe_title = _clean(title) if title else 'VNEWS - Tin tức'
     safe_img = _clean(img) if img else ''
     safe_url = _clean(url) if url else '/'
-    # Try to find post by slug (post ID)
     post = None
     try:
-        if slug and len(slug) > 5:  # Likely a post ID
             posts = _load_wall_posts()
             for p in posts:
                 if p.get('id') == slug:
@@ -613,14 +604,10 @@ async def _sh_slug(slug: str, request: Request, url: str = '', title: str = '',
                     break
     except:
         pass
     if post and post.get('slides'):
         return _render_slides_page(post, safe_title, safe_img, safe_url)
     if post and post.get('video'):
         return _render_video_page(post, safe_title, safe_img, safe_url)
-    # Otherwise redirect
     return HTMLResponse(f'''<!DOCTYPE html>
 <html lang="vi">
 <head>
@@ -640,8 +627,6 @@ async def _sh(url:str='',title:str='',img:str='',post_id:str=''):
     safe_title = _clean(title) if title else 'VNEWS - Tin tức'
     safe_img = _clean(img) if img else ''
     safe_url = _clean(url) if url else '/'
-    # Try to find wall post by post_id or URL (prioritize posts with slides/video)
     post = None
     try:
         posts = _load_wall_posts()
@@ -654,7 +639,6 @@ async def _sh(url:str='',title:str='',img:str='',post_id:str=''):
                     safe_url = p.get('url', safe_url) or safe_url
                     break
         elif url:
-            # Find matching URL - prioritize posts with slides or video
             for p in posts:
                 if p.get('url') == url and p.get('slides'):
                     post = p
@@ -663,7 +647,6 @@ async def _sh(url:str='',title:str='',img:str='',post_id:str=''):
                     safe_url = p.get('url', safe_url) or safe_url
                     break
             if not post:
-                # Fallback: find any matching URL
                 for p in posts:
                     if p.get('url') == url:
                         post = p
@@ -673,14 +656,10 @@ async def _sh(url:str='',title:str='',img:str='',post_id:str=''):
                         break
     except:
         pass
     if post and post.get('slides'):
         return _render_slides_page(post, safe_title, safe_img, safe_url)
     if post and post.get('video'):
         return _render_video_page(post, safe_title, safe_img, safe_url)
-    # Fallback: redirect to original URL
     return HTMLResponse(f'''<!DOCTYPE html>
 <html lang="vi">
 <head>
@@ -966,7 +945,6 @@ from urllib.parse import quote as _quote2
 _UA_RW = {'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36', 'Accept-Language': 'vi-VN,vi;q=0.9'}
-# Unique character markers for language detection
 _UNIQUE_CHARS = {
     'vietnamese': set('đăâêôơưàảãạáằẳẵặắầẩẫậấèẻẽẹéềễểệếìỉĩịíòỏõọóồổỗộốờởỡợớùủũụúừửữựứỳỷỹỵý'),
     'spanish': set('ñáéíóúü¿¡'),
@@ -981,1075 +959,31 @@ _STOPWORDS = {
 }
 def detect_language(text):
-    """Detect language from text content using stopword + character analysis."""
     if not text:
         return 'vietnamese'
     text_lower = text.lower()
     text_chars = set(text_lower)
-    # Strong signal: Vietnamese unique characters
     vn_chars = len(text_chars & _UNIQUE_CHARS['vietnamese'])
     if vn_chars >= 2:
         return 'vietnamese'
-    # Spanish unique chars (ñ, ¿, ¡)
     es_chars = len(text_chars & _UNIQUE_CHARS['spanish'])
     pt_chars = len(text_chars & _UNIQUE_CHARS['portuguese'])
-    # Stopword scoring
     words = set(re.findall(r'\b\w+\b', text_lower))
     scores = {}
     for lang, stops in _STOPWORDS.items():
         scores[lang] = len(words & stops) / max(len(stops), 1)
-    # Disambiguate Portuguese vs Spanish
     pt_markers = {'não', 'pelo', 'pela', 'isso', 'há', 'estão', 'num', 'numa', 'tenho', 'posso', 'você', 'nós', 'eles', 'elas', 'também', 'muito', 'já', 'só', 'até', 'entre', 'depois', 'sem', 'mesmo', 'aos', 'serão'}
     es_markers = {'pero', 'está', 'están', 'porque', 'también', 'hasta', 'donde', 'quien', 'fue', 'son', 'fueron', 'ese', 'eso', 'ante', 'ellos', 'ella', 'nosotros', 'usted', 'ustedes', 'tú', 'él', 'desde', 'todo', 'durante', 'todos', 'uno', 'les', 'ni', 'contra', 'otros', 'fueron'}
     pt_overlap = len(words & pt_markers)
     es_overlap = len(words & es_markers)
     if scores.get('portuguese', 0) > 0 and pt_overlap > es_overlap:
         return 'portuguese'
     if scores.get('spanish', 0) > 0 and es_overlap > pt_overlap:
         return 'spanish'
     if scores.get('english', 0) > 0.15:
         return 'english'
     best = max(scores, key=scores.get)
     return best if scores[best] > 0.05 else 'vietnamese'
-# Emotion keyword-based detection
-_EMOTION_KEYWORDS = {
-    'happy': {
-        'en': ['happy', 'joy', 'wonderful', 'great', 'amazing', 'fantastic', 'love', 'excellent', 'beautiful', 'glad', 'delighted', 'pleased', 'cheerful', 'celebrate', 'victory', 'win', 'success'],
-        'pt': ['feliz', 'alegria', 'maravilhoso', 'ótimo', 'incrível', 'fantástico', 'amor', 'excelente', 'lindo', 'contente', 'encantado', 'vitória', 'sucesso'],
-        'es': ['feliz', 'alegria', 'maravilloso', 'genial', 'increíble', 'fantástico', 'amor', 'excelente', 'hermoso', 'contento', 'encantado', 'victoria', 'éxito'],
-        'vi': ['vui', 'hạnh phúc', 'tuyệt vời', 'tuyệt', 'ý nghĩa', 'đẹp', 'thích', 'yêu', 'vui vẻ', 'hân hoan', 'phấn khích', 'chiến thắng', 'thành công'],
-    },
-    'sad': {
-        'en': ['sad', 'unhappy', 'terrible', 'awful', 'horrible', 'miserable', 'depressed', 'grief', 'sorrow', 'tragic', 'unfortunate', 'painful', 'death', 'die', 'kill'],
-        'pt': ['triste', 'infeliz', 'terrível', 'horrível', 'miserável', 'deprimido', 'dor', 'trágico', 'infelizmente', 'penoso', 'morte', 'morrer'],
-        'es': ['triste', 'infeliz', 'terrible', 'horrible', 'miserable', 'deprimido', 'dolor', 'trágico', 'desafortunado', 'penoso', 'muerte', 'morir'],
-        'vi': ['buồn', 'không vui', 'tồi tệ', 'kinh khủng', 'đau khổ', 'đau buồn', 'bi thương', 'khốn nạn', 'đau đớn', 'thảm họa', 'chết', 'mất'],
-    },
-    'excited': {
-        'en': ['excited', 'thrilling', 'amazing', 'wow', 'incredible', 'unbelievable', 'awesome', 'exhilarating', 'electrifying', 'breathtaking', 'breakthrough', 'record'],
-        'pt': ['animado', 'emocionante', 'incrível', 'impressionante', 'sensacional', 'eletrizante', 'empolgante', 'recorde'],
-        'es': ['emocionante', 'increíble', 'impresionante', 'sensacional', 'electrizante', 'emocionado', 'entusiasmado', 'récord'],
-        'vi': ['hào hứng', 'phấn khích', 'thú vị', 'tuyệt cú mèo', 'đỉnh cao', 'ngoạn mục', 'sục sôi', 'kỷ lục', 'đột phá'],
-    },
-    'humorous': {
-        'en': ['funny', 'hilarious', 'joke', 'laugh', 'comedy', 'humor', 'amusing', 'witty', 'sarcastic', 'ironic', 'ridiculous', 'absurd', 'lol', 'haha'],
-        'pt': ['engraçado', 'hilário', 'piada', 'rir', 'comédia', 'humor', 'divertido', 'irônico', 'ridículo', 'absurdo', 'kkk'],
-        'es': ['gracioso', 'hilarante', 'broma', 'risa', 'comedia', 'humor', 'divertido', 'irónico', 'ridículo', 'absurdo', 'jaja'],
-        'vi': ['hài hước', 'buồn cười', 'đùa', 'cười', 'hài', 'vui nhộn', 'hóm hỉnh', 'mỉa mai', 'lố bịch', 'vô lý', 'haha'],
-    },
-    'serious': {
-        'en': ['serious', 'critical', 'important', 'urgent', 'severe', 'grave', 'significant', 'crucial', 'vital', 'essential', 'alarming', 'concerning', 'crisis', 'war', 'conflict'],
-        'pt': ['sério', 'crítico', 'importante', 'urgente', 'grave', 'significativo', 'crucial', 'vital', 'essencial', 'preocupante', 'crise', 'guerra', 'conflito'],
-        'es': ['serio', 'crítico', 'importante', 'urgente', 'grave', 'significativo', 'crucial', 'vital', 'esencial', 'preocupante', 'crisis', 'guerra', 'conflicto'],
-        'vi': ['nghiêm trọng', 'quan trọng', 'khẩn cấp', 'nghiêm túc', 'đáng kể', 'thiết yếu', 'cần thiết', 'báo động', 'lo ngại', 'khủng hoảng', 'chiến tranh', 'xung đột'],
-    },
-}
-def detect_emotion(text, language='vietnamese'):
-    """Detect emotion from text using keyword matching."""
-    if not text:
-        return 'neutral'
-    text_lower = text.lower()
-    scores = {}
-    for emotion, lang_keywords in _EMOTION_KEYWORDS.items():
-        keywords = lang_keywords.get(language, lang_keywords.get('en', []))
-        score = sum(1 for kw in keywords if kw in text_lower)
-        scores[emotion] = score
-    if max(scores.values()) == 0:
-        return 'neutral'
-    return max(scores, key=scores.get)
-def detect_language_and_emotion(title, text):
-    """Detect both language and emotion from article content."""
-    combined = f"{title} {text}"
-    lang = detect_language(combined)
-    emotion = detect_emotion(combined, lang)
-    return lang, emotion
-# Voice selection based on language and emotion (using MultilingualNeural voices)
-VOICE_BY_LANG_EMOTION = {
-    'vietnamese': {
-        'happy': ('vi-VN-HoaiMyNeural', 'vui'),
-        'sad': ('vi-VN-NamMinhNeural', 'buồn'),
-        'excited': ('vi-VN-HoaiMyNeural', 'hào hứng'),
-        'humorous': ('vi-VN-HoaiMyNeural', 'vui'),
-        'serious': ('vi-VN-NamMinhNeural', 'nghiêm túc'),
-        'neutral': ('vi-VN-HoaiMyNeural', 'trung_tinh'),
-    },
-    'portuguese': {
-        'happy': ('pt-BR-ThalitaMultilingualNeural', 'feliz'),
-        'sad': ('pt-BR-ThalitaMultilingualNeural', 'triste'),
-        'excited': ('pt-BR-ThalitaMultilingualNeural', 'animado'),
-        'humorous': ('pt-BR-ThalitaMultilingualNeural', 'engraçado'),
-        'serious': ('pt-BR-ThalitaMultilingualNeural', 'sério'),
-        'neutral': ('pt-BR-ThalitaMultilingualNeural', 'neutro'),
-    },
-    'english': {
-        'happy': ('en-US-AndrewMultilingualNeural', 'happy'),
-        'sad': ('en-AU-WilliamMultilingualNeural', 'sad'),
-        'excited': ('en-US-AndrewMultilingualNeural', 'excited'),
-        'humorous': ('en-US-AndrewMultilingualNeural', 'funny'),
-        'serious': ('en-AU-WilliamMultilingualNeural', 'serious'),
-        'neutral': ('en-US-AndrewMultilingualNeural', 'neutral'),
-    },
-    'french': {
-        'happy': ('fr-FR-VivienneMultilingualNeural', 'heureux'),
-        'sad': ('fr-FR-RemyMultilingualNeural', 'triste'),
-        'excited': ('fr-FR-VivienneMultilingualNeural', 'excité'),
-        'humorous': ('fr-FR-VivienneMultilingualNeural', 'drôle'),
-        'serious': ('fr-FR-RemyMultilingualNeural', 'sérieux'),
-        'neutral': ('fr-FR-VivienneMultilingualNeural', 'neutre'),
-    },
-    'german': {
-        'happy': ('de-DE-SeraphinaMultilingualNeural', 'glücklich'),
-        'sad': ('de-DE-FlorianMultilingualNeural', 'traurig'),
-        'excited': ('de-DE-SeraphinaMultilingualNeural', 'aufgeregt'),
-        'humorous': ('de-DE-SeraphinaMultilingualNeural', 'lustig'),
-        'serious': ('de-DE-FlorianMultilingualNeural', 'ernst'),
-        'neutral': ('de-DE-SeraphinaMultilingualNeural', 'neutral'),
-    },
-    'korean': {
-        'happy': ('ko-KR-HyunsuMultilingualNeural', '행복'),
-        'sad': ('ko-KR-HyunsuMultilingualNeural', '슬픔'),
-        'excited': ('ko-KR-HyunsuMultilingualNeural', '흥분'),
-        'humorous': ('ko-KR-HyunsuMultilingualNeural', '유쾌'),
-        'serious': ('ko-KR-HyunsuMultilingualNeural', '진지'),
-        'neutral': ('ko-KR-HyunsuMultilingualNeural', '중립'),
-    },
-    'italian': {
-        'happy': ('it-IT-GiuseppeMultilingualNeural', 'felice'),
-        'sad': ('it-IT-GiuseppeMultilingualNeural', 'triste'),
-        'excited': ('it-IT-GiuseppeMultilingualNeural', 'emozionato'),
-        'humorous': ('it-IT-GiuseppeMultilingualNeural', 'divertente'),
-        'serious': ('it-IT-GiuseppeMultilingualNeural', 'serio'),
-        'neutral': ('it-IT-GiuseppeMultilingualNeural', 'neutro'),
-    },
-}
-# All valid voice IDs (new MultilingualNeural format)
-VALID_VOICES = {
-    'vi-VN-HoaiMyNeural', 'vi-VN-NamMinhNeural',
-    'en-US-AndrewMultilingualNeural', 'en-AU-WilliamMultilingualNeural',
-    'pt-BR-ThalitaMultilingualNeural',
-    'fr-FR-VivienneMultilingualNeural', 'fr-FR-RemyMultilingualNeural',
-    'de-DE-SeraphinaMultilingualNeural', 'de-DE-FlorianMultilingualNeural',
-    'ko-KR-HyunsuMultilingualNeural',
-    'it-IT-GiuseppeMultilingualNeural',
-}
-def get_voice_for_content(title, text, preferred_voice=None):
-    """Get appropriate voice based on content language and emotion."""
-    # Accept the new MultilingualNeural voices directly
-    if preferred_voice and preferred_voice in VALID_VOICES:
-        return preferred_voice
-    # Also accept old shorthand voice IDs and map them to new format
-    old_voice_map = {
-        'hoaimy': 'vi-VN-HoaiMyNeural',
-        'namminh': 'vi-VN-NamMinhNeural',
-        'andrew': 'en-US-AndrewMultilingualNeural',
-        'jenny': 'en-US-AndrewMultilingualNeural',
-        'thalita': 'pt-BR-ThalitaMultilingualNeural',
-        'pt_thalita': 'pt-BR-ThalitaMultilingualNeural',
-        'pt_francisco': 'pt-BR-ThalitaMultilingualNeural',
-        'ela': 'en-US-AndrewMultilingualNeural',
-        'es_carlos': 'en-US-AndrewMultilingualNeural',
-        'denise': 'fr-FR-VivienneMultilingualNeural',
-        'katja': 'de-DE-SeraphinaMultilingualNeural',
-        'nanami': 'en-US-AndrewMultilingualNeural',
-        'sunhee': 'ko-KR-HyunsuMultilingualNeural',
-        'xiaochen': 'en-US-AndrewMultilingualNeural',
-    }
-    if preferred_voice and preferred_voice in old_voice_map:
-        return old_voice_map[preferred_voice]
-    lang, emotion = detect_language_and_emotion(title, text)
-    lang_map = VOICE_BY_LANG_EMOTION.get(lang, VOICE_BY_LANG_EMOTION['vietnamese'])
-    voice, _ = lang_map.get(emotion, lang_map['neutral'])
-    return voice
-def _is_relevant_image(img_url, title, text):
-    """Check if an image is relevant to the article content."""
-    if not img_url:
-        return False
-    skip_patterns = ['pixel', 'analytics', 'tracking', '1x1.gif', 'spacer.gif',
-                     'logo', 'icon', 'avatar', 'emoji', 'smiley', 'sprite',
-                     'advertisement', 'ad-banner', 'sponsored', 'banner-ads']
-    img_lower = img_url.lower()
-    for p in skip_patterns:
-        if p in img_lower:
-            return False
-    if not any(img_lower.endswith(ext) for ext in ['.jpg', '.jpeg', '.png', '.webp', '.gif']):
-        return False
-    return True
-def _filter_relevant_images(images, title, text, max_images=8):
-    """Filter and rank images by relevance to article content."""
-    if not images:
-        return []
-    seen = set()
-    relevant = []
-    for img in images:
-        if img in seen:
-            continue
-        seen.add(img)
-        if _is_relevant_image(img, title, text):
-            relevant.append(img)
-    return relevant[:max_images]
-def _scrape_article_for_rewrite(url):
-    """Scrape article: extract title, paragraphs, RELEVANT images, OG image."""
-    try:
-        r = req.get(url, headers=_UA_RW, timeout=15, allow_redirects=True)
-        r.encoding = 'utf-8'
-        soup = BeautifulSoup(r.text, 'lxml')
-        for tag in soup.find_all(['script', 'style', 'nav', 'footer', 'aside', 'form']):
-            tag.decompose()
-        h1 = soup.find('h1')
-        ogt = soup.find('meta', property='og:title')
-        title = (h1.get_text(strip=True) if h1 else '') or (ogt.get('content', '') if ogt else '')
-        ogi = soup.find('meta', property='og:image')
-        og_img = ogi.get('content', '') if ogi else ''
-        if og_img and og_img.startswith('//'):
-            og_img = 'https:' + og_img
-        block = None
-        for sel in ['article', '.singular-content', '.detail-content', '.fck_detail', '.content-detail', '.knc-content', 'main', '.cms-body', '.article__body']:
-            el = soup.select_one(sel)
-            if el and len(el.find_all('p')) >= 2:
-                block = el
-                break
-        if not block:
-            block = soup.body or soup
-        paragraphs = []
-        all_images = []
-        seen_imgs = set()
-        if og_img and og_img not in seen_imgs:
-            all_images.append(og_img)
-            seen_imgs.add(og_img)
-        for el in block.find_all(['p', 'h2', 'h3', 'figure', 'img'], recursive=True):
-            if el.name == 'p':
-                t = _clean(el.get_text(strip=True))
-                if t and len(t) > 40:
-                    paragraphs.append(t)
-            elif el.name in ('figure', 'img'):
-                im = el if el.name == 'img' else el.find('img')
-                if im:
-                    src = im.get('data-src') or im.get('src') or im.get('data-original') or ''
-                    if src and 'base64' not in src:
-                        if src.startswith('//'):
-                            src = 'https:' + src
-                        if src not in seen_imgs:
-                            all_images.append(src)
-                            seen_imgs.add(src)
-        # Filter to relevant images only
-        relevant_images = _filter_relevant_images(all_images, title, ' '.join(paragraphs[:5]))
-        return {'title': _clean(title), 'paragraphs': paragraphs, 'images': relevant_images, 'og_img': og_img}
-    except Exception:
-        return None
-def _extract_key_points_rw(paragraphs, max_points=5):
-    r"""Extract key points from paragraphs - extracts ALL sentences, not just first one.
-    Fixes: Original regex `^(.+?[.!?])\s` only captured first sentence per paragraph.
-    Now splits on all sentence boundaries and takes valid sentences until max_points.
-    """
-    points = []
-    for p in paragraphs:
-        if len(points) >= max_points:
-            break
-        p = _clean(p)
-        if not p:
-            continue
-        # Split paragraph into sentences using Vietnamese + English punctuation
-        sentences = re.split(r'(?<=[.!?])\s+(?=[A-ZÀ-Ỹ0-9])', p)
-        sentences = [s.strip() for s in sentences if s.strip()]
-        for sentence in sentences:
-            if len(points) >= max_points:
-                break
-            # Clean sentence - remove extra whitespace
-            sentence = _clean(sentence)
-            if len(sentence) < 30:
-                continue
-            # Check for duplicates
-            if any(sentence[:60] in existing for existing in points):
-                continue
-            # Ensure sentence ends with punctuation
-            if not sentence.endswith(('.', '!', '?')):
-                sentence = sentence + '.'
-            points.append(sentence)
-    # If no valid sentences found, take chunks from raw text
-    if not points:
-        raw = '\n'.join(paragraphs)
-        for i in range(0, min(len(raw), max_points * 300), 280):
-            chunk = _clean(raw[i:i+280])
-            if len(chunk) >= 30 and chunk not in points:
-                points.append(chunk + ('.' if not chunk.endswith('.') else ''))
-            if len(points) >= max_points:
-                break
-    return points
-@app.post("/api/rewrite_slide")
-async def api_rewrite_slide(request: Request):
-    """Fast rewrite as SLIDES - no AI needed, instant response."""
-    body = await request.json()
-    url = _clean(body.get("url", ""))
-    context = body.get("context", "")
-    preferred_voice = body.get("voice", "")  # Accept custom voice selection
-    if not url and not context:
-        return JSONResponse({"error": "Cần URL hoặc nội dung"}, status_code=400)
-    data = None
-    if url and url.startswith("http"):
-        data = _scrape_article_for_rewrite(url)
-    if not data and context:
-        paragraphs = [_clean(p) for p in context.split('\n') if len(_clean(p)) > 40]
-        data = {'title': paragraphs[0][:80] if paragraphs else 'Bài viết', 'paragraphs': paragraphs, 'images': [], 'og_img': ''}
-    if not data or not data.get('paragraphs'):
-        return JSONResponse({"error": "Không đọc được bài viết"}, status_code=422)
-    points = _extract_key_points_rw(data['paragraphs'], max_points=12)
-    if not points:
-        return JSONResponse({"error": "Không tìm được ý chính"}, status_code=422)
-    images = data.get('images', [])
-    slides = []
-    for i, point in enumerate(points):
-        img = images[i] if i < len(images) else (images[-1] if images else '')
-        if img and 'cdnphoto.dantri' in img:
-            img = '/api/proxy/img?url=' + _quote2(img, safe='')
-        slides.append({'text': point, 'image': img, 'index': i + 1})
-    summary_text = '\n\n'.join([f"• {s['text']}" for s in slides])
-    # Auto-detect language and emotion
-    lang, emotion = detect_language_and_emotion(data['title'], summary_text)
-    # Use preferred voice if provided, otherwise auto-detect
-    voice = preferred_voice if preferred_voice else get_voice_for_content(data['title'], summary_text)
-    post = {
-        "id": str(int(time.time() * 1000)) + str(_random2.randint(100, 999)),
-        "title": data['title'],
-        "text": summary_text,
-        "img": images[0] if images else '',
-        "url": url,
-        "kind": "slide_summary",
-        "slides": slides,
-        "images": images[:10],
-        "video": "",
-        "voice": voice,
-        "emotion": emotion,
-        "language": lang,
-        "ts": int(time.time())
-    }
-    posts = _load_wall_posts()
-    posts.insert(0, post)
-    _save_wall_posts(posts)
-    return JSONResponse({"post": post, "slides": slides})
-@app.post("/api/rewrite_share")
-async def api_rewrite_share(request: Request):
-    """Rewrite article and post to Tường AI with SLIDES + AI text."""
-    body = await request.json()
-    url = _clean(body.get("url", ""))
-    ctx = _clean(body.get("context", ""))
-    preferred_voice = body.get("voice", "")  # Accept custom voice selection
-    if not url and not ctx:
-        return JSONResponse({"error": "Cần URL hoặc nội dung"}, status_code=400)
-    data = None
-    if url and url.startswith("http"):
-        data = _scrape_article_for_rewrite(url)
-    if not data and ctx:
-        paragraphs = [_clean(p) for p in ctx.split('\n') if len(_clean(p)) > 40]
-        data = {'title': paragraphs[0][:80] if paragraphs else 'Bài viết', 'paragraphs': paragraphs, 'images': [], 'og_img': ''}
-    if not data or not data.get('paragraphs'):
-        return JSONResponse({"error": "Không đọc được bài viết"}, status_code=422)
-    raw_text = '\n'.join(data['paragraphs'])
-    if len(raw_text) < 50:
-        raw_text = ctx[:14000]
-    if len(raw_text) < 50:
-        return JSONResponse({"error": "Bài viết quá ngắn"}, status_code=422)
-    domain = ''
-    try:
-        from urllib.parse import urlparse
-        domain = urlparse(url).netloc.replace('www.', '')
-    except:
-        pass
-    # Generate AI summary text
-    ai_text = None
-    try:
-        import ai_ext
-        if hasattr(ai_ext, 'qwen_generate'):
-            prompt = f'Tóm tắt đăng Tường AI:\nTiêu đề: {data["title"]}\n{raw_text[:14000]}\n\n4-6 ý chính. Cuối ghi nguồn.'
-            ai_text = await ai_ext.qwen_generate(prompt, max_tokens=1000)
-    except Exception:
-        pass
-    if not ai_text or len(ai_text) < 80:
-        key_pts = _extract_key_points_rw(data['paragraphs'], max_points=12)
-        if key_pts:
-            ai_text = '\n\n'.join([f"• {p}" for p in key_pts])
-        else:
-            ai_text = f"Tóm tắt: {data['title']}\n\n{raw_text[:1200]}\n\nNguồn: {domain}"
-    # Build slides from key points (FIX: include slides in rewrite_share too!)
-    points = _extract_key_points_rw(data['paragraphs'], max_points=12)
-    images = data.get('images', [])
-    slides = []
-    for i, point in enumerate(points):
-        img = images[i] if i < len(images) else (images[-1] if images else '')
-        if img and 'cdnphoto.dantri' in img:
-            img = '/api/proxy/img?url=' + _quote2(img, safe='')
-        slides.append({'text': point, 'image': img, 'index': i + 1})
-    # Auto-detect language and emotion
-    lang, emotion = detect_language_and_emotion(data['title'], ai_text)
-    # Use preferred voice if provided, otherwise auto-detect
-    voice = preferred_voice if preferred_voice else get_voice_for_content(data['title'], ai_text)
-    post = {
-        "id": str(int(time.time() * 1000)) + str(_random2.randint(100, 999)),
-        "title": data['title'],
-        "text": ai_text,
-        "img": images[0] if images else '',
-        "url": url,
-        "kind": "rewrite",
-        "slides": slides,
-        "images": images[:10],
-        "video": "",
-        "voice": voice,
-        "emotion": emotion,
-        "language": lang,
-        "ts": int(time.time())
-    }
-    posts = _load_wall_posts()
-    posts.insert(0, post)
-    _save_wall_posts(posts)
-    return JSONResponse({"post": post, "slides": slides})
-@app.post("/api/url_wall")
-async def api_url_wall(request: Request):
-    """Submit URL to add to Tường AI."""
-    body = await request.json()
-    url = _clean(body.get("url", ""))
-    if not url or not url.startswith('http'):
-        return JSONResponse({"error": "URL không hợp lệ"}, status_code=400)
-    # Reuse rewrite_share logic
-    req._body = json.dumps({"url": url}).encode()
-    return await api_rewrite_share(request)
-# ===== PERSONAL OPINION POST v3: AI VIET BAI MOI, dao nau quan diem + nguon tin =====
-async def _scrape_source_for_opinion(url):
-    """Scrape article content for opinion synthesis."""
-    try:
-        art = _scrape_article_for_rewrite(url)
-        if art:
-            paras = art.get("paragraphs", [])[:6]
-            imgs = art.get("images", [])[:2]
-            proxied_imgs = []
-            for img in imgs[:2]:
-                if any(x in img for x in ["cdnphoto.dantri", "vnexpress", "vcdn", "refooty"]):
-                    img = "/api/proxy/img?url=" + _quote2(img, safe="")
-                proxied_imgs.append(img)
-            return {
-                "title": art.get("title", ""),
-                "paragraphs": paras,
-                "images": proxied_imgs,
-                "og_image": art.get("og_img", ""),
-                "via": ""
-            }
-    except:
-        pass
-    return None
-def _split_into_slides(text, source_images, max_slides=6):
-    """Split article text into slides with relevant images."""
-    import re as _re
-    paragraphs = [p.strip() for p in text.split("\n") if p.strip() and len(p.strip()) > 50]
-    slides = []
-    img_idx = 0
-    for p in paragraphs:
-        if len(slides) >= max_slides:
-            break
-        p = _re.sub(r"^#+\s*", "", p)
-        if p.startswith("---") or (p.startswith("*") and p.endswith("*")):
-            continue
-        img = source_images[img_idx] if img_idx < len(source_images) else ""
-        if img:
-            img_idx += 1
-        slides.append({"text": p, "image": img, "index": len(slides) + 1})
-    if not slides:
-        slides.append({"text": "Dang tao bai viet...", "image": "", "index": 1})
-    return slides
-def _build_fallback_article(opinion, source_details, title):
-    """Build proper synthesized article when AI is unavailable."""
-    import random
-    lines = []
-    lines.append(title)
-    lines.append("")
-    # Intro - paraphrase the opinion
-    lines.append("Trong boi canh hien nay, van de '" + opinion[:80] + "' dang nhan duoc nhieu su quan tam. Bai viet duoi day se phan tich goc nhin ca nhan dua tren cac nguon tin bao chi.")
-    lines.append("")
-    # Body - synthesize from sources with personal take
-    for sd in source_details[:4]:
-        paras = sd.get("paragraphs", [])
-        via = sd.get("via", "Nguon tin")
-        sd_title = sd.get("title", "")
-        if paras:
-            key_point = paras[0][:200]
-            lines.append("Theo " + via + ", " + key_point + ". Dieu nay cho thay " + opinion[:40] + " la mot goc nhin dang duoc ban luan.")
-            lines.append("")
-    # Conclusion
-    lines.append("Tom lai, " + opinion[:120] + ". Day la quan diem ca nhan duoc tong hop tu nhieu nguon tin khac nhau.")
-    lines.append("")
-    lines.append("---")
-    vias = [sd.get("via","") for sd in source_details[:4] if sd.get("via")]
-    if vias:
-        lines.append("Nguon tham khao: " + ", ".join(vias) + ".")
-    return "\n".join(lines)
-@app.post("/api/personal_post/preview")
-async def api_personal_post_preview(request: Request):
-    """Preview: AI writes NEW article from opinion + HOT sources."""
-    body = await request.json()
-    opinion = _clean(body.get("opinion", ""))
-    selected_topics = body.get("selected_topics", []) or []
-    selected_sources = body.get("selected_sources", []) or []
-    if not opinion or len(opinion) < 10:
-        return JSONResponse({"error": "Quan diem ca nhan qua ngan"}, status_code=400)
-    if not selected_topics:
-        hot = _get_hot_topics()
-        selected_topics = [t.get("topic", "") for t in hot[:3] if t.get("topic")]
-    all_sources = []
-    seen_urls = set()
-    for topic in selected_topics[:3]:
-        for s in _search_all(topic, limit=5):
-            if s.get("url") and s["url"] not in seen_urls:
-                seen_urls.add(s["url"])
-                all_sources.append(s)
-                if len(all_sources) >= 5:
-                    break
-        if len(all_sources) >= 5:
-            break
-    for src in selected_sources:
-        if src.get("url") and src["url"] not in seen_urls:
-            all_sources.insert(0, src)
-    source_details = []
-    source_images = []
-    for src in all_sources[:5]:
-        sd = await _scrape_source_for_opinion(src.get("url", ""))
-        if sd:
-            sd["via"] = src.get("via", "")
-            source_details.append(sd)
-            for img in sd.get("images", []):
-                if img not in source_images:
-                    source_images.append(img)
-    # Title
-    words = re.findall(r"[A-Za-z0-9\u00c0-\u1ef9]+", opinion)
-    tw = words[:6] if len(words) >= 6 else words[:3]
-    title = " ".join(w[0].upper() + w[1:] for w in tw) if tw else "Quan diem ca nhan"
-    title = title[:80]
-    # AI generate
-    ai_success = False
-    ai_text = None
-    try:
-        import ai_ext
-        if hasattr(ai_ext, 'qwen_generate'):
-            src_lines = []
-            for i, sd in enumerate(source_details[:5]):
-                src_lines.append("[" + sd.get("via","Nguon") + "] " + sd.get("title",""))
-                for p in sd.get("paragraphs", [])[:3]:
-                    src_lines.append("  -> " + p[:250])
-            src_text = "\n".join(src_lines)
-            prompt = "Hay viet MOT BAI BAO HOAN TOAN MOI dua tren y kien ca nhan va tu lieu tham khao duoi day.\n\nY KIEN CA NHAN (chu de):\n" + opinion[:1000] + "\n\nTU LIEU THAM KHAO:\n" + src_text + "\n\nYEU CAU:\n- Viet bai moi, KHONG sao chep nguyen van tu lieu.\n- Dung y kien ca nhan lam chu dao.\n- Long ghep dan chung tu bao mot cach tu nhien. Ghi ro nguon khi trich dan.\n- Giong van bao chi, mach lac.\n- 4-5 doan, moi doan 2-3 cau.\n- Ket thuc bang ket luan ca nhan."
-            ai_text = await ai_ext.qwen_generate(prompt, max_tokens=1500)
-            if ai_text and len(ai_text) > 150:
-                ai_success = True
-    except:
-        pass
-    if not ai_success:
-        ai_text = _build_fallback_article(opinion, source_details, title)
-    slides = _split_into_slides(ai_text, source_images)
-    return JSONResponse({"preview": {"title": title, "text": ai_text, "opinion": opinion, "images": source_images, "sources": source_details[:5], "slides": slides}})
-@app.post("/api/personal_post")
-async def api_personal_post(request: Request):
-    """Publish: AI writes new article, saves to wall."""
-    body = await request.json()
-    opinion = _clean(body.get("opinion", ""))
-    selected_topics = body.get("selected_topics", []) or []
-    selected_sources = body.get("selected_sources", []) or []
-    custom_title = body.get("custom_title", "")
-    custom_slides = body.get("custom_slides", [])
-    if not opinion or len(opinion) < 10:
-        return JSONResponse({"error": "Quan diem ca nhan qua ngan"}, status_code=400)
-    if not selected_topics:
-        hot = _get_hot_topics()
-        selected_topics = [t.get("topic", "") for t in hot[:3] if t.get("topic")]
-    all_sources = []
-    seen_urls = set()
-    for topic in selected_topics[:3]:
-        for s in _search_all(topic, limit=5):
-            if s.get("url") and s["url"] not in seen_urls:
-                seen_urls.add(s["url"])
-                all_sources.append(s)
-                if len(all_sources) >= 5:
-                    break
-        if len(all_sources) >= 5:
-            break
-    for src in selected_sources:
-        if src.get("url") and src["url"] not in seen_urls:
-            all_sources.insert(0, src)
-    source_details = []
-    source_images = []
-    for src in all_sources[:5]:
-        sd = await _scrape_source_for_opinion(src.get("url", ""))
-        if sd:
-            sd["via"] = src.get("via", "")
-            source_details.append(sd)
-            for img in sd.get("images", []):
-                if img not in source_images:
-                    source_images.append(img)
-    if custom_title:
-        title = custom_title[:80]
-    else:
-        words = re.findall(r"[A-Za-z0-9\u00c0-\u1ef9]+", opinion)
-        tw = words[:6] if len(words) >= 6 else words[:3]
-        title = " ".join(w[0].upper() + w[1:] for w in tw) if tw else "Quan diem ca nhan"
-        title = title[:80]
-    ai_success = False
-    ai_text = None
-    try:
-        import ai_ext
-        if hasattr(ai_ext, 'qwen_generate'):
-            src_lines = []
-            for i, sd in enumerate(source_details[:5]):
-                src_lines.append("[" + sd.get("via","Nguon") + "] " + sd.get("title",""))
-                for p in sd.get("paragraphs", [])[:3]:
-                    src_lines.append("  -> " + p[:250])
-            src_text = "\n".join(src_lines)
-            prompt = "Hay viet MOT BAI BAO HOAN TOAN MOI dua tren:\n\nY KIEN CA NHAN:\n" + opinion[:1000] + "\n\nTU LIEU THAM KHAO:\n" + src_text + "\n\nYEU CAU: Bai moi, lay y kien ca nhan lam chu dao, long dan chung tu bao. Ghi nguon. 4-5 doan. Ket luan ca nhan."
-            ai_text = await ai_ext.qwen_generate(prompt, max_tokens=1500)
-            if ai_text and len(ai_text) > 150:
-                ai_success = True
-    except:
-        pass
-    if not ai_success:
-        ai_text = _build_fallback_article(opinion, source_details, title)
-    if custom_slides and len(custom_slides) > 0:
-        slides = [{"text": s.get("text",""), "image": s.get("image",""), "index": i+1} for i, s in enumerate(custom_slides)]
-    else:
-        slides = _split_into_slides(ai_text, source_images)
-    lang, emotion = detect_language_and_emotion(title, ai_text)
-    voice = get_voice_for_content(title, ai_text)
-    post = {
-        "id": str(int(time.time() * 1000)) + str(_random2.randint(100, 999)),
-        "title": title, "text": ai_text,
-        "img": source_images[0] if source_images else "",
-        "url": "", "kind": "personal_opinion",
-        "slides": slides, "images": source_images[:10],
-        "video": "", "voice": voice, "emotion": emotion,
-        "language": lang, "ts": int(time.time()),
-        "sources": source_details[:5]
-    }
-    posts = _load_wall_posts()
-    posts.insert(0, post)
-    _save_wall_posts(posts)
-    return JSONResponse({"post": post, "slides": slides})
-# ===== END PERSONAL OPINION POST v3 =====
-def _bg():
-    time.sleep(15)
-    while True:
-        try:get_wc2026_all()
-        except:pass
-        time.sleep(90)
-threading.Thread(target=_bg,daemon=True).start()
-# ===== AUTO SCHEDULER: rewrite AI + short at 7/13/19 VN time =====
-_AUTO_SCHEDULE_TIMES = [(7, '07:00'), (13, '13:00'), (19, '19:00')]
-_AUTO_LOG = os.path.join(DATA_DIR, 'auto_rewrite_log.json')
-def _load_auto_log():
-    try:
-        if os.path.exists(_AUTO_LOG):
-            with open(_AUTO_LOG, 'r') as f:
-                return json.load(f)
-    except: pass
-    return {}
-def _save_auto_log(log):
-    try:
-        tmp = _AUTO_LOG + '.tmp'
-        with open(tmp, 'w') as f:
-            json.dump(log, f)
-        os.replace(tmp, _AUTO_LOG)
-    except: pass
-async def _auto_fetch_short(post_id):
-    """Try to auto-generate a short for a post."""
-    try:
-        import httpx
-        async with httpx.AsyncClient(timeout=180) as cl:
-            r = await cl.post(
-                f"http://localhost:7860/api/ai/short/{post_id}",
-                json={"voice":"vi-VN-HoaiMyNeural","emotion":"neutral","speed":1.2},
-                headers={"Content-Type":"application/json"}
-            )
-            if r.status_code < 300:
-                sj = r.json()
-                if sj.get('video'):
-                    posts = _load_wall_posts()
-                    for p in posts:
-                        if p.get('id') == post_id:
-                            p['video'] = sj['video']
-                            break
-                    _save_wall_posts(posts)
-                    return True
-    except: pass
-    return False
-async def _auto_rewrite_one(topic, slot_label, used_urls=None, post_index=0):
-    """Rewrite one topic: find articles, summarize, post to wall, trigger short.
-    used_urls: shared set to avoid duplicate articles across topics.
-    post_index: 0-based index to create multiple posts per topic (0,1,2 = up to 3 posts)."""
-    from urllib.parse import quote as _q
-    # Get MORE items to support 1-3 posts per topic
-    items = _search_all(topic, limit=12)
-    # Skip URLs already used by another topic
-    if used_urls is not None:
-        filtered = [it for it in items if it.get('url') not in used_urls]
-        if filtered:
-            items = filtered
-    if not items or post_index >= len(items):
-        return False
-    # Get article at post_index (0,1,2 for multiple posts)
-    item = items[post_index]  # post_index allows multiple articles per topic
-    url = item.get('url', '')
-    title = item.get('title', topic)
-    if url and used_urls is not None:
-        used_urls.add(url)
-    if not url.startswith('http'):
-        return False
-    data = _scrape_article_for_rewrite(url)
-    if not data or not data.get('paragraphs'):
-        return False
-    raw_text = '\n'.join(data['paragraphs'])
-    ai_text = None
-    # Try AI generation
-    try:
-        import ai_ext
-        prompt = f"Tóm tắt tin tức (tự động {slot_label}):\nTiêu đề: {data['title']}\n{raw_text[:10000]}\n\n4-6 ý chính dạng bullet. Cuối ghi nguồn."
-        ai_text = await ai_ext.qwen_generate(prompt, max_tokens=1000)
-    except: pass
-    if not ai_text or len(ai_text) < 80:
-        pts = data['paragraphs'][:6]
-        ai_text = '\n\n'.join([f"• {p[:300]}" for p in pts])
-        via = item.get('via', '') or urlparse(url).netloc.replace('www.', '')
-        ai_text += f"\n\nNguồn tham khảo: {via}"
-    # Build slides
-    images = data.get('images', [])
-    pts = data['paragraphs'][:10]
-    slides = []
-    for i, p in enumerate(pts[:8]):
-        img = images[i] if i < len(images) else (images[-1] if images else data.get('og_img', ''))
-        slides.append({'text': p[:300], 'image': img, 'index': i + 1})
-    post_id = str(int(time.time() * 1000)) + str(_random2.randint(100, 999))
-    post = {
-        "id": post_id, "title": data.get('title', title)[:200],
-        "text": ai_text, "img": images[0] if images else data.get('og_img', ''),
-        "url": url, "kind": "auto_rewrite", "slides": slides,
-        "images": images[:10], "video": "",
-        "voice": "vi-VN-HoaiMyNeural", "emotion": "neutral",
-        "language": "vietnamese", "ts": int(time.time()),
-        "auto_scheduled": True, "slot": slot_label,
-    }
-    posts = _load_wall_posts()
-    posts.insert(0, post)
-    _save_wall_posts(posts)
-    # Trigger short generation async
-    threading.Thread(target=lambda: asyncio.run(_auto_fetch_short(post_id)), daemon=True).start()
-    return True
-async def _do_scheduled_run(slot_label):
-    """Main scheduled run: 1-3 posts from 3 different HOT topics (3-9 total), no duplicates."""
-    print(f"[auto] Starting scheduled rewrite for {slot_label}")
-    # Get top hot topics, skip duplicates
-    all_topics = _get_hot_topics()
-    seen_topics = set()
-    unique_topics = []
-    for t in all_topics:
-        kw = t.get('topic', '').lower().strip()
-        if kw and len(kw) > 5 and kw not in seen_topics:
-            is_dup = False
-            for s in seen_topics:
-                # Check if one topic is substring of another
-                if kw in s or s in kw:
-                    is_dup = True
-                    break
-            if not is_dup:
-                seen_topics.add(kw)
-                unique_topics.append(t)
-        if len(unique_topics) >= 3:
-            break
-    job_topics = [t['topic'] for t in unique_topics[:3] if t.get('topic')]
-    if not job_topics:
-        print(f"[auto] No hot topics found, skipping")
-        return
-    print(f"[auto] Running 3 topics: {job_topics}")
-    # Track used URLs to avoid cross-topic duplicates
-    _used_urls = set()
-    results = []
-    # Process each topic, create 1-3 posts per topic
-    for jt in job_topics:
-        for post_idx in range(3):  # Try up to 3 posts per topic
-            try:
-                ok = await asyncio.wait_for(_auto_rewrite_one(jt, slot_label, _used_urls, post_idx), timeout=120)
-                if ok:
-                    results.append((jt, post_idx, True))
-                    print(f"[auto] Created post {post_idx+1} for '{jt}'")
-                else:
-                    # No more articles for this topic
-                    break
-            except Exception as e:
-                print(f"[auto] Error on '{jt}' post {post_idx}: {e}")
-                results.append((jt, post_idx, False))
-            await asyncio.sleep(1)  # Small delay between posts
-    # Ensure at least 3 posts total (fallback if needed)
-    successful_posts = sum(1 for _, _, ok in results if ok)
-    print(f"[auto] Done {slot_label}: {successful_posts} posts created")
-    # Log
-    from datetime import datetime, timezone, timedelta
-    VN_TZ_SCHED = timezone(timedelta(hours=7))
-    today_str = datetime.now(VN_TZ_SCHED).strftime('%Y-%m-%d')
-    log = _load_auto_log()
-    if today_str not in log: log[today_str] = {}
-    log[today_str][slot_label] = {
-        'time': datetime.now(VN_TZ_SCHED).strftime('%H:%M:%S'),
-        'count': successful_posts,
-        'total': len(job_topics),
-    }
-    _save_auto_log(log)
-def _scheduler_loop():
-    """Check every 60s; trigger at 7:00, 13:00, 19:00 VN time.
-    On startup, check for any missed slots today and run them immediately."""
-    time.sleep(35)
-    from datetime import datetime, timezone, timedelta
-    VN_TZ_SCHED = timezone(timedelta(hours=7))
-    _last_run_date = ""
-    _last_run_slots = set()
-    # On startup: check log for missed slots today
-    try:
-        start_now = datetime.now(VN_TZ_SCHED)
-        today_str = start_now.strftime('%Y-%m-%d')
-        current_hour = start_now.hour
-        current_minute = start_now.minute
-        log = _load_auto_log()
-        today_log = log.get(today_str, {})
-        for h, label in _AUTO_SCHEDULE_TIMES:
-            # Run if slot is past (either strictly earlier hour, or same hour but window has passed)
-            should_run = False
-            if h < current_hour:
-                should_run = True
-            elif h == current_hour and current_minute > 10:
-                should_run = True
-            if should_run and label not in today_log:
-                print(f"[auto] Detected missed slot {label} (h={h} < now={current_hour}:{current_minute}), running catch-up now")
-                _run_scheduled_sync(label)
-                _last_run_slots.add(label)
-    except Exception as e:
-        print(f"[auto] Catch-up check error: {e}")
-    while True:
-        try:
-            now = datetime.now(VN_TZ_SCHED)
-            today = now.strftime('%Y-%m-%d')
-            hour = now.hour
-            minute = now.minute
-            if today != _last_run_date:
-                _last_run_date = today
-                _last_run_slots = set()
-            slot = None
-            for h, label in _AUTO_SCHEDULE_TIMES:
-                if hour == h and 0 <= minute < 5:
-                    slot = label
-                    break
-            if slot and slot not in _last_run_slots:
-                _last_run_slots.add(slot)
-                _run_scheduled_sync(slot)
-        except Exception as e:
-            print(f"[auto] Loop error: {e}")
-        time.sleep(60)
-threading.Thread(target=_scheduler_loop, daemon=True, name='auto-rewrite-scheduler').start()
-@app.get('/api/debug/auto_schedule')
-async def debug_auto_schedule(slot: str = '07:00'):
-    """Manually trigger auto scheduler for debugging."""
-    try:
-        # Check if we can access the data directory
-        log = _load_auto_log()
-        topics = _get_hot_topics()[:3]
-        job_topics = [t['topic'] for t in topics if t.get('topic')]
-        return JSONResponse({
-            "slot": slot,
-            "log": log,
-            "hot_topics": job_topics,
-            "wall_posts_count": len(_load_wall_posts()),
-            "data_dir_writable": os.access(DATA_DIR, os.W_OK) if os.path.isdir(DATA_DIR) else False,
-            "data_dir_exists": os.path.isdir(DATA_DIR),
-        })
-    except Exception as e:
-        return JSONResponse({"error": str(e)}, status_code=500)
-def _run_scheduled_sync(slot):
-    """Run _do_scheduled_run in a separate event loop (for background thread)."""
-    loop = asyncio.new_event_loop()
-    asyncio.set_event_loop(loop)
-    try:
-        loop.run_until_complete(_do_scheduled_run(slot))
-    except Exception as e:
-        print(f"[auto] Background run error: {e}")
-    finally:
-        loop.close()
-@app.get('/api/debug/trigger_auto')
-async def debug_trigger_auto(slot: str = '19:00'):
-    """Trigger _do_scheduled_run in background thread (non-blocking)."""
-    threading.Thread(target=_run_scheduled_sync, args=(slot,), daemon=True).start()
-    return JSONResponse({"status": "started", "slot": slot})
-# ===== SHORTS RSS PROXY ENDPOINT =====
-@app.get("/api/shorts/rss")
-def shorts_rss():
-    """Get shorts from YouTube RSS feeds server-side"""
-    import xml.etree.ElementTree as ET
-    import html as html_lib2
-    import re as re2
-    YOUTUBE_CHANNELS = {
-        "baodantri7941": "UC_x5TKhOgd6GhYvv5z4I3jg",
-        "baosuckhoedoisongboyte": "UCBsY5fXTQLkF_JnH9kLkL4g",
-    }
-    shorts = []
-    seen = set()
-    for handle, channel_id in YOUTUBE_CHANNELS.items():
-        try:
-            rss_url = f"https://www.youtube.com/feeds/videos.xml?channel_id={channel_id}"
-            r = req.get(rss_url, headers=HEADERS, timeout=15)
-            if r.status_code != 200:
-                continue
-            root = ET.fromstring(r.text)
-            ns = {
-                'atom': 'http://www.w3.org/2005/Atom',
-                'yt': 'http://www.youtube.com/xml/schemas/2015',
-                'media': 'http://search.yahoo.com/mrss/'
-            }
-            for entry in root.findall('atom:entry', ns)[:30]:
-                title_el = entry.find('atom:title', ns)
-                title = html_lib2.unescape(title_el.text) if title_el is not None and title_el.text else ''
-                link_el = entry.find('atom:link', ns)
-                link = link_el.get('href', '') if link_el is not None else ''
-                vid_el = entry.find('yt:videoId', ns)
-                vid = vid_el.text if vid_el is not None else ''
-                if not vid or vid in seen:
-                    continue
-                # Check if it's a short
-                is_short = '#shorts' in title.lower() or '#short' in title.lower() or '/shorts/' in link
-                if not is_short:
-                    desc_el = entry.find('media:description', ns)
-                    if desc_el is not None and desc_el.text:
-                        if '#shorts' in desc_el.text.lower():
-                            is_short = True
-                if not is_short:
-                    continue
-                seen.add(vid)
-                # Get thumbnail
-                thumb = f"https://i.ytimg.com/vi/{vid}/hqdefault.jpg"
-                media_group = entry.find('media:group', ns)
-                if media_group is not None:
-                    thumb_el = media_group.find('media:thumbnail', ns)
-                    if thumb_el is not None:
-                        thumb = thumb_el.get('url', thumb)
-                shorts.append({
-                    'id': vid,
-                    'title': title.replace('#shorts', '').replace('#short', '').strip()[:120],
-                    'img': thumb,
-                    'link': f'https://www.youtube.com/shorts/{vid}',
-                    'channel': handle,
-                    'source': 'yt'
-                })
-                if len(shorts) >= 40:
-                    break
-        except Exception as e:
-            print(f"RSS error for {handle}: {e}")
-            continue
-    return {"shorts": shorts, "count": len(shorts)}
-app.mount('/static',StaticFiles(directory=STATIC_DIR),name='vnews_static')

+"""VNEWS v2 Entry Point - with fast bongda proxy + rewrite endpoints + multilingual TTS + opinion v3"""
 import sys, os
 from main import app, HEADERS, BONGDA_HEADERS, fetch_bongda_api, HL_LEAGUES
 except Exception as e:
     print(f"[WARN] ai_patch import failed: {e}")
+# PERSONAL OPINION POST v3 - AI synthesis from user opinion + hot news sources
+try:
+    import opinion_v3_patch
+    print("[app_v2_entry] opinion_v3_patch loaded: POST /api/opinion/post + GET /api/opinion/hot_context")
+except Exception as e:
+    print(f"[WARN] opinion_v3_patch import failed: {e}")
 from fastapi.responses import HTMLResponse, JSONResponse, FileResponse, Response
 from fastapi.staticfiles import StaticFiles
 from starlette.routing import Mount
 HL_LEAGUES['friendly'] = {"path": "giai-khac/friendly", "name": "Giao hữu", "emoji": "🤝"}
 STATIC_DIR = os.path.join(os.path.dirname(os.path.abspath(__file__)), 'static')
+SPACE = "https://bep40-vnews.hf.space"
 app.router.routes=[r for r in app.router.routes if not(getattr(r,'path',None)=='/' and hasattr(r,'methods') and 'GET' in getattr(r,'methods',set()))]
 app.routes[:]=[r for r in app.routes if not isinstance(r, Mount)]
 app.router.routes=[r for r in app.router.routes if not isinstance(r, Mount)]
 def _clean(s): return re.sub(r"\s+"," ",html_lib.unescape(str(s or""))).strip()
 _match_cache = {}
 def _get_match_detail(event_id, slug=None):
     headers = {"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36", "Accept": "text/html", "Referer": "https://bongda.com.vn/"}
     if slug:
 # ===== SHARE HELPERS: render content pages for shared links =====
 def _render_slides_page(post, safe_title, safe_img, safe_url):
     slides = post.get('slides', [])
     if not safe_img and slides and slides[0].get('image'):
         safe_img = slides[0].get('image', '')
     description = _clean((post.get('text') or '')[:200]) or "Tin tức tóm tắt, AI rewrite, World Cup 2026"
     if safe_url and safe_url != '/':
         canonical_url = f"{SPACE}/s?url={quote(safe_url)}&title={quote(safe_title[:100])}"
     else:
         canonical_url = f"{SPACE}/s?post_id={post.get('id') or ''}"
     h = f'''<!DOCTYPE html>
 <html lang="vi">
 <head>
 def _render_video_page(post, safe_title, safe_img, safe_url):
     video_url = post.get('video', '')
     description = _clean((post.get('text') or '')[:200]) or "Tin tức tóm tắt, AI rewrite, World Cup 2026"
     if safe_url and safe_url != '/':
         canonical_url = f"{SPACE}/s?url={quote(safe_url)}&title={quote(safe_title[:100])}"
     else:
         canonical_url = f"{SPACE}/s?post_id={post.get('id') or ''}"
     h = f'''<!DOCTYPE html>
 <html lang="vi">
 <head>
 @app.get('/s/{slug}')
 async def _sh_slug(slug: str, request: Request, url: str = '', title: str = '', img: str = ''):
     safe_title = _clean(title) if title else 'VNEWS - Tin tức'
     safe_img = _clean(img) if img else ''
     safe_url = _clean(url) if url else '/'
     post = None
     try:
+        if slug and len(slug) > 5:
             posts = _load_wall_posts()
             for p in posts:
                 if p.get('id') == slug:
                     break
     except:
         pass
     if post and post.get('slides'):
         return _render_slides_page(post, safe_title, safe_img, safe_url)
     if post and post.get('video'):
         return _render_video_page(post, safe_title, safe_img, safe_url)
     return HTMLResponse(f'''<!DOCTYPE html>
 <html lang="vi">
 <head>
     safe_title = _clean(title) if title else 'VNEWS - Tin tức'
     safe_img = _clean(img) if img else ''
     safe_url = _clean(url) if url else '/'
     post = None
     try:
         posts = _load_wall_posts()
                     safe_url = p.get('url', safe_url) or safe_url
                     break
         elif url:
             for p in posts:
                 if p.get('url') == url and p.get('slides'):
                     post = p
                     safe_url = p.get('url', safe_url) or safe_url
                     break
             if not post:
                 for p in posts:
                     if p.get('url') == url:
                         post = p
                         break
     except:
         pass
     if post and post.get('slides'):
         return _render_slides_page(post, safe_title, safe_img, safe_url)
     if post and post.get('video'):
         return _render_video_page(post, safe_title, safe_img, safe_url)
     return HTMLResponse(f'''<!DOCTYPE html>
 <html lang="vi">
 <head>
 _UA_RW = {'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36', 'Accept-Language': 'vi-VN,vi;q=0.9'}
 _UNIQUE_CHARS = {
     'vietnamese': set('đăâêôơưàảãạáằẳẵặắầẩẫậấèẻẽẹéềễểệếìỉĩịíòỏõọóồổỗộốờởỡợớùủũụúừửữựứỳỷỹỵý'),
     'spanish': set('ñáéíóúü¿¡'),
 }
 def detect_language(text):
     if not text:
         return 'vietnamese'
     text_lower = text.lower()
     text_chars = set(text_lower)
     vn_chars = len(text_chars & _UNIQUE_CHARS['vietnamese'])
     if vn_chars >= 2:
         return 'vietnamese'
     es_chars = len(text_chars & _UNIQUE_CHARS['spanish'])
     pt_chars = len(text_chars & _UNIQUE_CHARS['portuguese'])
     words = set(re.findall(r'\b\w+\b', text_lower))
     scores = {}
     for lang, stops in _STOPWORDS.items():
         scores[lang] = len(words & stops) / max(len(stops), 1)
     pt_markers = {'não', 'pelo', 'pela', 'isso', 'há', 'estão', 'num', 'numa', 'tenho', 'posso', 'você', 'nós', 'eles', 'elas', 'também', 'muito', 'já', 'só', 'até', 'entre', 'depois', 'sem', 'mesmo', 'aos', 'serão'}
     es_markers = {'pero', 'está', 'están', 'porque', 'también', 'hasta', 'donde', 'quien', 'fue', 'son', 'fueron', 'ese', 'eso', 'ante', 'ellos', 'ella', 'nosotros', 'usted', 'ustedes', 'tú', 'él', 'desde', 'todo', 'durante', 'todos', 'uno', 'les', 'ni', 'contra', 'otros', 'fueron'}
     pt_overlap = len(words & pt_markers)
     es_overlap = len(words & es_markers)
     if scores.get('portuguese', 0) > 0 and pt_overlap > es_overlap:
         return 'portuguese'
     if scores.get('spanish', 0) > 0 and es_overlap > pt_overlap:
         return 'spanish'
     if scores.get('english', 0) > 0.15:
         return 'english'
     best = max(scores, key=scores.get)
     return best if scores[best] > 0.05 else 'vietnamese'
+# ===== EXPORT FUNCTIONS NEEDED BY opinion_v3_patch =====
+# These are imported by opinion_v3_patch to avoid circular deps