rewrite_slide.py

"""Fast rewrite as slides - no AI needed, extracts key points + images from article."""
from main import app
from fastapi import Request
from fastapi.responses import JSONResponse
import requests, re, time, random, json, os
from bs4 import BeautifulSoup
from urllib.parse import quote

UA = {'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36', 'Accept-Language': 'vi-VN,vi;q=0.9'}

try:
    from main import _load_wall, _save_wall
except:
    _data_dir = "/data" if os.path.isdir("/data") else "/app/data"
    _wall_file = os.path.join(_data_dir, "wall_posts.json")
    def _load_wall():
        try:
            if os.path.exists(_wall_file):
                with open(_wall_file, 'r', encoding='utf-8') as f: return json.load(f)
        except: pass
        return []
    def _save_wall(posts):
        try:
            os.makedirs(os.path.dirname(_wall_file), exist_ok=True)
            with open(_wall_file+'.tmp', 'w', encoding='utf-8') as f: json.dump(posts[:100], f, ensure_ascii=False)
            os.replace(_wall_file+'.tmp', _wall_file)
        except: pass


def _clean(s): return re.sub(r'\s+', ' ', str(s or '')).strip()


def _scrape_article_full(url):
    """Scrape article: extract paragraphs + ALL images."""
    try:
        r = requests.get(url, headers=UA, 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()
        
        # Title
        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 '')
        
        # OG image
        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
        
        # Find content block
        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
        
        # Extract paragraphs and images IN ORDER
        paragraphs = []
        images = []
        seen_imgs = set()
        
        if og_img and og_img not in seen_imgs:
            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:
                            images.append(src)
                            seen_imgs.add(src)
        
        return {'title': _clean(title), 'paragraphs': paragraphs, 'images': images, 'og_img': og_img}
    except Exception as e:
        return None


def _extract_key_points(paragraphs, max_points=5):
    """Extract key points: take first sentence of each significant paragraph."""
    points = []
    for p in paragraphs:
        if len(points) >= max_points: break
        # Take first complete sentence (ends with . ! ?)
        m = re.match(r'^(.+?[.!?])\s', p)
        if m:
            sentence = m.group(1)
        else:
            sentence = p[:150] + ('.' if not p.endswith('.') else '')
        
        # Skip if too short or duplicate
        if len(sentence) < 30: continue
        if any(sentence[:50] in existing for existing in points): continue
        
        points.append(sentence)
    
    return points


@app.post("/api/rewrite_slide")
async def api_rewrite_slide(request: Request):
    """
    Fast rewrite as SLIDES:
    - Extract key points from article (1 sentence each, full and complete)
    - Pair each point with an image from the article
    - Return as slides array for frontend to display
    - Save to Tường AI
    NO AI NEEDED - instant response.
    """
    body = await request.json()
    url = _clean(body.get("url", ""))
    context = body.get("context", "")
    
    if not url and not context:
        return JSONResponse({"error": "Cần URL hoặc nội dung"}, status_code=400)
    
    # Scrape article
    data = None
    if url and url.startswith("http"):
        data = _scrape_article_full(url)
    
    if not data and context:
        # Use context passed from frontend
        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)
    
    # Extract key points
    points = _extract_key_points(data['paragraphs'], max_points=6)
    if not points:
        return JSONResponse({"error": "Không tìm được ý chính"}, status_code=422)
    
    # Build slides: pair each point with an image
    images = data.get('images', [])
    slides = []
    for i, point in enumerate(points):
        img = images[i] if i < len(images) else (images[-1] if images else '')
        # Proxy dantri images
        if img and 'cdnphoto.dantri' in img:
            img = '/api/proxy/img?url=' + quote(img, safe='')
        slides.append({
            'text': point,
            'image': img,
            'index': i + 1
        })
    
    # Create post for Tường AI
    summary_text = '\n\n'.join([f"• {s['text']}" for s in slides])
    # Auto voice + emotion based on topic (reuse ai_ext detector if available)
    try:
        from ai_ext import _detect_voice_emotion
        _voice, _emotion = _detect_voice_emotion(data['title'], summary_text)
    except Exception:
        _voice, _emotion = "hoaimy", "trung_tinh"
    post = {
        "id": str(int(time.time() * 1000)) + str(random.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,
        "ts": int(time.time())
    }
    
    # Save to wall
    posts = _load_wall()
    posts.insert(0, post)
    _save_wall(posts)
    
    return JSONResponse({"post": post, "slides": slides})