ai_runtime.py

import os, re, subprocess, json, time, hashlib
import ai_patch as old
from ai_patch import app
import ai_ext as base
from fastapi import Request
from fastapi.responses import JSONResponse, HTMLResponse, FileResponse
try:
    from PIL import Image, ImageDraw, ImageFont
except Exception:
    Image = ImageDraw = ImageFont = None


def clean(s):
    import html as html_lib
    return re.sub(r"\s+", " ", html_lib.unescape(s or "")).strip()


def _domain(url):
    try:
        from urllib.parse import urlparse
        return urlparse(url or '').netloc.replace('www.','')
    except Exception:
        return ''


def _strip_bullet_prefix(s):
    # remove bullets, numbered prefixes, leading dots commonly produced by AI summaries
    return clean(re.sub(r'^[\s•\-\*·▪▫●○\d\.\)\(]+', '', s or ''))


def source_line(sources):
    names=[]
    for s in (sources or [])[:5]:
        via=s.get('via') or _domain(s.get('url','')) or s.get('title','')
        if via and via not in names:names.append(via)
    return 'Nguồn tham khảo: '+', '.join(names[:5]) if names else 'Nguồn tham khảo: tổng hợp internet'


def _source_badge(post):
    sources=post.get('sources') or []
    for s in sources:
        via=s.get('via') or _domain(s.get('url',''))
        if via:return via
    return _domain(post.get('url','')) or post.get('source') or 'VNEWS'


def _collect_all_images(data):
    imgs=[]
    def add(u):
        u=(u or '').strip()
        if not u or u.startswith('data:') or 'base64' in u:return
        if u.startswith('//'):u='https:'+u
        if u not in imgs:imgs.append(u)
    add(data.get('image') or data.get('og_image') or data.get('img'))
    for u in data.get('images') or []:add(u)
    for b in data.get('body') or []:
        if isinstance(b,dict) and b.get('type')=='img':add(b.get('src'))
    return imgs[:20]


def _scrape_url_with_images(url):
    data=base.scrape_any_url(url)
    # extra pass: collect every useful image from original HTML, because some readers only return one image
    try:
        import requests
        from bs4 import BeautifulSoup
        r=requests.get(url,headers=base.HEADERS,timeout=18);r.encoding='utf-8'
        soup=BeautifulSoup(r.text,'lxml')
        extra=[]
        for im in soup.find_all('img'):
            src=im.get('data-src') or im.get('data-original') or im.get('data-lazy-src') or im.get('src') or ''
            if src.startswith('//'):src='https:'+src
            if src and 'base64' not in src and src not in extra:
                # skip tiny icons/logos as much as possible
                low=src.lower()
                if any(x in low for x in ['logo','icon','avatar','sprite']):
                    continue
                extra.append(src)
            if len(extra)>=20:break
        data['images']=_collect_all_images(data)+[u for u in extra if u not in _collect_all_images(data)]
    except Exception:
        data['images']=_collect_all_images(data)
    data['images']=_collect_all_images(data)
    if data['images'] and not data.get('image'):
        data['image']=data['images'][0]
    return data


def rich_context(topic, limit=5):
    try: ctx,sources=base.web_context(topic, limit=limit)
    except Exception: ctx,sources='',[]
    rich=[];rs=[];seen=set()
    for s in (sources or [])[:limit*2]:
        url=s.get('url') or ''
        if not url.startswith('http') or url in seen:continue
        seen.add(url)
        try:
            data=base.scrape_any_url(url)
            raw=(data.get('summary','')+'\n'+data.get('text','')).strip()
            if len(raw)<180:continue
            title=data.get('title') or s.get('title') or url
            via=data.get('via') or s.get('via') or _domain(url)
            rich.append(f"### {title} ({via})\n{raw[:2600]}")
            rs.append({'title':title,'url':url,'excerpt':raw[:700],'via':via})
            if len(rich)>=limit:break
        except Exception:continue
    if rich:return '\n\n'.join(rich),rs
    return ctx or f'Chủ đề: {topic}', sources or []


def postprocess(text):
    if hasattr(old,'_postprocess_ai_text'):
        out=old._postprocess_ai_text(text, max_units=7)
    else:
        out=clean(text)
    # keep wall text readable, but ensure short generation later won't show bullets
    return out


# Remove old routes we must override.
_PATCH={('/api/topic_post','POST'),('/api/url_wall','POST'),('/api/rewrite_share','POST'),('/api/ai/url','POST'),('/api/ai/short/{post_id}','POST'),('/api/ai/short-file/{file_id}','GET'),('/','GET')}
app.router.routes=[r for r in app.router.routes if not any(getattr(r,'path',None)==p and m in getattr(r,'methods',set()) for p,m in _PATCH)]


@app.post('/api/url_wall')
async def url_wall_only(request:Request):
    body=await request.json();url=base._clean_text(body.get('url',''))
    if not url.startswith('http'):return JSONResponse({'error':'missing url'},status_code=400)
    try:data=_scrape_url_with_images(url)
    except Exception as e:return JSONResponse({'error':'Không scrape được URL: '+str(e)[:180]},status_code=422)
    raw=(data.get('summary','')+'\n'+data.get('text','')).strip()
    if len(raw)<120:return JSONResponse({'error':'URL không có đủ nội dung để tóm tắt'},status_code=422)
    prompt=f"""Tóm tắt bài viết nguồn dưới đây để đăng lên Tường AI VNEWS.

Yêu cầu bắt buộc:
- Chỉ tóm tắt nội dung chính, không viết lại toàn bộ bài.
- Ngắn gọn, cụ thể, dễ hiểu.
- Không lặp lại ý và không thêm chi tiết ngoài nguồn.
- Tối đa 5 ý chính hoặc 2 đoạn ngắn.
- Tránh dùng dấu đầu dòng nếu không thật cần thiết.

Tiêu đề gốc: {data.get('title','')}
Nguồn: {data.get('via','') or _domain(url)}
Nội dung gốc:
{raw[:16000]}"""
    text=await base.qwen_generate(prompt,image_url=(data.get('image') or None),max_tokens=900)
    if not text:text=old._fallback_summary_from_prompt(prompt,max_units=5) if hasattr(old,'_fallback_summary_from_prompt') else raw[:900]
    text=postprocess(text)
    src=[{'title':data.get('title'), 'url':url, 'excerpt':raw[:500], 'via':data.get('via') or _domain(url)}]
    if 'Nguồn tham khảo:' not in text:text+='\n\n'+source_line(src)
    images=_collect_all_images(data)
    post=base.make_post(data.get('title') or 'Bài viết',text,images[0] if images else (data.get('image') or ''),url,'url',sources=src)
    post['images']=images
    posts=base._load_ai_wall();posts.insert(0,post);base._save_ai_wall(posts)
    return JSONResponse({'post':post})


@app.post('/api/rewrite_share')
async def rewrite_share_url_only(request:Request):
    return await url_wall_only(request)


@app.post('/api/ai/url')
async def ai_url_compat(request:Request):
    return await url_wall_only(request)


@app.post('/api/topic_post')
async def topic_disabled(request:Request):
    return JSONResponse({'error':'Đã tắt tạo bài theo chủ đề. Vui lòng dán URL bài viết để AI tóm tắt.'},status_code=410)


def split_segments(post,max_segments=8):
    text=clean(post.get('text') or post.get('title') or '')
    text=re.sub(r'Nguồn tham khảo:.*$','',text,flags=re.I|re.S).strip()
    lines=[]
    for ln in text.splitlines():
        ln=_strip_bullet_prefix(ln)
        if len(ln)>=18:lines.append(ln)
    if len(lines)<2:
        lines=[_strip_bullet_prefix(s) for s in re.split(r'(?<=[\.\!\?])\s+',text) if len(_strip_bullet_prefix(s))>=25]
    segs=[];cur=''
    for ln in lines:
        ln=_strip_bullet_prefix(ln)
        if not ln:continue
        if len(cur)+len(ln)<180:cur=(cur+' '+ln).strip()
        else:
            if cur:segs.append(_strip_bullet_prefix(cur))
            cur=ln
    if cur:segs.append(_strip_bullet_prefix(cur))
    return segs[:max_segments] or [_strip_bullet_prefix(post.get('title','VNEWS'))]


def wrap_text(draw,text,font,maxw,max_lines):
    words=clean(text).split();lines=[];cur=''
    for w in words:
        test=(cur+' '+w).strip()
        try:width=draw.textbbox((0,0),test,font=font)[2]
        except Exception:width=len(test)*20
        if width<=maxw:cur=test
        else:
            if cur:lines.append(cur)
            cur=w
        if len(lines)>=max_lines:break
    if cur and len(lines)<max_lines:lines.append(cur)
    return lines


def _draw_center(draw, lines, font, y, fill, W, line_h):
    for ln in lines:
        try:
            box=draw.textbbox((0,0),ln,font=font);tw=box[2]-box[0]
        except Exception:
            tw=len(ln)*24
        x=max(30,(W-tw)//2)
        draw.text((x,y),ln,fill=fill,font=font)
        y+=line_h
    return y


def make_frame(post,seg,idx,total,img_path,out_path):
    if Image is None:raise RuntimeError('Pillow not ready')
    W,H=1080,1920;bg=Image.new('RGB',(W,H),(12,12,12))
    hero_h=760
    try:
        im=Image.open(img_path).convert('RGB');ratio=im.width/max(1,im.height)
        target=(W,hero_h);tr=target[0]/target[1]
        if ratio>tr:nh=target[1];nw=int(nh*ratio)
        else:nw=target[0];nh=int(nw/ratio)
        im=im.resize((nw,nh));left=(nw-target[0])//2;top=(nh-target[1])//2
        bg.paste(im.crop((left,top,left+target[0],top+target[1])),(0,0))
    except Exception:pass
    draw=ImageDraw.Draw(bg)
    try:
        fb=ImageFont.truetype('/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf',58)
        ft=ImageFont.truetype('/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf',38)
        fs=ImageFont.truetype('/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf',30)
        fsmall=ImageFont.truetype('/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf',28)
    except Exception:fb=ft=fs=fsmall=None
    # source badge on top image corner
    badge='Nguồn: '+_source_badge(post)
    try:
        b=draw.textbbox((0,0),badge,font=fsmall);bw=b[2]-b[0];bh=b[3]-b[1]
    except Exception:
        bw=len(badge)*16;bh=34
    bx=W-bw-42;by=24
    draw.rounded_rectangle((bx-16,by-8,W-24,by+bh+14),radius=18,fill=(0,0,0,170))
    draw.text((bx,by),badge,fill=(255,255,255),font=fsmall)
    # bottom text area
    draw.rectangle((0,hero_h-20,W,H),fill=(12,12,12))
    # progress bars centered
    total_w=total*38-14;start=(W-total_w)//2
    for i in range(total):
        fill=(92,184,122) if i==idx else (70,70,70)
        draw.rounded_rectangle((start+i*38,820,start+i*38+24,832),radius=6,fill=fill)
    brand='VNEWS AI SHORT'
    try:
        bb=draw.textbbox((0,0),brand,font=ft);tx=(W-(bb[2]-bb[0]))//2
    except Exception:tx=360
    draw.text((tx,870),brand,fill=(110,231,143),font=ft)
    clean_seg=_strip_bullet_prefix(seg)
    lines=wrap_text(draw,clean_seg,fb,W-120,8)
    block_h=len(lines)*74
    y=max(980, 1250-block_h//2)
    _draw_center(draw,lines,fb,y,(255,255,255),W,74)
    # small title centered near bottom
    title_lines=wrap_text(draw,_strip_bullet_prefix(post.get('title','')),fs,W-120,3)
    y2=1640
    draw.line((80,y2-26,W-80,y2-26),fill=(70,70,70),width=2)
    _draw_center(draw,title_lines,fs,y2,(220,220,220),W,42)
    bg.save(out_path,quality=92)


def make_tts(text,voice,out_path):
    v={'nam':'vi-VN-NamMinhNeural','male':'vi-VN-NamMinhNeural','nu':'vi-VN-HoaiMyNeural','female':'vi-VN-HoaiMyNeural','mien-nam':'vi-VN-HoaiMyNeural'}.get(voice,'vi-VN-HoaiMyNeural')
    text=_strip_bullet_prefix(text)
    try:subprocess.run(['python','-m','edge_tts','--voice',v,'--text',text,'--write-media',out_path],check=True,stdout=subprocess.PIPE,stderr=subprocess.PIPE,timeout=160)
    except Exception:
        tld='com.vn' if voice in ('nu','female','mien-nam') else 'com'
        try:base.gTTS(text,lang='vi',tld=tld,slow=False).save(out_path)
        except TypeError:base.gTTS(text,lang='vi',slow=False).save(out_path)


@app.post('/api/ai/short/{post_id}')
async def short_segments(post_id:str,request:Request):
    try:body=await request.json()
    except Exception:body={}
    voice=str(body.get('voice','nu')).lower().strip();emotion=str(body.get('emotion','neutral')).lower().strip();speed=max(0.85,min(1.35,float(body.get('speed',1.2) or 1.2)))
    posts=base._load_ai_wall();post=next((p for p in posts if str(p.get('id'))==str(post_id)),None)
    if not post:return JSONResponse({'error':'post not found'},status_code=404)
    segs=split_segments(post,8)
    os.makedirs(base.SHORTS_DIR,exist_ok=True);suffix=f'_{voice}_{emotion}_{str(speed).replace(".","p")}_centered_source_nobullet'
    out=os.path.join(base.SHORTS_DIR,base._safe_name(post_id+suffix)+'.mp4')
    if os.path.exists(out):post['video']='/api/ai/short-file/'+post_id+suffix;base._save_ai_wall(posts);return JSONResponse({'video':post['video'],'segments':len(segs),'subtitles':False})
    work=os.path.join(base.SHORTS_DIR,base._safe_name(post_id+suffix));os.makedirs(work,exist_ok=True)
    img=os.path.join(work,'image.jpg');base._download_image(post.get('img'),post.get('title','AI news'),img)
    clips=[]
    try:
        for i,seg in enumerate(segs):
            frame=os.path.join(work,f'f{i}.jpg');aud=os.path.join(work,f'a{i}.mp3');aud2=os.path.join(work,f'a{i}_fast.mp3');clip=os.path.join(work,f'c{i}.mp4')
            seg=_strip_bullet_prefix(seg)
            make_frame(post,seg,i,len(segs),img,frame)
            prefix={'urgent':'Tin nhanh.','warm':'Câu chuyện đáng chú ý.','serious':'Bản tin nghiêm túc.','energetic':'Cập nhật nổi bật.'}.get(emotion,'')
            spoken=(prefix+' '+seg).strip() if i==0 and prefix else seg
            make_tts(spoken,voice,aud)
            subprocess.run(['ffmpeg','-y','-i',aud,'-filter:a',f'atempo={speed}','-vn',aud2],check=True,stdout=subprocess.PIPE,stderr=subprocess.PIPE,timeout=120)
            subprocess.run(['ffmpeg','-y','-loop','1','-i',frame,'-i',aud2,'-shortest','-c:v','libx264','-tune','stillimage','-pix_fmt','yuv420p','-c:a','aac','-b:a','128k','-vf','scale=1080:1920',clip],check=True,stdout=subprocess.PIPE,stderr=subprocess.PIPE,timeout=180)
            clips.append(clip)
        lf=os.path.join(work,'list.txt')
        with open(lf,'w',encoding='utf-8') as f:
            for c in clips:f.write("file '{}".format(c.replace("'","'\\''"))+"'\n")
        subprocess.run(['ffmpeg','-y','-f','concat','-safe','0','-i',lf,'-c','copy',out],check=True,stdout=subprocess.PIPE,stderr=subprocess.PIPE,timeout=240)
        post['video']='/api/ai/short-file/'+post_id+suffix;post['short_subtitles']=False;post['short_segments']=segs;post['short_speed']=speed;base._save_ai_wall(posts)
        return JSONResponse({'video':post['video'],'segments':len(segs),'subtitles':False})
    except Exception as e:return JSONResponse({'error':'Không tạo được shorts: '+str(e)[:200]},status_code=500)


@app.get('/api/ai/short-file/{file_id}')
def short_file(file_id:str):
    path=os.path.join(base.SHORTS_DIR,base._safe_name(file_id)+'.mp4')
    if not os.path.exists(path):return JSONResponse({'error':'not found'},status_code=404)
    return FileResponse(path,media_type='video/mp4',filename=f'vnews-ai-{file_id}.mp4')


# Rebuild / with old UI injection plus final UI overrides.
app.router.routes=[r for r in app.router.routes if not (getattr(r,'path',None)=='/' and 'GET' in getattr(r,'methods',set()))]
@app.get('/')
async def index_runtime():
    with open('/app/static/index.html','r',encoding='utf-8') as f:html=f.read()
    inject=getattr(old,'PATCH_INJECT','')+r'''
<style>
/* Hide old topic UI, keep URL input only */
#ai-topic-input{display:none!important}
#ai-topic-input,*[onclick*="createTopicPost"]{display:none!important}
.ai-topic-row,.topic-row,.ai-compose-topic{display:none!important}
.ai-wall-gallery{display:grid;grid-template-columns:repeat(2,1fr);gap:6px;margin:10px 0}.ai-wall-gallery img{width:100%;aspect-ratio:16/9;object-fit:cover;border-radius:8px;background:#222}.ai-wall-gallery img:first-child{grid-column:1/-1}.ai-url-only-note{font-size:11px;color:#888;margin:5px 0 8px}
</style>
<script>
(function(){
function esc(s){return String(s||'').replace(/[&<>"']/g,m=>({'&':'&amp;','<':'&lt;','>':'&gt;','"':'&quot;',"'":'&#39;'}[m]));}
function hideTopicControls(){
  document.querySelectorAll('#ai-topic-input').forEach(e=>{let p=e.closest('.ai-compose,.ai-compose-topic,.topic-row,div'); if(p&&p.querySelector('#ai-url-input')) e.style.display='none'; else if(p) p.style.display='none';});
  document.querySelectorAll('button').forEach(b=>{let t=(b.textContent||'').toLowerCase();let oc=b.getAttribute('onclick')||'';if(oc.includes('createTopicPost')||t.includes('chủ đề'))b.style.display='none';});
  let url=document.getElementById('ai-url-input'); if(url&&!document.getElementById('ai-url-only-note')){let n=document.createElement('div');n.id='ai-url-only-note';n.className='ai-url-only-note';n.textContent='Dán URL bài viết để AI tóm tắt và lấy ảnh từ bài.';url.insertAdjacentElement('afterend',n);}
}
window.createTopicPost=function(){alert('Đã tắt ô nhập chủ đề. Vui lòng dán URL bài viết.');};
window.createUrlPost=function(){let inp=document.getElementById('ai-url-input');let url=(inp&&inp.value||'').trim();if(!url)return alert('Dán URL trước');if(!/^https?:\/\//i.test(url))return alert('URL cần bắt đầu bằng http:// hoặc https://');fetch('/api/url_wall',{method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify({url})}).then(r=>r.json().then(j=>({ok:r.ok,j}))).then(({ok,j})=>{if(ok&&j.post){if(typeof prependWallPost==='function')prependWallPost(j.post);if(window.patchedWall)window.patchedWall=[j.post].concat(window.patchedWall||[]);if(inp)inp.value='';alert('Đã tóm tắt URL, lấy ảnh trong bài và đăng lên Tường AI');location.reload();}else alert(j.error||'Lỗi URL')}).catch(e=>alert(e.message||'Lỗi URL'));};
function galleryHtml(p){let imgs=(p.images||[]).filter(Boolean);if(!imgs.length&&p.img)imgs=[p.img];if(!imgs.length)return '';return '<div class="ai-wall-gallery">'+imgs.slice(0,12).map(u=>`<img src="${esc(u)}" loading="lazy">`).join('')+'</div>';}
function patchReaders(){
  let oldRead=window.aiReadWallPatched||window.aiReadWall;
  window.aiReadWallPatched=window.aiReadWall=function(i){let arr=window.patchedWall||window.aiWall||[];let p=arr[i];if(!p&&oldRead)return oldRead(i);if(!p)return;showView('view-article');let sources='';if(p.sources&&p.sources.length){sources='<div class="article-summary"><b>Nguồn tham khảo:</b><br>'+p.sources.slice(0,5).map(s=>`• ${esc(s.title||s.url||'Nguồn')} ${s.url?`(${esc(new URL(s.url).hostname.replace('www.',''))})`:''}`).join('<br>')+'</div>'}let h=`<button class="back-btn" onclick="switchCat('home')">← Quay lại</button><div class="article-view"><span class="badge badge-ai">AI</span><h1 class="article-title">${esc(p.title)}</h1>${galleryHtml(p)}${sources}<p class="article-p" style="white-space:pre-wrap">${esc(p.text)}</p>${p.video?`<video class="article-img" src="${p.video}" controls playsinline></video>`:''}<div class="article-actions">${p.url?`<button onclick="window.open('${p.url}','_blank')">🔗 Nguồn</button>`:''}<button onclick="aiMakeShortPatched?aiMakeShortPatched(${i}):aiMakeShort(${i})">🎬 Tạo video shorts</button></div></div>`;document.getElementById('view-article').innerHTML=h;window.scrollTo(0,0);};
}
setInterval(hideTopicControls,1000);setTimeout(hideTopicControls,300);setTimeout(patchReaders,1600);
})();
</script>
'''
    return HTMLResponse(html.replace('</body>',inject+'\n</body>') if '</body>' in html else html+inject)